Glossary

2380 terms from every lesson, each in one plain sentence. Tap a lesson name to see the term in context.

Find a term

A

A/A tests: A test where both groups get exactly the same version, so every difference between them is chance.In Hypothesis tests without the fog
A/B test: An experiment that randomly splits users into groups that see different versions at the same time, so any difference in results can be credited to the version.In Experiments: A/B tests, holdouts and store pilots
ABAC: Attribute-based access control: the decision uses attributes of the user, the record and the situation, such as user.siteIds containing order.siteId.In Security essentials
abort conditions: A condition, agreed before an experiment, that stops it at once and removes the fault, such as checkout errors above 2% for 2 minutes.In Capacity, resilience and game days
AbortController: A browser and Node object with an abort() method and a signal property. You pass the signal to fetch or other APIs; calling abort() cancels them.In Async JavaScript: callbacks, promises and async/await, Building the app
absolute path: A path from the very top of the file system, such as /Users/you/work or C:\Users\you\work. It means the same place from anywhere.In The terminal without fear
acceptance checks: A short, testable statement of when a feature is done, often written as Given a situation, When the user acts, Then this result happens.In From idea to one-page spec, A spec the AI can follow, Building in small steps
Acceptance criteria: The conditions a story must meet to be accepted as done, agreed before work starts and checked at the end.In User stories and acceptance criteria, Bugs, reports and acceptance
acceptance criterion: The conditions a story must meet to be accepted, written so anyone can check them, often as Given/When/Then.In What testing is for
acceptance measure: The measurable check that shows a requirement is met, written so a tester can say pass or fail.In Requirements that work
ACCESS EXCLUSIVE: The strongest table lock in Postgres. While it is held or waited for, no other query can read or write the table.In Changing the schema without downtime, Partitioning big tables
access review: A regular check where owners confirm, account by account, that each person still needs the access they have.In Accounts, SSO and MFA
access token: A short-lived token, often a JWT valid for 5 to 15 minutes, sent with every API call.In Security essentials
Accessibility: Designing so that people with different abilities, including low vision, colour blindness and limited dexterity, can use a product.In Design for PMs
accessibility tree: A simplified copy of the page that the browser gives to assistive technology. Each node has a role (button, table, heading), a name and a state.In Accessible, fast pages, Testing React like a user
accessible name: The text assistive technology uses to identify an element, computed from its label, aria-labelledby, aria-label or its own text.In Testing React like a user, Accessibility, devices and the real world
accessible role: What an element is to assistive technology, such as button, link, checkbox or heading, plus its accessible name, usually its visible text or label.In End-to-end tests with Playwright
accountability without direct control: Being answerable for a result even though other people do the work and make most of the daily decisions that produce it.In The department head's job
accrual: A cost recorded in the period it was incurred, before the invoice arrives or is paid. It sits on the balance sheet as a liability until paid.In The three statements, Budgets, capex and opex
accrual accounting: Accounting that records revenue when it is earned and costs when they are incurred, whatever the date cash moves.In The three statements
accumulating snapshot: A fact table with one row per process instance, such as one app order, that is updated as the process reaches each milestone.In From questions to a reporting model
accuracy: The share of all predictions that are correct. Misleading when one class is rare.In Classification and churn
ACID: The four promises of a database transaction: all or nothing (atomic), rules kept (consistent), transactions do not see each other's half-done work (isolated), and committed data survives a crash (durable). Transactions
ACK: Acknowledgement: a small message that says which data arrived safely.In Networking you actually need
acknowledges: Acknowledge: the worker tells the queue the job is finished, so the queue can remove it and will not hand it to anyone else.In Queues, background jobs and pub/sub
ACME: Automatic Certificate Management Environment: a standard protocol (RFC 8555) that lets a server request, prove control of a name and renew certificates with no human steps.In Cryptography without the maths
action: A reusable piece of automation published in a repository, used in a step with uses, for example actions/checkout@v7.In GitHub Actions: continuous integration, State, effects and where data should live
action titles: A slide or section title written as a full sentence that states the point of the slide.In Executive updates and presentations, Charts that answer the question
actionability checks: Checks Playwright makes before an action: the element is visible, stable (not moving), enabled, and would receive the click (nothing covers it).In End-to-end tests with Playwright
Activation: The moment a new user first gets the product's core value, defined as a specific event that predicts whether they keep using it.In Funnels, activation and adoption
activation metric: A measurable early event, such as a second order within 14 days, that new users who go on to stay are much more likely to reach than those who leave.In Onboarding and the first-time experience
Active structure: ArchiMate elements that do things: actors, roles, application components, nodes and devices.In Frameworks and governance
active-active: A setup where two or more regions all serve live traffic at the same time, so if one fails, the others take its users.In Netflix: streaming to every screen
adaptive bitrate: Adaptive bitrate streaming: the player measures the network and its buffer, and for each segment downloads the rung that fits, switching up or down during playback.In Netflix: streaming to every screen
additive: A measure you can sum across every dimension, such as units sold or sales in cents, and still get a meaningful total.In From questions to a reporting model
ADKAR: A model of individual change from the consulting firm Prosci: Awareness, Desire, Knowledge, Ability and Reinforcement, which a person needs in that order.In Change management and adoption
ADM: Architecture Development Method: TOGAF's cycle of phases for developing and governing an architecture, from vision to change management.In Frameworks and governance
admission controller: A Kubernetes component that checks every new or changed object, such as a pod, against rules before the cluster accepts it.In Secure supply chain and delivery metrics
ADR: Architecture decision record: a one-page note that records one important technical decision, its context, the options considered and the consequences.In Running a design review, Technical vision and design docs
ADR (architecture decision record): A one-page note recording one decision, its context, options and consequences. ADRs and C4
advance shipping notice: Advance shipping notice: a message a supplier sends before a delivery listing what is on the truck, often linked to a barcode on the delivery note.In Designing for frontline work
advice process: A way of deciding where anyone may make an architectural decision, as long as they first ask for advice from everyone affected and from people with expertise, and record that advice.In What architects actually do, Decisions, diagrams and reviews, Frameworks and governance
advisory database: A public database of known vulnerabilities in packages, recording which package, which versions are affected and which version fixes it.In Supply chain and secrets
advisory lock: A lock on a number you choose, not on a row or table. pg_advisory_xact_lock waits for the lock and releases it at commit or rollback.In Offline-first apps and data sync
Affinity mapping: Grouping research notes by similarity, usually by moving them around on a wall or board until clear clusters appear.In From research to opportunities
agent: A shop or kiosk that exchanges cash for wallet balance and back, earning a commission, acting like a small bank branch.In Teardown: sending money with a mobile wallet, AI agents in operations, Designing AI features: LLMs, RAG and vector search, What makes an agent
AGENTS.md: A Markdown file in a repo that tells AI coding assistants how the project works: commands, layout and rules. Working with AI coding assistants
Aggregate: A function that turns many rows into one value, such as SUM, COUNT or AVG. Everyday SQL
Aggregation: Turning many rows into one number, such as a sum, count or average. Pivot tables
AI incident: An event where an AI system causes or nearly causes harm: wrong or unsafe outputs at scale, a data leak, an action taken without proper approval, or costs running out of control.In Governance and accountability
AI inventory: A maintained list of every AI system a company builds or uses, with its purpose, users, model, data, risk tier, owner, test status and review dates.In Governance and accountability
alert: A notice raised by a detection rule that something may be worth a human look. Most alerts turn out to be harmless.In Inside a SOC
alert fatigue: When people get so many alerts, many of them false or useless, that they start ignoring them, including the real ones.In Observability: logs, metrics and traces, Monitoring, drift and retraining, Writing detections
Alertmanager: The Prometheus component that receives firing alerts, groups and de-duplicates them, silences or inhibits some, and sends notifications to receivers such as a paging tool or a ticket system.In Alerting that does not burn people out
Aliases: A named pointer to one version of a registered model, such as champion, that you can move to another version.In Experiments, versions and the model registry
All-k pass: The share of tasks that passed on every one of k runs. It answers whether the agent can be trusted to do this task every time, and falls as k grows.In Evaluating agents
Allocated costs: A share of a shared cost assigned to a product, channel or unit by a rule, such as share of revenue, headcount or orders. The total is real; the slice usually does not disappear when the unit does.In Unit economics
allow-list: A fixed list of the only values your code accepts. Anything not on the list is refused or replaced with a default.In Injection and unsafe input
alternative hypothesis: The claim you accept if the data make the null hypothesis hard to believe: here, that the true conversion rates differ.In Hypothesis tests without the fog
always-on feedback: A standing weekly process that collects, tags and reviews feedback from everyday sources such as tickets, reviews and chat groups.In From research to opportunities
amortisation: Spreading the cost of an intangible asset, such as capitalised software, over its useful life.In Budgets, capex and opex
amortised cost: Cost with up-front commitment payments spread evenly over the months they cover, so each month shows what it really used.In Cloud and SaaS costs with FinOps
Analytical: A dashboard built for exploring data, with many filters and drill paths, used by analysts to find causes.In Designing a dashboard for decisions
Android App Links: Android's verified links: https URLs on a domain you own that open your app directly once Android has checked a file on that domain.In Navigation and app structure
annotated tag: A tag stored as its own object, with the tagged object's ID, the tag name, the tagger, a date and a message. Created with git tag -a.In Inside the .git folder: objects, refs and HEAD
anomalies: A wrong result that can appear when transactions run at the same time, but could never appear if they ran one after another.In Transactions, ACID, isolation and locks
anomaly check: A check that compares a measure, such as total sales or the share of empty values, with its usual range and flags values far outside it.In Fresh, tested, trusted
Anomaly detection: Flagging records that look unlike the rest, often without labelled examples of what is wrong.In Framing a problem as a model
ANR: Application Not Responding: Android shows a dialog when the app's main thread is blocked, for example when it ignores a touch for 5 seconds.In Testing and releasing
Anscombe's quartet: Four datasets published by Francis Anscombe in 1973 with the same means, variances, correlation and fitted line, but completely different shapes when plotted.In Regression and correlation
Anti-corruption layer: A translation layer that keeps a legacy or vendor model from leaking into your own design. Legacy migration
API: Application programming interface: a set of web addresses a program can call to read or change data.In Data and security basics, Observability with OpenTelemetry
API client: A program that sends requests to an API and shows or uses the answers. A phone app, a web page and a test tool can all be API clients.In Testing APIs and data
API gateway: A single entry point in front of many services that routes each request to the right one and handles shared concerns like authentication and rate limits.In Monolith, modular monolith, microservices, Netflix: streaming to every screen
API key: A long secret code that lets a program use a service, such as a database or a payment provider. Whoever has it can act as you.In What AI coding tools can and cannot do
API server: A program on a server that receives requests from clients, applies the business rules, talks to the database and sends back answers.In What system design really is
API test: A test that sends real HTTP requests to the service and checks the response and the data it stored, with only outside services replaced.In Unit and API tests
APNs: Apple Push Notification service: the only way to send push notifications to iPhones. FCM can forward to it for you.In Real-time updates: polling, WebSockets, SSE and push, Device features
appetite: The amount of time you decide a problem is worth before you plan the work. The time stays fixed and the scope changes to fit it.In From idea to one-page spec, Outcome roadmaps
application inventory: A list of every application that supports the business, with the same facts recorded for each one: owner, users, capabilities, cost, technology, support dates and the data it owns.In Application portfolio
ArchiMate: A modelling language for enterprise architecture from The Open Group, with standard element types, layers and relationships. The current version is ArchiMate 3.2, from 2022.In Frameworks and governance
architect elevator: Gregor Hohpe's picture of the architect's job: travelling between the executives on the top floor and the engineers in the engine room, and translating between them.In What architects actually do
architecturally significant: A decision that shapes the structure or qualities of a system and would be costly to reverse, so it deserves options, a written record and a review.In What architects actually do
architecturally significant decision: A decision that has a measurable effect on the structure of the system or its quality attributes, and that is expensive to change later.In Decisions, diagrams and reviews
architecturally significant requirements: A requirement that has a big effect on the structure of the system, so getting it wrong is expensive to fix later.In Quality attributes and constraints
architecture decision records: Architecture decision record: a short note that records one significant decision with its context, the decision, its status and its consequences.In Frameworks and governance
Architecture governance: The light rules and reviews that keep decisions consistent across teams. Governance
architecture principles: A rule that guides architecture decisions across the organisation, written with a statement, a rationale and implications.In Frameworks and governance, Technology strategy
architecture quantum: The smallest part of a system that can be deployed on its own and has its own quality-attribute needs. The term comes from Building Evolutionary Architectures and Fundamentals of Software Architecture.In Choosing an architecture style
architecture style: A family of architectures that share the same kinds of parts and the same rules for how those parts may connect, such as microservices or event-driven.In Choosing an architecture style
argon2id: A password hashing algorithm designed to be slow and to use a lot of memory, so guessing billions of passwords on graphics cards is expensive. The recommended choice today.In Security essentials
arguments: The things a command works on, such as file names or a branch name, written after the command.In The terminal without fear
ARIA: Accessible Rich Internet Applications: a W3C standard set of HTML attributes, such as role, aria-label and aria-live, that add to or change what the accessibility tree says.In Accessible, fast pages
ARIMA: AutoRegressive Integrated Moving Average: a classic statistical model that forecasts a series from its own past values and past errors.In Forecasting demand
arrow function: A short function syntax, (a, b) => a + b, that has no this, arguments or prototype of its own and cannot be called with new.In Functions, scope and closures
artifact: A file or folder saved from a workflow run, such as a build output or a test report, that you can download or pass to a later job.In GitHub Actions: continuous integration, Continuous delivery, Experiments, versions and the model registry
As-is and to-be: The process as it works today and as it should work after the change. Process models
as-is model: A model of how the work is really done today, including workarounds, before any change.In Modelling processes
as-of date: The date a prediction is made for. Features may only use data from before it, and the label only uses data from after it.In Classification and churn
as-of rule: The rule that a feature for a prediction made at a given moment may only use data that existed, in its final form, at that moment.In Features and leakage
assertion: The signed statement the IdP sends to the app about who the user is, when they logged in and which groups they are in.In Accounts, SSO and MFA, Testing APIs and data
asserts: A statement in a test that a value must be something specific, such as expect(total).toBe(7200). If it is not, the test fails.In Errors, debugging and testing
asset: Anything of value to an organisation that could be harmed: data, money, systems, people's time or its reputation.In Thinking like an attacker
Asset register: The list of every device with its serial, owner, location and status. Device fleet
asset tag: A short unique label, often a barcode sticker, that the company puts on each device so it can be tracked.In Looking after 1,400 devices
assistive technology: Hardware or software that helps people with disabilities use a device, such as a screen reader, a magnifier, a switch or voice control.In Accessibility, devices and the real world
association: In TAO, a typed, directed edge from one object to another, such as user 7 likes post 9, with a timestamp and optional fields.In Facebook: the social graph, TAO and memcache at planet scale
assume breach: Working as though an attacker is already present in the network, rather than only trying to keep them out, so you actively look for them.In Threat hunting
assumption test: The smallest piece of work that shows whether one belief behind an idea is true, with a pass mark set in advance.In Testing ideas cheaply
Asymmetric encryption: Encryption with a key pair: a public key anyone may have and a private key only the owner holds. What one key does, only the other can undo.In Cryptography without the maths
async iteration: Looping over values that arrive one at a time, asynchronously, with for await...of, often from an async generator function.In Async JavaScript: callbacks, promises and async/await
asynchronous: The primary confirms a COMMIT as soon as its own disk has the WAL, without waiting for any replica.In Replication and read replicas, Queues, background jobs and pub/sub
asynchronous replication: Asynchronous replication: the primary confirms a commit as soon as it is saved locally and sends it to replicas afterwards, so replicas can be slightly behind.In CAP, PACELC and consistency
at least once: A delivery promise: every job is processed, but some may be processed more than once, for example after a worker crashes before it reports success.In Background jobs and idempotency
at-least-once delivery: A delivery promise: every message will be processed, but some may be processed more than once.In Queues, background jobs and pub/sub, Case study: syncing POS sales from hundreds of stores
ATAM: Architecture Tradeoff Analysis Method: the Software Engineering Institute's method for checking an architecture against prioritised quality attribute scenarios to find risks and trade-offs.In Decisions, diagrams and reviews
atomic network: The smallest network that is stable and can grow by itself, such as one neighbourhood, one campus or one team.In Marketplaces, platforms and moats, Growth loops and local go-to-market
atomic notes: Research notes that each hold one observation and the ID of the source it came from.In From research to opportunities
atomic publish: A change that happens completely or not at all, so no reader can ever see it half done.In Serving: batch and online
attack chain: The ordered steps an attacker goes through, from first getting in to reaching their goal. Each step leaves traces and each can be blocked.In How breaches really happen
attack graph: A map of accounts, groups, computers and their rights, drawn as a graph so you can find the shortest chain of steps from where you are to a target such as Domain Admin.In Privilege escalation and lateral movement
attack surface: All the places where an attacker could try to get into a system or get data out of it: apps, APIs, networks, accounts, devices and suppliers.In Thinking like an attacker, Reconnaissance
Attack surface management: The ongoing practice of discovering, listing and reducing an organisation's own externally visible assets, using the same passive and active methods an attacker would.In Reconnaissance
attributes: Key and value pairs on a span, such as the model name or token count, that you can filter and group by later.In Observability for LLM systems, Observability with OpenTelemetry
audit and risk committee: A smaller group of board members who look closely at financial reporting, internal controls, audits and the main risks before the full board meets.In The board and the executive team
Audit log: An append-only record of who ran which action, on what target, when, and with what result.In Toil and automation
audit trail: A permanent record of every action: what triggered it, what the system read and proposed, which checks passed, who approved it, what was done and what happened next.In AI agents in operations
authenticated mode: An encryption mode that also detects any change to the ciphertext, so tampered data fails to decrypt instead of returning garbage.In Cryptography without the maths
Authentication: Checking who the caller is, for example with a password, an OTP or a signed token. Often shortened to AuthN.In Security essentials, Who are you and what may you do?, Data and security basics, Identity: passwords, MFA and passkeys, Broken access control
Authentication and authorization: Authentication proves who you are; authorization decides what you may do. Most breaches of web apps are authorization bugs. Access control
Authorisation: Deciding what a signed-in person is allowed to see and do, usually from their role.In Identity: passwords, MFA and passkeys, Broken access control, Ethics, law and scoping
Authorization: Checking what an authenticated caller is allowed to do or see. Often shortened to AuthZ.In Security essentials, Who are you and what may you do?, Data and security basics
auto_explain: A module shipped with Postgres that writes the execution plan of slow statements to the server log automatically.In Finding and fixing slow queries
auto-instrumentation: Code that patches common libraries such as http, express, pg and ioredis at startup so they create spans automatically, without changes to your code.In Observability: logs, metrics and traces
automation bias: The habit of accepting a machine's suggestion without checking it, especially when it is usually right.In Picking the right problems for AI, Responsible ML in a business
autonomy: How much an AI system may do without a person: from only answering, to suggesting, to acting after approval, to acting alone within limits.In AI agents in operations
Autoscaling: The platform adds or removes copies of your app automatically, based on a metric such as CPU use or request rate.In Scaling up vs scaling out
Autovacuum: The background process that runs VACUUM and ANALYZE on a table once enough rows have changed.In Finding and fixing slow queries
Availability: The share of time (or requests) a service works. 99.9% availability is called three nines.In Measuring speed and reliability: p99, SLOs and nines
availability zones: A separate data centre inside one cloud region, with its own power and network, so a failure in one zone does not take down the others.In Capacity, resilience and game days
Average basket: Total sales divided by the number of checkout transactions in the same period: how much a shopper spends in one visit.In Asking a question data can answer
average of averages: Taking the mean of several group averages without weighting each by its size, which gives small groups as much say as big ones.In Pivot tables and summaries
AWS Trainium: A chip designed by Amazon Web Services for training and running machine learning models.In Claude: how Anthropic serves Claude at scale
AWS Well-Architected Framework: AWS's set of design principles and review questions, grouped into six pillars, for judging a workload in the cloud.In Cloud architecture and cost

B

B-tree: Balanced tree: the default Postgres index. Sorted keys are stored in pages arranged as a shallow tree, so any key is found by reading one page per level.In Indexes from the inside
BABOK Guide: A Guide to the Business Analysis Body of Knowledge, published by the International Institute of Business Analysis (IIBA). Version 3 came out in 2015.In What business analysts do
back-of-the-envelope math: A quick, rough calculation, the kind you could write on the back of an envelope, to find the size of a number to within about 2 to 10 times.In Speed numbers and back-of-the-envelope math
back-translation: Translating a translated text back into the original language to check that the meaning survived.In User interviews that don't lie
backbone: The row of big user activities across the top of a story map, in the order a user does them.In User stories and acceptance criteria
backend: Where Terraform stores the state file, for example an S3 bucket, a cloud storage bucket or a managed service, instead of the local disk.In Infrastructure as code, Inside a PostgreSQL server
backend-as-a-service: A ready-made back end you rent: database, sign-in and file storage, reached over the internet, so you write little or no server code yourself.In Data and security basics
backfill: Filling a new column or table with values for rows that already existed before the change.In Changing the schema without downtime, Fresh, tested, trusted, From notebook to pipeline, Data and feature pipelines
background jobs: A unit of work saved in a queue and done later by a separate worker process, after the HTTP request has already been answered.In Background jobs and idempotency
Background location: Permission to read the phone's position while the app is not on screen, for example while another app is open or the screen is off.In Device features
background processes: Server processes that do housekeeping for everyone, such as writing pages to disk, flushing the log and starting vacuum.In Inside a PostgreSQL server
backlog: The ordered list of work a team might do next, such as features, fixes and improvements.In What a product manager actually does
Backlog refinement: The ongoing work of breaking backlog items into smaller ones and adding detail, such as examples, estimates and order, so they are ready to build.In What business analysts do
Backpressure: Signals that slow down or reject new work when a system cannot keep up, so it does not drown.In Queues, background jobs and pub/sub
backtest: Replaying the past: standing at each past prediction time, forecasting with only the data available then, and comparing with what happened.In Framing a problem as a model
Backtesting: Testing a forecast by replaying the past: train up to a date, forecast after it, repeat. Forecasting
backup manifest: A file listing every file in a base backup with its size and checksum, so a tool can later prove nothing is missing or changed.In Backups you have actually restored
bad strategy: Rumelt's name for plans that look like strategy but are not: fluff, failing to face the challenge, mistaking goals for strategy, and objectives that cannot be achieved or do not address the problem.In Technology strategy
Baggage: Key-value pairs that OpenTelemetry passes to every downstream service in a baggage header, alongside the trace context.In Observability with OpenTelemetry
bagging: Training each model on a random sample of rows drawn with replacement, then averaging the models. Short for bootstrap aggregating.In Training and comparing models
bake time: The time an update runs on a wave before you judge it, long enough to include real trading hours.In Looking after 1,400 devices
bake-off: A fair comparison of several options on the same test cases and traffic, measuring quality, cost, latency and upkeep, so the decision follows the evidence.In Prompting, retrieval or fine-tuning?
balance sheet: Also called the statement of financial position: what a company owns (assets) and owes (liabilities) on one date, plus the owners' share (equity).In The three statements
balancing algorithm: The rule a load balancer uses to choose which backend server gets the next request.In Load balancers
bandwidth: The amount of data your app sends to users, measured in gigabytes per month.In Ship it and know your limits
banner: A short line a service sends when you connect, often naming the software and its version.In Attacking networks and services
bare repository: A repository with only the .git data and no checked-out working folder. Servers like GitHub store repositories this way.In GitHub: remotes, clone, push, pull and authentication
base backup: A copy of the whole data directory taken while the server runs, plus the WAL needed to make it consistent. It is the starting point of every physical restore.In Backups you have actually restored
base branch: The branch a pull request will be merged into, usually main.In Pull requests and code review
base rate: How common something is in the whole population before any test or alert, such as the share of refunds that are fraudulent.In Traps that fool smart people, Classification and churn
base value: The model's average prediction over the data used as background, the starting point of a SHAP explanation.In Explaining and choosing models
base62: Writing a number with 62 symbols: 0-9, a-z and A-Z. Each character holds one of 62 values, so 7 characters give 62 to the power 7, about 3.5 trillion codes.In Case study: a URL shortener
base64: A way to write any bytes as plain letters and digits. Anyone can decode it in one command; it hides nothing.In Kubernetes for application teams
base64url: A way of writing bytes as letters, digits, dash and underscore so they fit safely in URLs and headers. It is an encoding anyone can reverse, not encryption.In Authentication, sessions and tokens
baseline: The value of a metric before you change anything, so you can tell later how much it moved.In From idea to one-page spec, Framing a problem as a model, Releasing safely, Security in the pipeline and in AI features, From change control to acceptance, Business cases
baseline architecture: A description of the architecture as it is today: the capabilities, applications, data and technology in use now.In Target state and roadmap
Batch ETL: Extract, transform, load: a scheduled job that reads data from the source, reshapes it, and writes it to the warehouse, for example every night.In Beyond SQL: choosing the right data store
batch interface: A way to submit many model requests at once and collect the results later, within a window such as 24 hours, at a lower price than live calls.In Cost and latency at scale
Batch scoring: Computing predictions for every case you can list in advance, on a schedule, and storing them in a table for later reads.In Serving: batch and online
batches: Grouping several state updates into one re-render instead of rendering after each one.In State, effects and where data should live
Batching: Giving one rider two or more orders on the same trip, to save rider time at the cost of a slightly later arrival for some orders.In Teardown: ride-hailing and food delivery
BATNA: Best alternative to a negotiated agreement: what you will do if you do not reach a deal.In Stakeholders, negotiation and leading without authority
BEAM: The Erlang virtual machine. It schedules millions of lightweight Erlang processes across all CPU cores.In WhatsApp: billions of messages with a tiny team
bearer token: A token sent in the Authorization header as Bearer followed by the token. Whoever holds it can use it, like cash.In Who are you and what may you do?
Benjamini-Hochberg: A method that ranks p-values from smallest to largest and compares the k-th with k divided by the number of tests, times the chosen rate; it controls the expected share of false discoveries.In Traps that fool smart people
bi-encoder: A model that embeds the question and each document separately, so document vectors can be computed once and searched fast.In Retrieval-augmented generation
bias: A systematic error that pushes an estimate the same way every time, for example because the sample favours certain units.In Samples, uncertainty and confidence intervals, Framing a problem as a model, Training and comparing models, Forecasting demand
Bikeshedding: Spending a long time on a trivial topic, such as naming, because everyone has an opinion about it, while hard topics get little attention.In Running a design review
bimodal: Having two peaks. Often a sign that two different groups are mixed in one column.In Describing data honestly
binary files: A file that stores bytes meant for a program, not for people: images, fonts, zip files, compiled apps. Opening it in an editor shows unreadable symbols.In How code becomes a running program
birthright access: The default access every person in a given job gets automatically, without a request.In Accounts, SSO and MFA
bisection: A search that keeps halving an interval known to contain the answer, until the interval is tiny.In Business cases
bitrates: How many bits per second a video stream uses. More bits usually means better picture but a bigger download. Measured in kbps or Mbps.In Netflix: streaming to every screen
BIZBOK Guide: A Guide to the Business Architecture Body of Knowledge, from the Business Architecture Guild: practical guidance on capabilities, value streams, information, organisation and how to use them.In Frameworks and governance
blameless: A review that looks for the conditions that made an error possible and likely, and does not punish the people who made the decisions.In Postmortems and learning, The board and the executive team
blameless post-incident review: A review that looks at how the system and process let the incident happen, not at who to blame, so people tell the truth.In Incident response
Blameless postmortem: A written incident review focused on causes in the system, not on who to blame. Postmortems
blameless review: A review after an incident that looks for the causes in systems, processes and decisions rather than blaming individuals, so that people share what really happened.In Risk, security and compliance
blast radius: How much can break if one change goes wrong.In Infrastructure as code, Releasing safely, Capacity, resilience and game days, Triage and investigation
blended CAC: One CAC figure for all channels together: total spend divided by all new customers. It hides channels that lose money behind ones that do well.In Unit economics
Blind receiving: Receiving where staff count what arrived without seeing the expected quantity, so the count is independent; the system compares it with the order afterwards.In Designing for frontline work
bloat: Wasted space in a table or index from dead rows that were not cleaned up in time. It makes scans read more pages and the cache hold less useful data.In Running Postgres in production, Upgrades, maintenance and on-call
blob report: Playwright's raw report format: a zip of results per shard that merge-reports turns into one HTML, JUnit or other report.In Stable suites in CI
block-scoped: Visible only inside the nearest pair of curly braces where the variable was declared, such as a loop body or an if block.In JavaScript in one sitting
blocker: A usability problem that stops the user finishing the task, or makes them finish it wrongly, such as sending the wrong quantities.In Screens and flows
blocklist: A list of passwords that must be refused, such as ones seen in data breaches, very common ones, and words tied to the service or the user.In Identity: passwords, MFA and passkeys
Blue-green: Running two full copies of production, blue and green. One is live; the new version starts on the other, and traffic switches over in one step.In Shipping safely: containers, CI/CD and releases, Releasing safely
Blue-green deployment: Running the new version beside the old one and switching all traffic at once. Safe releases
BM25: Best Matching 25: a standard keyword ranking formula that rewards rare words and repeated matches, and adjusts for document length.In Retrieval-augmented generation
Bonferroni: Divide the significance level by the number of tests, so the chance of even one false alarm across all of them stays at or below the original level.In Traps that fool smart people
bootstrap: A way to estimate uncertainty by resampling your own data with replacement many times and looking at how much the result varies.In Samples, uncertainty and confidence intervals
Borg: Google's internal cluster manager. It decides which machine runs which task, restarts failed tasks, and packs many jobs onto shared machines.In Google Search: crawling, indexing and answering in a fraction of a second
bottleneck: The one part of a system that limits how much the whole system can handle, like a single narrow door at a busy shop.In What system design really is
bottom-up adoption: A way a product spreads through a company: one team starts using it for free, colleagues join, and only later does a buyer pay for it company-wide.In Teardown: tools businesses run on
bottom-up size: Estimating a market by multiplying real counts of buyers, how many will use it, how often and at what price, instead of taking a share of a big industry number.In Idea lab: judging and pitching ideas
boundary: The value where a rule changes its answer, such as 2,500 cents for free delivery. Bugs cluster on and around it.In Unit and API tests
Boundary value: A value at the edge of a rule, such as the maximum quantity, where off-by-one bugs live. Test design
Boundary value analysis: A test technique that tests the smallest and largest values of each ordered partition, and their neighbours, because mistakes cluster at the edges.In Designing tests that find bugs
bounded context: A part of the system where one model and one vocabulary apply. The same word can mean different things in different contexts.In Monolith, modular monolith, microservices
box plot: A chart that shows the median, the middle half of the values as a box, and the spread beyond it as lines, so many groups can be compared side by side.In Charts that answer the question
BPMN: Business Process Model and Notation, a standard diagram language for processes published by the Object Management Group.In Modelling processes
branch: A name that points to one commit. When you commit on a branch, the name moves forward to the new commit.In Branches and merging, Screens and flows, Reading and fixing what the AI wrote
Branch by abstraction: Replacing a part of a system gradually: put an interface in front of the old code, move all callers to the interface, build the new implementation behind it, switch over, then delete the old code.In Evolving a legacy system
branch protection rules: The older GitHub feature for protecting one branch pattern with rules. Rulesets replace and extend it.In Branching strategies, releases and versions
breadcrumb: The path of headings above a piece of text, such as Delivery > Festival week > Later slots, so the text makes sense alone.In Retrieval-augmented generation
break-even: The volume at which two options cost the same; here, the monthly answers at which a fixed server cost equals the per-answer API bill.In How AI features work, just enough for PMs, Telling the story
break-glass account: An emergency admin account that does not depend on SSO, kept locked away and used only when normal login is broken.In Accounts, SSO and MFA
breakpoint: A marker on a line of code. When the program reaches that line under the debugger, it pauses there.In Errors, debugging and testing
bridge: A table or chart that starts from one number, lists every difference with its amount, and ends exactly at another number.In Metrics everyone agrees on, Incidents, problems and changes, Leading an incident
BRIN: Block Range INdex: stores only the minimum and maximum value for each group of table pages. Tiny, and useful when the table's physical order follows the column, such as time in an append-only table.In Indexes from the inside
broadcast address: The last address in a subnet. A message sent to it reaches every device on that subnet.In Networks from the till to the data centre
Broken Object Level Authorization: Broken Object Level Authorization: the API acts on a record because the caller knows its id, without checking the caller may access that record. Number one on the OWASP API Top 10 (2023).In Who are you and what may you do?
Brooks's law: Fred Brooks's rule that adding people to a late software project makes it later, because of ramp-up time and extra communication.In Managing up and across
browser console: A panel in the browser developer tools where you can type JavaScript and where web pages print their errors and messages.In Reading and fixing what the AI wrote
browser context: A fresh, isolated browser profile inside one browser process, with its own cookies, local storage and cache. Playwright creates a new one for every test.In End-to-end tests with Playwright
Brute force: Trying many passwords against one account until one works.In How breaches really happen
bubbling: After an event fires on an element, it travels up through every parent to the top of the document, running each one's listeners on the way.In JavaScript in the browser
buckets: A named container for objects, with its own access rules and settings.In Files, object storage and CDNs
budget: The plan of income and spending for the next financial year, broken down by team and by month, approved by the board.In Budgets, capex and opex
buffer: Video downloaded ahead of what is being shown. If the network stalls, playback continues from the buffer. When the buffer is empty, the spinner appears.In Netflix: streaming to every screen
bug life cycle: The set of statuses a bug report passes through, from first logged to closed, and the rules for moving between them.In Bugs, reports and acceptance
build cache: Docker's store of layers from earlier builds. A step whose instruction and inputs are unchanged reuses its old layer instead of running again.In Containers from the inside
build context: The set of files sent to the builder with docker build, usually the current folder minus whatever .dockerignore excludes.In Containers from the inside
built: The step that turns your source files into the files a server or browser can run, for example by installing packages and bundling JavaScript.In Ship it and know your limits
bulkhead: A limit that keeps one dependency or feature from using up all shared resources, such as connections, threads or memory.In Failure is normal: timeouts, retries, backoff and circuit breakers
bullet graph: A compact bar that shows a measure against its target mark and shaded bands for poor, fair and good ranges; designed by Stephen Few to replace gauges.In Designing a dashboard for decisions
bundle: One file that contains all of your app's JavaScript, ready for the phone to run.In Building the app
bundler: A tool that follows the imports in your code and joins many files into a few optimised files for the browser.In How code becomes a running program
burn rate: How fast you are spending your error budget compared with the allowed pace. A burn rate of 10 means the whole month's budget would be gone in about 3 days.In Observability: logs, metrics and traces, Alerting that does not burn people out
bus factor: The number of people who could leave or fall ill before nobody on the team understands a part of the system.In What a tech lead is (and is not)
bus matrix: A grid with one row per business process (fact table) and one column per dimension, with a mark where they meet; columns with several marks must be conformed.In From questions to a reporting model
business analyst: A person who finds out what an organisation really needs, defines the change that would meet that need, and checks that the change delivers value.In What business analysts do
Business architecture: The part of enterprise architecture that describes the business itself: what it does, how it delivers value, the information it uses and how it is organised.In Why enterprise architecture exists
business capability: Something the business does, described in business words, such as ordering, stock keeping, pricing or supplier management.In Monolith, modular monolith, microservices, Capability maps
business case: A short written argument that an investment is worth making: what it costs, what it returns and when, how sure you are, and why it beats the alternatives.In Business cases a managing director will sign, Business cases
business date: The calendar day a sale belongs to for reporting, decided by the store's local time, not by the server's clock or by UTC.In Working with data: arrays, Maps, JSON, dates and money, Metrics everyone agrees on
Business email compromise (BEC): Business email compromise: fraud where an attacker uses a fake or taken-over business email account to trick staff or customers into sending money or data.In How breaches really happen
Business fit: How well an application supports the capabilities and users it serves today and the goals of the next few years, judged from evidence such as errors, workarounds and coverage.In Application portfolio
business model: How a product earns money or saves it: who pays, for what, and when.In How products create value
business process: An activity the company performs that creates data, such as selling at a till, taking an app order or counting stock.In From questions to a reporting model
business rules: A rule that comes from how the business works, such as a minimum order value or a delivery cut-off, rather than from the technology.In A spec the AI can follow, Modelling processes, Requirements that work, User stories and acceptance criteria
Business-logic flaws: A flaw in the rules of a feature rather than in its code: the application does exactly what it was told, but the sequence or the values should not have been allowed.In Attacking web applications
bytecode: A compact list of simple instructions for a virtual machine. Faster to run than source text, but not yet CPU machine code.In How code becomes a running program, Performance and memory in JavaScript

C

%C&A: Percent complete and accurate: the share of work a stage passes on that the next stage can use without correcting, adding or clarifying anything.In Value streams and processes
C4 model: Simon Brown's way of describing software as a model with four levels of zoom: system context, containers, components and code.In Decisions, diagrams and reviews
CAB: Change advisory board: a group (often a weekly meeting) that reviews risky changes before they happen. ITIL 4 calls whoever approves a change the change authority.In Incidents, problems and changes
CAC: Customer acquisition cost: what you spend to win one new customer, such as ads, referral rewards and sign-up offers, divided by the new customers those channels brought in.In Unit economics: does each order make money?, Unit economics
CAC payback: How many months of a customer's contribution it takes to earn back what you spent to acquire them.In Unit economics: does each order make money?, Unit economics
cache: A fast store holding copies of data that is slower to fetch or compute from its real source.In Caching
cache hit: A request where the data was found in the cache.In Caching
cache hit rate: The share of input tokens read from the provider's prompt cache instead of processed fresh.In Observability for LLM systems, Cost and latency at scale
Cache invalidation: Marking cached server data as stale so it is fetched again. Server data
cache key: The parts of a request a cache uses to decide whether two requests are for the same thing, usually the host and path, sometimes query strings or headers.In Files, object storage and CDNs, Loading server data
cache miss: A request where the data was not in the cache, so it had to be fetched from the real source.In Caching, Netflix: streaming to every screen
cache mount: A RUN option that mounts a folder which survives between builds on the same builder but is never stored in the image.In Containers from the inside
cache stampede: Many requests miss the cache for the same key at the same moment and all load it from the database together. Also called thundering herd or dog-piling.In Caching
Cache-aside: The app checks the cache first, and on a miss reads the database and fills the cache with a TTL. Caching
Cadence: Uber's open-source engine for long-running workflows. You write the steps as normal code and the engine saves progress so a crash resumes where it stopped.In Uber: matching riders and drivers in real time
Calculated column: A column computed once per row and stored in the model. Warehouse models
calculated field: A field you add to a pivot table with a formula, such as sales divided by transactions, computed from the other fields in each cell.In Pivot tables and summaries
calibrated: When a model's predicted probabilities match how often the event really happens: of customers scored 0.3, about 30% churn.In Classification and churn
calibrated trust: Trust that matches how reliable a system really is: relying on it where it is usually right and checking it where it is not.In Designing AI features people can trust
Calibration: A meeting where managers compare proposed ratings and promotions against the same ladder, so that a level or rating means the same thing on every team.In Performance and growth
call stack: The engine's list of functions that are currently running. Calling a function pushes it on top; returning pops it off. Only the top one is executing.In The event loop
callback: A function you hand to another function, to be called later when the work is finished.In Async JavaScript: callbacks, promises and async/await
callback hell: Deeply nested callbacks, each step inside the previous one, which makes code hard to read and errors easy to miss.In Async JavaScript: callbacks, promises and async/await
Canary: Sending a small share of real traffic, such as 5%, to the new version first, and increasing it only if its error rate and latency stay healthy.In Shipping safely: containers, CI/CD and releases, Claude: how Anthropic serves Claude at scale, Cost, latency and safety, Evals that never stop, Releasing safely
canary token: A unique random string placed in a system prompt or document. If it ever appears in an output or in a log outside its place, you know that text leaked.In Guardrails and red-teaming
cannibalisation: When a promotion on one product takes sales from similar products instead of creating new demand.In Forecasting demand
canonical: The one URL Google picks to show for a group of pages with the same or very similar content.In Google Search: crawling, indexing and answering in a fraction of a second
canonical URL: The one URL that names a resource, such as /v1/orders/90412. Other URLs may link to it, but this is its address.In Designing an API people can use
CAP theorem: During a network partition, a distributed system must choose between staying consistent and staying available. CAP and consistency
Capabilities: Something the business does or must be able to do, named as a noun such as Partner credit management, independent of who does it or which system supports it.In Why enterprise architecture exists
Capability: Something the business must be able to do, independent of who does it or which system supports it. Capability maps
capability-based planning: Planning change as improvements to named business capabilities, and judging initiatives by which capabilities they lift and by how much.In Capability maps
Capacity planning: Measuring how fast a resource such as disk, connections or CPU is being used up, and acting before it runs out.In Upgrades, maintenance and on-call
capex: Capital expenditure: money spent on assets that last for years, such as devices or servers, recorded on the balance sheet.In Business cases a managing director will sign, The three statements, Budgets, capex and opex
capitalisation: Recording a cost as an asset on the balance sheet instead of an expense, then spreading it over its useful life.In The three statements
capitalised: Recorded as an asset on the balance sheet instead of as an expense, then amortised over its useful life.In Budgets, capex and opex
card data environment: The people, processes and devices that store, process or send card data, plus anything that can connect to them without a firewall in between.In Networks, endpoints and the cloud
cardinality: The number of unique label combinations a metric has. High cardinality, such as one series per user, makes a metrics system slow and expensive.In Observability: logs, metrics and traces, Observability for LLM systems
cardinality budget: An agreed maximum number of time series a service or metric may create, checked in review and enforced by limits.In Observability with OpenTelemetry
Career framework: Written expectations for each level, used for growth, promotion and pay. Performance
career ladder: A written description of the levels in a job family and what is expected at each one.In Growing as a PM, Performance and growth
cascade: A routing pattern where the cheap model answers first and its answer is accepted only if checks pass; otherwise the same request goes to a stronger model.In Cost and latency at scale
cash flow statement: The statement of where cash came from and went in a period, in three sections: operating, investing and financing.In The three statements
Cassandra: An open-source distributed database that spreads rows across many servers by a partition key, replicates each row to several of them, and stays writeable when some servers fail. It favours availability over strict consistency.In Netflix: streaming to every screen
cause: An option of the Error constructor, new Error(message, { cause }), that stores the original error inside the new one so no detail is lost.In Errors, debugging and testing, Alerting that does not burn people out
CD: Continuous delivery or deployment: every change that passes CI is packaged and released to staging, and then to production with or without a human approval.In Shipping safely: containers, CI/CD and releases
CD (continuous delivery or deployment): Automatically releasing tested changes to an environment, with or without a final approval. Deploying and securing with GitHub Actions
CD3: Cost of delay divided by duration: a way to order work so the most value is delivered soonest when one team does the items one after another.In Prioritisation that survives the meeting
CDC: Change data capture: reading a database's own change log to stream every insert, update and delete to other systems.In Integration and data ownership
CDC (change data capture): Reading a database's change log, such as the Postgres WAL, to stream every change to other systems. Data stores
CDN: Content delivery network: many cache servers spread around the world that keep copies of your files close to users.In Files, object storage and CDNs, Netflix: streaming to every screen
CDN (content delivery network): Servers close to users that keep copies of files and pages, so most requests never reach your origin. Files and CDNs
cell-based architecture: Splitting one service into many complete, independent copies (cells), each serving a slice of customers, so a failure hurts only one slice.In Amazon: the store that never closes
cells: One block of a notebook: some code or text that you run on its own, with its result shown right below it.In The same analysis in Python
censored demand: Demand you only see up to a cap: when an item runs out, recorded sales stop at the stock that was there, so true demand was at least that and possibly more.In Framing a problem as a model
central limit theorem: The result that the means of fairly large random samples follow a roughly normal (bell-shaped) distribution, whatever the shape of the data they come from.In Samples, uncertainty and confidence intervals
certificate: A signed file that proves a server owns a domain name. Browsers and phones trust it only if a known authority signed it.In What happens when you open an app or website, Networking you actually need, Cryptography without the maths
certificate authority: An organisation that checks a requester controls a name and then signs certificates for it. Browsers and phones trust a fixed list of root CAs.In Cryptography without the maths
certificate authority (CA): An organisation that phones and browsers trust to check domain ownership and sign certificates, for example Let's Encrypt.In Networking you actually need
Certificate transparency: Public, append-only logs of the TLS certificates that certificate authorities issue, created so mis-issued certificates can be caught; anyone can search them.In Reconnaissance
Certification: A metric or dataset that a named reviewer has checked against the agreed definition and marked as the official one to use.In Metrics everyone agrees on
CGNAT: Carrier-grade NAT: a mobile or internet provider shares one public IP address among many customers.In Load balancers, Case study: a URL shortener
cgroup: Control group: a Linux kernel feature that limits and measures how much CPU, memory and I/O a group of processes may use.In Containers from the inside
chain of custody: A documented record of who handled a piece of evidence, when, and what they did to it, so it can be trusted later.In Triage and investigation
champion: A trusted member of staff in a site who learns a change first, helps colleagues during rollout and sends feedback to the product team.In Change management and adoption, From notebook to pipeline
champion alias: A named pointer in the model registry, such as champion, that says which model version the pipeline and the service should load.In Monitoring, drift and retraining
change: Adding, changing or removing anything that could affect a live service: software, configuration, hardware or a network rule.In Incidents, problems and changes
Change advisory: The review that approves risky changes and their timing and rollback plans. Service desk
change board: The group or person with authority to approve or reject changes above an agreed size. In agile teams it is often the product owner for small changes and the sponsor for large ones.In From change control to acceptance
Change data capture: Change data capture: reading every insert, update and delete from the database's own log, such as the Postgres WAL, and sending each change to other systems within seconds.In Beyond SQL: choosing the right data store, Events, the outbox pattern and sagas
Change fail rate: The share of deployments that need an immediate fix, such as a hotfix or a rollback. One of DORA's four keys. Your team's engineering playbook
Change failure rate: The share of deployments that need immediate intervention afterwards, such as a rollback or a hotfix.In Team design and delivery health
change freeze: A period, such as a festival rush, when no planned changes or updates are allowed.In Looking after 1,400 devices, Incidents, problems and changes
change request: A written request to add, remove or change agreed scope, which is estimated for cost, time and risk before anyone decides.In Working with vendors and agencies, From change control to acceptance
channel: In EAS Update, a named stream such as preview or production that each build listens to for updates. A build only gets updates from its own channel.In Continuous delivery
chaos engineering: Testing a system by deliberately causing failures in production, such as killing servers or adding delay, to prove it survives them.In Netflix: streaming to every screen, Capacity, resilience and game days
Chargeback: Moving each team its share of cost into its own budget, so the team pays for what it uses.In Cloud and SaaS costs with FinOps
Chargeback and showback: Billing teams for their cloud use, or just showing them the cost. FinOps
charter: A short statement of the aim of an exploratory session: what you will explore, with what, and what you want to find out.In Exploratory testing
checkpoint: The moment when all dirty pages up to a point have been written to the data files, so older WAL is no longer needed for crash recovery.In Inside a PostgreSQL server
checks: In k6, a named true or false assertion on a response, such as status is 201. Failed checks are counted but do not stop the test or fail it unless a threshold uses them.In Performance and load testing
chi-square test: A test that compares observed counts with the counts you expected, adding up (observed - expected)² / expected over every group.In Designing and reading an A/B test
choreography: A saga where there is no central coordinator: each service listens for events and reacts by doing its step and emitting the next event.In Events, the outbox pattern and sagas
chunk: One piece of a document, small enough to match one topic, stored with its own vector and metadata.In Retrieval-augmented generation
churn: Customers who stop being customers. In an app without subscriptions it is a definition you choose, such as no order in the next 30 days.In Classification and churn
CI: Continuous integration: every push is automatically built and tested, so problems are found within minutes of the change.In Shipping safely: containers, CI/CD and releases, Transforming data in the warehouse, From notebook to pipeline
CI (continuous integration): Automatically building and testing every change, usually on each push and pull request. GitHub Actions: continuous integration
CI/CD: Continuous integration (merge small changes often, tested automatically) and continuous delivery or deployment (every good build can go, or goes, to production). Trusted CI
CIA triad: Confidentiality, integrity and availability: the three properties security tries to protect.In Thinking like an attacker
CIDR: Classless Inter-Domain Routing: writing a network as an address plus a slash and the number of bits that name the network, such as 10.20.142.0/24.In Networks from the till to the data centre
circuit breaker: A wrapper around remote calls that stops sending them for a while when most recent calls failed, and fails fast instead.In Failure is normal: timeouts, retries, backoff and circuit breakers, Postmortems and learning
circular import: Two or more modules that import each other, directly or through a chain, so one of them must run before the other has finished.In Prototypes, classes and modules
CISA KEV: The CISA Known Exploited Vulnerabilities catalogue: a public list of vulnerabilities confirmed to be exploited by attackers in the real world.In Attacking networks and services
Citus: An open-source Postgres extension that spreads tables across several Postgres servers and routes queries to the right shards automatically.In Sharding and consistent hashing
claims: One fact inside a token, such as sub (the user id), aud (who the token is for) or exp (when it expires).In Who are you and what may you do?, Authentication, sessions and tokens
Class imbalance: When one outcome is much rarer than the other, such as fraud. Churn
Classification: A prediction task whose output is a category or a yes or no, usually with a probability.In Framing a problem as a model
cleaning log: A short list of every change made to a dataset: what was wrong, how many rows, what you did, and the row count after.In Cleaning data in a spreadsheet
cleanup: The function an effect returns. React runs it before the effect runs again with new values, and when the component leaves the screen.In State, effects and where data should live
client: The program the user touches: a mobile app, a web page, a handscanner app. It sends requests and shows the answers.In What system design really is, What happens when you open an app or website
closed model: A load model where a fixed number of users each wait for a reply before sending the next request, so the request rate falls whenever the system slows down.In Performance and load testing
closing keyword: A word such as Fixes, Closes or Resolves followed by an issue reference in a PR description. When the PR merges into the default branch, GitHub closes the issue.In Pull requests and code review, Planning on GitHub: issues, labels and Projects
closure: A function together with the variables from the scope where it was created. The function can still read and change those variables after the outer function has returned.In Functions, scope and closures
CloudEvents: An open specification from the CNCF for describing events in a common way, so any tool can route and read them.In Integration and data ownership
cluster: A group of machines managed together by Kubernetes: a control plane that stores the desired state and makes decisions, plus worker nodes that run the containers.In Kubernetes for application teams
CNAME: Canonical name record: says this name is an alias of another name, which the resolver then looks up.In Networks from the till to the data centre
Code coverage: The share of code lines or branches that tests ran; it shows what is untested, not that tested code is right. Test strategy
code editor: An app on your computer for opening, reading and changing code files.In What AI coding tools can and cannot do
codebook: A short shared list of tags and their meanings, used to tag research notes and feedback the same way every time.In From research to opportunities
codec: Software that compresses video for storage and sending, and decompresses it for playback. H.264 (AVC), VP9, HEVC and AV1 are common video codecs.In Netflix: streaming to every screen
CODEOWNERS: A file in the repo that maps paths to the people or teams who own them. GitHub automatically requests their review when a PR touches those paths, and a ruleset can require their approval.In Your team's engineering playbook
coefficient of variation: The standard deviation of a set of numbers divided by their average. A small value means the numbers cluster tightly around the average, which is what regular beaconing looks like.In Threat hunting
cognitive load: How much a person or team has to hold in mind to do their work. Too much leads to slow, error-prone work.In Team design and delivery health
COGS: Cost of goods sold: the direct cost of what was sold. Three statements
Cohesion: How closely the things inside one module belong together; higher is better. Architecture styles
cohort: A group of users who started in the same period, such as everyone whose first order was in the week of 3 March, tracked together over time.In Product-market fit before growth, Unit economics
cohort table: A table with one row per group of users who started in the same period and one column per period since they started, showing the share still active.In Retention and cohorts
cold start: Predicting for a new product, store or customer that has little or no history.In Explaining and choosing models, Navigation and app structure, Fast on a cheap phone, Choosing an architecture style
cold start problem: The problem of starting a marketplace or network: neither side will join until the other side is already there.In Marketplaces, platforms and moats
collector: A service you run between your applications and your tracing backend that receives telemetry, processes it (redact, sample, route) and exports it.In Observability for LLM systems
collision: Two different inputs that end up with the same short code. The system must detect it and pick another code.In Case study: a URL shortener, Cryptography without the maths
column: One field that every row in a table has, such as name or price, with a fixed type such as text or whole number.In Data and security basics
Columnar warehouses: A database that stores each column separately and compressed, so a query that needs 3 of 20 columns reads only those 3. Very fast for analytics, slow for single-row updates.In Beyond SQL: choosing the right data store
command: A message asking one specific receiver to do something, named as an instruction, such as reserve-stock. The receiver may refuse.In Events, the outbox pattern and sagas
Command Palette: A searchable list of every command in VS Code and its extensions. Open it with Ctrl+Shift+P (Cmd+Shift+P on macOS).In VS Code essentials
command substitution: Writing $(command) so the shell runs the command and puts its printed output in that place.In Shell scripting for daily jobs
commit: One saved snapshot of the project in Git, together with its author, date, message and a link to the previous commit.In What Git really is: snapshots and the three areas, Building in small steps, Reading and fixing what the AI wrote
commit ID: The unique ID of a commit: a 40-character fingerprint (hash) calculated from the commit's snapshot, author, date, message and parent. Usually shortened to its first 7 characters.In What Git really is: snapshots and the three areas
commit SHA: The unique id git gives each commit, a long hexadecimal string usually shown by its first 7 characters.In Experiments, versions and the model registry
common schema: An agreed set of field names and meanings that logs from every source are mapped into, so one query can read them all.In Logs that matter
compaction: Replacing older parts of an agent's history with a short summary so the context stays small while key facts survive.In Memory, state and multi-step plans
comparator: A function given to sort that receives two items, a and b, and returns a negative number if a comes first, a positive number if b comes first, or 0 if they are equal.In Working with data: arrays, Maps, JSON, dates and money
comparison group: Stores (or customers) that did not get the change, measured over the same weeks, so season, weather and pay days affect both groups.In Telling the story
compensating actions: A step that undoes the business effect of an earlier, already committed step, such as releasing reserved stock or refunding a payment.In Events, the outbox pattern and sagas
compensating control: A different control that lowers the same risk while the normal one is missing, such as network isolation for a system that cannot take updates.In Risk, compliance and the first hour of an incident
compiler: A program that translates code into another form, such as machine code, bytecode or simpler JavaScript, before it runs.In How code becomes a running program
compliance policy: A set of conditions a device must meet, such as a minimum OS version, encryption on and a recent check-in. Devices that fail are marked non-compliant.In Looking after 1,400 devices
components: A function that takes data (props) and returns a description of the UI. React calls it again whenever its data changes.In JavaScript in the browser, Components, props and how React renders
composite action: A GitHub Action made of several steps defined in its own action.yml file. Its steps can call other actions.In Secure supply chain and delivery metrics
composite index: An index on two or more columns, sorted by the first column, then by the second within each value of the first, and so on.In Indexes from the inside
Composition: Building an object from small, independent pieces of behaviour, each passed in or mixed in, instead of inheriting everything from a parent class.In Prototypes, classes and modules
Computer Fraud and Abuse Act: US federal law that makes accessing a computer without authorisation, or beyond the access you were given, an offence.In Ethics, law and scoping
Computer Misuse Act: UK law that makes unauthorised access to computer material, and unauthorised acts that impair a computer, offences.In Ethics, law and scoping
Concentration risk: The risk that comes from depending on one supplier for something critical, so that their failure, price rise or exit hurts you badly.In Money: budgets, headcount and vendors
Concept drift: A change in the relation between inputs and the outcome, so the same inputs now lead to different sales.In Monitoring, drift and retraining
concerns: A question or interest a stakeholder has about the system, such as cost, risk, timing or who owns some data.In Why enterprise architecture exists
concierge test: A test where a person delivers the value by hand and the user knows a person is doing it.In Testing ideas cheaply
concurrency limit: The most tasks allowed to be running at the same moment. Extra tasks wait until a running one finishes.In Async JavaScript: callbacks, promises and async/await
conditional aggregation: An aggregate such as sum that only counts the rows meeting a condition, written in PostgreSQL as sum(x) FILTER (WHERE condition), so one query can produce one column per period or category.In Pivot tables and summaries
conditional sign-off: Approval to go live with a written list of known open issues, each with an owner and a fix date.In From change control to acceptance
cone of uncertainty: The idea that early estimates can be wrong by a large factor and become more accurate only as decisions remove unknowns.In Working with engineers
confidence interval: A range computed from a sample by a method that, over many samples, contains the true value a stated share of the time, usually 95%.In Samples, uncertainty and confidence intervals, Hypothesis tests without the fog
config plugin: A function that changes the generated native project during prebuild, for example to add a permission or set up a native SDK.In How React Native works
ConfigMap: A Kubernetes object holding plain, non-secret settings as key and value pairs, which pods read as environment variables or files.In Kubernetes for application teams
configuration profiles: A bundle of settings the management server pushes to a group of devices, such as Wi-Fi details, passcode rules or allowed apps.In Looking after 1,400 devices
Confirmation testing: Running the steps that showed a defect again on the fixed build, to confirm the failure is gone. Often called retesting.In Bugs, reports and acceptance
conflict: A place where two branches changed the same lines differently since their merge base, so Git cannot combine them automatically and asks you to choose.In Merge conflicts without panic
conflict markers: Lines that Git writes into a conflicted file to show the two versions: seven less-than signs, seven equals signs and seven greater-than signs.In Merge conflicts without panic
conformed: A dimension used by several fact tables with the same keys, attribute names and values, so their numbers can be compared side by side.In From questions to a reporting model
confounder: A third variable that affects both the x and the y, creating a correlation between them even if neither causes the other.In Regression and correlation
confusion matrix: A 2 by 2 table of predictions against reality: true positives, false positives, false negatives and true negatives.In Classification and churn
connection pool: A set of open database connections that an app reuses for many queries, so it does not open a new connection for every request.In Scaling up vs scaling out
connection pooler: A small server between the apps and Postgres that holds a few real database connections and lends them to many app connections in turn.In Running Postgres in production
Connection pooling: Sharing a small number of database connections among many app requests, for example with PgBouncer. Postgres ops
consensus store: A small, strongly consistent key-value store (here etcd or Consul) that the Patroni agents use to agree on which node holds the leader key.In Replication and failover
Consistent hashing: A way to map keys to servers on a circle of hash values, so adding or removing a server moves only the keys next to it, about 1/N of all keys.In Sharding and consistent hashing
constrained decoding: Limiting which tokens the model may pick at each step so that the output must follow a grammar or schema.In Prompts and structured output
constraint: A rule stored in the database, such as NOT NULL, UNIQUE, CHECK or FOREIGN KEY, that every insert and update must satisfy.In Modeling data in tables, Data model and API, Quality attributes and constraints
consumer: The side of an API call that sends requests and uses the responses, such as the Mangrove app.In Contract tests
consumer group: A named group of readers on a Redis Stream. Each entry goes to one reader in the group, and stays pending until that reader acknowledges it.In Case study: a URL shortener
consumer version selectors: Rules a provider uses to pick which consumer pacts to verify, such as the latest from the main branch or every version deployed or released.In Contract tests
consumer-driven contract: An approach where each client of an API writes down, as tests, exactly which requests it sends and which parts of the responses it reads; the API is then checked against all of these.In Contract tests, Integration and data ownership
container: In C4, an application or data store that must be running for the system to work, such as a web app, an API, a database or a queue. It is not the same as a Docker container.In Decisions, diagrams and reviews
container image: A packaged, read-only file system with your app, its runtime and its dependencies, plus the command to start it. Containers are started from images.In Shipping safely: containers, CI/CD and releases, Continuous delivery
Content Security Policy: Content Security Policy: an HTTP header that tells the browser which sources of scripts, styles and connections a page may use, so injected scripts are blocked.In Security and quality in JavaScript projects, Injection and unsafe input
content-addressable storage: Storage where the address (name) of each item is computed from the item's own content, so equal content gets the same address.In Inside the .git folder: objects, refs and HEAD
context: Everything the model is given for one request: your instruction, open files, instruction files, earlier messages and tool results. It has a size limit, and the model knows nothing outside it.In Working with AI coding assistants, What AI coding tools can and cannot do, Building in small steps, Shipping an agent safely
Context engineering: Choosing and arranging exactly what goes into each model call so the model has what it needs and little else.In Memory, state and multi-step plans
context propagation: Sending the trace id and parent span id to the next service, usually in the traceparent HTTP header, so all services add spans to the same trace.In Observability: logs, metrics and traces, Observability for LLM systems, Observability with OpenTelemetry
context window: The most text, counted in tokens, that a model can read in one call, including the instructions, the documents, the chat so far and its own answer.In How AI features work, just enough for PMs, How LLM APIs work, Designing AI features: LLMs, RAG and vector search
contextual inquiry: A field method where you watch people do their real work in their real place and ask about it as it happens, learning like an apprentice.In Watching the work
continuous batching: A serving method where the batch is updated after every model step: finished requests leave and waiting requests join at once, so GPU slots are never idle for long.In ChatGPT: serving a large language model to hundreds of millions
Continuous delivery: Continuous delivery or deployment: automatically releasing every change that passes CI to an environment, with production either automatic or behind an approval.In Deploying and securing with GitHub Actions, Continuous delivery
continuous deployment: Continuous delivery where every change that passes all checks goes to production automatically, with no human approval step.In Continuous delivery
Continuous integration: Continuous integration: automatically building and testing every change as soon as it is pushed, so problems are found within minutes.In GitHub Actions: continuous integration, Continuous delivery, Continuous integration that people trust
Continuous Native Generation: Continuous Native Generation: the android and ios folders are generated from app.json and config plugins whenever needed, instead of being edited by hand.In How React Native works
contract test: An automated check that two systems agree on the requests and responses between them, run against each system separately instead of against both together.In Contract tests, Managing up and across
contributing factors: Conditions that together made an incident possible or made it worse. None of them is enough alone; removing any of them changes the outcome.In Postmortems and learning
contribution: The part of a total change that comes from one row, such as one store or category, often shown as its share of the total change.In Pivot tables and summaries
Contribution margin: What one unit earns after all its variable costs: revenue per unit minus the costs that come with that unit. It is what each unit contributes towards fixed costs and profit.In Unit economics: does each order make money?
control: Anything that reduces a risk: a technical setting, a process step or a physical lock.In Thinking like an attacker
Control group: The group that does not get the change, so you can see what would have happened anyway. A/B test design
control plane: The part of a system that makes decisions: who you are, what you may watch, which server to use. It handles many small requests.In Netflix: streaming to every screen
Controlled input: An input whose value comes from React state and changes only through onChange. Forms
controllers: A loop inside Kubernetes that keeps comparing the desired state of some object with the actual state, and acts to close the gap.In Kubernetes for application teams
convenience sample: A sample made of whatever units were easy to reach, such as volunteers or the nearest stores; it can differ from the population in ways you cannot see.In Samples, uncertainty and confidence intervals
Conventional Commits: A commit message format: type(optional scope): short description, for example fix(receipt): show discount line. Tools can read it to build changelogs and version numbers.In Everyday Git: status, diff, log and good commits
Conway's law: The observation that a system's structure ends up mirroring the communication structure of the organisation that built it.In Monolith, modular monolith, microservices, Team design and delivery health
cookie: A small piece of text the server asks the browser to store and send back with every later request to that site.In What happens when you open an app or website, JavaScript in the browser
cooldown: A setting in Dependabot (cooldown) or Renovate (minimumReleaseAge) that waits a number of days after a package version is published before proposing it.In Secure supply chain and delivery metrics
Coordinated omission: A measuring error where the load tool waits for slow replies and so skips the requests it should have sent meanwhile, which makes slowness look smaller than users feel it.In Performance and load testing
copy-on-write: A file from a lower read-only layer is copied into the writable layer the first time a process changes it; the original stays untouched.In Containers from the inside
Core components: React Native's built-in native building blocks: View, Text, Image, TextInput, Pressable and lists. RN basics
core concept model: The Business Analysis Core Concept Model: six ideas (change, need, solution, stakeholder, value, context) that every piece of business analysis work involves.In What business analysts do
core loop: The repeating cycle of actions that delivers value to users and brings them back, so each turn makes the next one more likely.In How to break down any product
Core Web Vitals: Three metrics Google defines for real-user experience: LCP for loading, INP for responsiveness and CLS for visual stability.In Accessible, fast pages
Correctness: An SLI that counts whether the system produced the right result, not only a fast or successful-looking one.In How reliable is reliable enough?
CORS: Cross-origin resource sharing: browser rules that decide whether a web page from one site may call an API on another site. It protects users in browsers, not your API.In Security essentials, JavaScript in the browser
cosign: A tool from the Sigstore project that signs and verifies container images and other artifacts, with stored keys or with short-lived keyless certificates.In Secure supply chain and delivery metrics
cosine distance: One minus the cosine of the angle between two vectors: 0 means same direction (very similar), larger means less similar. pgvector writes it as the operator <=>.In Designing AI features: LLMs, RAG and vector search
cost centre: A unit in the accounts, usually a team or department, that costs are assigned to and that one owner answers for.In Budgets, capex and opex
cost model: A spreadsheet or small program that estimates the monthly bill from business numbers, such as orders a day and peak hours, with each assumption written down.In Cloud architecture and cost
cost of capital: In plain words, what the company pays, on average, for the money it uses: interest to lenders and the return shareholders expect.In Business cases
cost of change: How much time, money and risk it takes to undo or change a decision after the system has been built around it.In What architects actually do
cost of credit: What it costs a seller to wait for payment: the amount owed times the yearly cost of money times the days outstanding divided by 365.In Unit economics
Cost of delay: The value lost for each week (or month) that a piece of work is not yet live. It turns urgency into money.In Prioritisation that survives the meeting
cost of goods sold: What the goods Mangrove sold cost to buy from suppliers, including inbound freight, minus supplier discounts.In The three statements
Cost to serve: The full cost of handling one unit of work, such as one order, delivery, goods received note or support ticket, including staff time, systems and the cost of fixing errors.In Unit economics: does each order make money?, What business analysts do, Unit economics
counter: A metric that only goes up, such as total requests served. You look at how fast it grows.In Observability: logs, metrics and traces
counter-metric: A metric watched next to a target metric, chosen because it gets worse when the target is hit the wrong way.In North star, input and guardrail metrics
Coupling: How much one part must change when another changes; lower is better. Architecture styles
Coverage: The share of your code's lines and branches that ran at least once during the tests.In Errors, debugging and testing, Designing tests that find bugs, Cloud and SaaS costs with FinOps
coverage gaps: A place where an attack could happen but no log records it, so no rule or search could ever catch it.In Logs that matter
coverage heat map: A view of the ATT&CK matrix with each technique shaded by how well your detections cover it, used to find gaps.In Writing detections
Covering index: An index that also stores extra columns (with INCLUDE) so a query can be answered from the index alone, without reading the table.In Indexes from the inside
CrashLoopBackOff: A pod status meaning a container keeps crashing; the kubelet waits longer before each restart (10 s, 20 s, 40 s and so on, up to 5 minutes by default).In Kubernetes for application teams
CRDT: Conflict-free replicated data type: a data structure designed so that copies edited separately can always be merged automatically into the same result, whatever order the edits arrive in.In Offline-first apps and data sync
credential helper: A helper program that stores your Git login securely, for example in Windows Credential Manager or the macOS Keychain, so you do not type it each time.In GitHub: remotes, clone, push, pull and authentication
credential stuffing: Trying email and password pairs leaked from one site against the login of other sites, by automation, hoping people reused their passwords.In How breaches really happen, Identity: passwords, MFA and passkeys, Authentication, sessions and tokens
critical path: The chain of tasks that sets the earliest finish date; if any of them is late, the whole project is late.In What a tech lead is (and is not), The manager's job
critical ratio: Shortage cost divided by shortage cost plus excess cost: the share of days you should be able to meet demand.In Forecasting demand
CRM: Customer relationship management: the system marketing uses to send offers and messages to customers.In Responsible ML in a business
cron: The classic Linux scheduler. It reads lines of five time fields plus a command and runs the command at those times.In Shell scripting for daily jobs
cross-encoder: A model that reads the question and one document together and outputs one relevance score. More accurate than comparing vectors, but too slow to run on every document.In Retrieval-augmented generation
Cross-site scripting: Cross-site scripting: an attacker gets their own script to run inside your web page, in another user's browser, with that user's session and permissions.In Security and quality in JavaScript projects, Attacking web applications
Cross-validation: Training and testing several times on different splits to get a steadier score. Models
CRUD: Create, read, update and delete: the four basic things most apps do with records, such as adding, showing, changing and removing an order line.In What AI coding tools can and cannot do, A spec the AI can follow
crux: The one obstacle that, once dealt with, unlocks progress on most of the other problems.In Vision, strategy and why now
CSP (Content Security Policy): A response header that tells the browser which sources of script, style and frames to allow. JS security
CSPM: Cloud security posture management: tools that check cloud accounts against secure-configuration rules and report or fix drift.In Networks, endpoints and the cloud
CSRF: Cross-site request forgery: another website makes the user's browser send a request to your site, with the user's cookies attached, that the user never meant to send.In Authentication, sessions and tokens
CTE: Common Table Expression: a named sub-query written with WITH name AS (...) at the top of a query, used like a temporary table inside that one query.In SQL you'll actually use
CTE (common table expression): A named subquery written with WITH, so a long query reads in steps. Everyday SQL
ctid: A row's physical address in the table file: the page number and the slot inside that page, written like (36573,66).In Indexes from the inside
current directory: The folder the shell is in right now. Relative paths start from here. Also called the working directory.In The terminal without fear
cursor: A bookmark that says how far a reader has got. In sync, the last change number the device has applied, sent back to ask only for newer changes.In Offline-first apps and data sync
Cursor pagination: Paging by value: the server returns a cursor (often the last id seen) and the next page starts right after it, for example WHERE id > 4711.In APIs done right
custom format: pg_dump output format -Fc: one compressed file that pg_restore can read selectively and restore with several parallel jobs.In Backups you have actually restored
customer acquisition cost (CAC): Customer acquisition cost: what it costs, in marketing, rewards and sales time, to win one new customer.In Growth loops and local go-to-market
cut-over: The moment users switch from the old system to the new one, after which the old one can be turned off.In Build, buy or outsource
cutover: The planned moment when the old system stops being used and the new one takes over, usually with a timed checklist, a go/no-go decision and a rollback plan.In Platforms, integrations and systems of record
CVE: Common Vulnerabilities and Exposures: a public catalogue that gives each known software vulnerability a unique identifier.In Attacking networks and services
CVE and CVSS: A CVE is the public id of one known vulnerability; CVSS scores how severe it is from 0 to 10. Supply chain
CVSS: Common Vulnerability Scoring System: a 0 to 10 score for how serious a security hole is. 9.0 and above is Critical, 7.0 to 8.9 High, 4.0 to 6.9 Medium.In Looking after 1,400 devices, Networks, endpoints and the cloud, Security in the pipeline and in AI features, Attacking networks and services, Reporting that gets things fixed
cyber kill chain: A model of an intrusion as a sequence of stages from reconnaissance to actions on the objective; stopping any one stage can stop the attack.In Inside a SOC
Cycle time: The elapsed time from the start of a process to its end for one case, including all waiting.In Modelling processes

D

DACI: A way to assign roles for a decision: Driver, Approver, Contributors and Informed.In Stakeholders, negotiation and leading without authority
DAG: Directed acyclic graph: a map of which model reads which, with no loops, so there is always an order that builds every model after its inputs.In Transforming data in the warehouse, Fresh, tested, trusted, From notebook to pipeline
dark patterns: An interface designed to trick or pressure people into choices they would not make if they understood them, such as hidden costs or a cancellation that is hard to find.In Engagement loops, habits and notifications without dark patterns
dark store: A small warehouse closed to shoppers that only picks and packs app orders for fast delivery nearby.In Unit economics: does each order make money?
DAST: Dynamic application security testing: a tool that sends requests to a running copy of your app, the way an attacker would, and reports responses that show a weakness.In Security in the pipeline and in AI features
data checksums: A small check value stored in each data page, so PostgreSQL can detect a page that was corrupted on disk.In Upgrades, maintenance and on-call
data classes: Groups of data with the same handling rules, for example public, internal, confidential and restricted.In Leading AI adoption
data contract: A written agreement between the team that produces data and the teams that use it, covering its fields, meaning, delivery time, quality and how changes are announced.In Fresh, tested, trusted
data dictionary: A short table that lists every column in a dataset with its meaning, type, unit and any known problems.In Analysing with an AI assistant, Requirements that work
Data drift: A change in the distribution of a model's inputs, such as prices, store mix or promo share, compared with the training data.In Monitoring, drift and retraining
Data minimisation: Collecting and keeping only the personal data a task really needs.In Responsible ML in a business
Data model: Tables and the relationships between them that a pivot or BI tool reads. Reporting model
Data ownership: The rule that exactly one module or service may write a given piece of data. Data ownership
data plane: The part of a system that moves the actual payload, here the video bytes, as fast and cheaply as possible, following the control plane's decisions.In Netflix: streaming to every screen
data snapshot: A fixed copy, or a fixed list of files, of the exact data a run used, with an id the run records.In Experiments, versions and the model registry
data test: In dbt, a SELECT statement that returns the rows breaking a rule. Zero rows means the test passes.In Transforming data in the warehouse
data type: What kind of value a cell holds, such as a number, a date or text. Formulas treat each type differently.In Cleaning data in a spreadsheet
data-flow diagram: A drawing of a system that shows its processes, data stores, outside actors and the data that moves between them.In Threat modelling a feature
database: The program that stores data on disk and finds it again quickly, for example PostgreSQL.In What system design really is, What AI coding tools can and cannot do, A spec the AI can follow, Data and security basics
database per service: Each service or module is the only one that reads and writes its tables; others go through its API or events.In Monolith, modular monolith, microservices
DataFrame: The pandas name for a table: rows and named columns, where every column has one data type.In The same analysis in Python
datasheets for datasets: A document for a dataset that says why it was collected, what it contains, how it was collected and cleaned, what it should and should not be used for, and who maintains it.In Responsible ML in a business
DBIR: Verizon's Data Breach Investigations Report, a widely cited yearly study of thousands of real security incidents and breaches.In How breaches really happen
dead row: An old row version that no running or future transaction can see any more. It still takes space until VACUUM clears it.In Running Postgres in production
dead tuple: A row version that no running or future transaction can see. It still takes space until VACUUM removes it.In Inside a PostgreSQL server
dead-letter queue: Dead-letter queue: a separate place for jobs that failed all their attempts, kept so a person can inspect, fix and replay them.In Queues, background jobs and pub/sub, Platforms, integrations and systems of record
Dead-letter queue (DLQ): A holding place for jobs that failed all their attempts, kept for a person to inspect and replay. Queues and jobs
deadline: An absolute time by which a whole request must finish, passed from service to service so each one knows how much time is left.In Failure is normal: timeouts, retries, backoff and circuit breakers
deadlock: Two transactions each wait for a lock the other holds, so neither can ever continue. Postgres detects this and cancels one of them.In Transactions, ACID, isolation and locks, Case study: ordering without overselling stock
debrief: A short talk right after a session where the tester and a lead go through what was tested, found and still open.In Exploratory testing
debt register: A short shared list of known technical debt items, each with its cost, its effort to fix and an owner.In Quality, debt and AI tools
debugger: A tool that pauses a running program at a chosen line so you can inspect variables, the call stack and step through code one line at a time.In Errors, debugging and testing
decider: The one person named before a meeting who makes the final call if the group cannot agree.In Running a design review
decision document: A short written document, usually one page, that asks for a decision and gives the recommendation, options and risks.In Writing for decisions
decision log: A dated record of each approval, rejection, promotion or rollback: what was decided, by whom, on what evidence and why.In Responsible ML in a business, Leading with product and management
Decision memo: A short document that asks for one decision with options, a recommendation and a deadline. Decision memos
decision rule: A rule written before the test starts that says which result leads to ship, iterate or stop.In Designing and reading an A/B test
decision table: A short table listing each user of a dashboard, when they use it, the decision they make, the question that decision needs answered, and the part of the dashboard that answers it.In Designing a dashboard for decisions, Designing tests that find bugs, Requirements that work
decision tree: A model that predicts by asking a sequence of yes or no questions about the features. Each final group, a leaf, predicts the average of its training rows.In Training and comparing models
declarative: You describe the end result you want, and the tool works out which steps get there from what exists now.In Infrastructure as code
declare: To say out loud, in the agreed place, that something is an incident, which starts the incident process: a commander, a channel and updates.In Leading an incident
Decode: The second phase: the model generates the answer one token per step. Each step reads all model weights from memory but does little math, so it is limited by memory speed.In ChatGPT: serving a large language model to hundreds of millions
deemed acceptance: A contract clause that treats a delivery as accepted if the buyer has not formally rejected it within a set number of days.In Working with vendors and agencies
deep copy: A copy that duplicates every level, so the copy shares no nested objects or arrays with the original.In Objects, arrays and references
deep link: A link that opens an app on one specific screen, such as a single job, instead of its home screen.In Navigation and app structure
default gateway: The router address a device sends all traffic to when the destination is outside its own subnet.In Networks from the till to the data centre
defect: A flaw in a work product (code, a story, a test, a config file) that can make the system behave wrongly. Also called a fault or a bug.In What testing is for, Bugs, reports and acceptance
Defence in depth: Layering several independent controls so that when one fails, another still stops or catches the attack.In Thinking like an attacker
Definition of Done: A team-wide checklist every story must meet before it counts as finished, such as reviewed code, automated tests and release notes.In User stories and acceptance criteria, User stories and acceptance criteria
definition of ready: A short, team-agreed checklist a story should meet before the team commits to it in a sprint.In User stories and acceptance criteria
degenerate dimension: An identifier such as a receipt or order number kept on the fact table with no dimension table of its own.In From questions to a reporting model
degrees of freedom: A number, worked out from the sample sizes, that sets the shape of the t distribution; with thousands of rows the t distribution is almost the normal one.In Hypothesis tests without the fog
delayed replica: A replica configured with recovery_min_apply_delay to replay WAL only after a fixed delay, such as one hour.In Replication and read replicas
delegation brief: A short written note that hands over a piece of work: the outcome, why it matters, the context, the limits, who decides what, and when you will check in.In Delegation and mentoring
Delegation level: How much decision power is handed over, from "I tell you" to "you decide". Delegation
delimiters: Markers such as tags or headings that show where a piece of data starts and ends, so the model can tell data from instructions.In Prompts and structured output
delivery: The work of building, testing and releasing the product.In What a product manager actually does
delta compression: Storing an object as the differences from a similar object instead of in full.In Inside the .git folder: objects, refs and HEAD
denial of wallet: Abuse or bugs that run up a pay-per-use bill: scripted requests, agent loops or retry storms, so the damage is money rather than downtime.In Cost and latency at scale, Guardrails and red-teaming
density-independent pixels: A unit that is scaled by the screen's pixel density, so 16 of them look about the same size on a cheap phone and an expensive one.In How React Native works
deoptimisation: Deoptimisation: the engine discards optimised machine code because an assumption it made is no longer true, and falls back to slower code.In Performance and memory in JavaScript
dependency: A library your project needs in order to run or build, such as NestJS or date-fns. Listed in package.json.In How code becomes a running program
dependency array: The array after the effect function. React runs the effect again only when one of these values changed since the last render, compared with Object.is.In State, effects and where data should live
dependency injection: A pattern where an object receives the services it needs (through its constructor) instead of creating them itself. NestJS does this automatically.In Prototypes, classes and modules, From script to service: a small API in Node, Your first NestJS service
deploy: To put a version of your app on a server so that other people can use it at a public web address.In Ship it and know your limits, Releasing safely
Deployment: A Kubernetes object that describes a set of identical pods (image, replicas, update strategy) and manages ReplicaSets to roll out changes.In Kubernetes for application teams
Deployment (Kubernetes): The object that keeps a set number of pod copies running and rolls out new versions. Kubernetes
Deployment frequency: How often a team deploys changes to production.In Team design and delivery health
depreciated: Spreading the cost of an asset over the years it is used, so each year's accounts carry a share of it.In Business cases a managing director will sign
depreciation: Spreading the cost of a physical asset, such as a till, over the years it is used.In Budgets, capex and opex
depreciation and amortisation: Depreciation spreads the cost of a physical asset over its useful life; amortisation does the same for an intangible asset such as software.In The three statements
derived value: A value calculated from other data during render, such as the low-stock count from the stock rows. It is never stored, so it can never be out of date.In State, effects and where data should live
design doc: A short written proposal for a system or feature: the problem, requirements, numbers, design, alternatives and risks. Some teams call it an RFC.In Running a design review, Technical vision and design docs
design effect: How many times bigger the variance is when you randomise clusters instead of individuals: 1 + (m - 1) x ICC, where m is the cluster size.In Designing and reading an A/B test
design review: A meeting where a team checks a proposed design before building it, to find risks early and agree on the key decisions.In Running a design review
design system: A shared library of reusable components, design tokens such as colours and spacing, and guidelines that keep a company's products consistent.In Design for PMs
Destructuring: Unpacking properties of an object or items of an array into separate variables in one statement, such as const { siteId, qty } = sale.In Objects, arrays and references
detached: The state where HEAD holds a commit ID directly instead of naming a branch. New commits are not on any branch until you create one.In Inside the .git folder: objects, refs and HEAD
Detached HEAD: When HEAD points straight at a commit instead of a branch; new commits there are easy to lose unless you create a branch. Inside the .git folder: objects, refs and HEAD
detection as code: Keeping detection rules as text files in version control, with metadata, test cases and automated checks, like any other software.In Writing detections
Detection rule: A saved query over logs that raises an alert when it matches suspicious activity. Detections
Dev Containers: A VS Code feature that runs your editor's tools inside a Docker container defined by .devcontainer/devcontainer.json, so everyone gets the same Node version, tools and extensions.In VS Code for teams: formatting, linting and shared settings
development build: Your own app compiled with the expo-dev-client library: it contains your chosen native code and loads JavaScript from Metro during development.In How React Native works, Navigation and app structure
DevEx: A framework that says developer experience depends on three things: feedback loops, cognitive load and flow state.In Team design and delivery health
device inventory: A list of every device the company owns, with its serial number, location, owner and current state. Also called an asset register.In Looking after 1,400 devices
device matrix: The short list of devices, operating system versions and screen sizes a team tests on, chosen to represent its real users.In Accessibility, devices and the real world
device sequence: A counter on each device that goes up by one for every sale and is never reused, so a missing number at HQ means a missing sale.In Case study: syncing POS sales from hundreds of stores
device token: An id that the push service gives one app install on one device. Your server stores it and sends pushes to it. It can change or stop working.In Real-time updates: polling, WebSockets, SSE and push
DHCP: Dynamic Host Configuration Protocol: a device asks the network for an address and settings, and a DHCP server answers.In Networks from the till to the data centre
DHCP options: Extra settings a DHCP server hands out with the address, such as the default gateway, the DNS servers and the lease time.In Networks from the till to the data centre
DHCP reservation: A DHCP setting that always gives one device, recognised by its MAC address, the same IP address.In Networks from the till to the data centre
diary study: A study where participants record short notes, photos or voice messages each time something happens, over days or weeks.In Watching the work
diff: A text report of the differences between two versions of files: which lines were removed (-) and which were added (+).In Everyday Git: status, diff, log and good commits, Building in small steps
difference in differences: An estimate of an effect: the change in the pilot group minus the change in a comparable control group over the same period.In Experiments: A/B tests, holdouts and store pilots
digest: The content hash of an image manifest, written sha256: followed by 64 hex characters. The same digest always means the same bytes.In Containers from the inside, Secure supply chain and delivery metrics
digital signature: A value made from a message with a private key. Anyone with the matching public key can check that the message came from the key holder and was not changed.In Cryptography without the maths
dimensional modelling: A way of designing reporting tables, made popular by Ralph Kimball, that separates measurements (facts) from the descriptions used to slice them (dimensions).In From questions to a reporting model
dimensions: A table of descriptive attributes, such as a store's city or a product's category, used to filter and group facts.In From questions to a reporting model
directory: The central list of people and their accounts, groups and details such as job title, store and manager.In Accounts, SSO and MFA
directory-based sharding: Sharding where a lookup table records which shard each key (or group of keys) lives on, instead of computing it with a hash.In Sharding and consistent hashing
dirty page: A page in shared buffers that has been changed in memory but not yet written back to its data file.In Inside a PostgreSQL server
discount rate: The yearly return a company requires from its money, set by finance; future cash is discounted at this rate when calculating NPV. Also called the hurdle rate.In Business cases a managing director will sign
Discount rate (hurdle rate): The return the company requires, used to bring future cash to today's value. Business cases
Discounted payback: Payback calculated on discounted cash flows: the first period in which the running total of discounted cash is at zero or above.In Business cases
Discounting: Converting future money into today's value by dividing each future amount by (1 + rate) raised to the number of periods.In Unit economics
discovery: The work of deciding what to build: learning about users and problems and testing ideas before committing engineers to them.In What a product manager actually does, User interviews that don't lie
discriminated union: A union of object types that all share one field with a different literal value, such as method: 'cash' or 'wallet', so a check on that field tells TypeScript which object it is.In TypeScript for a JavaScript team
Dispatch: The system that decides which driver or rider gets which trip or order, and when.In Teardown: ride-hailing and food delivery
dispatcher: The code between the model and your tools. It looks up the tool, validates the arguments, adds facts from the session, checks permissions and runs the tool with a timeout.In Tools and the Model Context Protocol
Distillation: Training a small model on the outputs of a larger model for one task, so the small model copies the large one's behaviour at lower cost.In Prompting, retrieval or fine-tuning?
distributed monolith: A system split into separately running services that still have to change and release together, usually because they share data or call each other in tight chains.In Choosing an architecture style
distributed system: A system whose parts run on several machines that talk over a network, so they can fail separately and can be cut off from each other.In CAP, PACELC and consistency
Distribution: How often each value occurs; its shape matters as much as its middle. Describing data
distribution centre (DC): A large warehouse that receives goods from suppliers and sends them on to stores. Mangrove has four: North, East, South and West.In Asking a question data can answer
Distroless: Container base images that contain only an application runtime and its libraries: no shell, no package manager, no extra tools.In Containers from the inside
DNS: Domain Name System: the internet's phone book. It turns a name like api.example.com into an IP address.In What happens when you open an app or website, Ship it and know your limits
Docker image: A packaged file system with the system libraries, Python, packages and code a program needs, which runs the same way on any machine.In From notebook to pipeline
Docstore: Uber's distributed database built on MySQL. Data is split into partitions, each copied to several nodes kept in step by the Raft protocol.In Uber: matching riders and drivers in real time
DOM: Document Object Model: the browser's live tree of objects built from the HTML. JavaScript reads and changes the page through it.In JavaScript in the browser, Components, props and how React renders
DOMA: Domain-Oriented Microservice Architecture: Uber's way of grouping related microservices into domains with one gateway each, arranged in layers.In Uber: matching riders and drivers in real time
domain: A name such as mangrove-reorder.com that you rent yearly from a registrar and point at your app.In Ship it and know your limits
DORA: DevOps Research and Assessment: a long-running research programme, now part of Google Cloud, that studies what makes software delivery teams effective, and publishes the metrics and yearly reports.In Your team's engineering playbook, Networks from the till to the data centre, Secure supply chain and delivery metrics
DORA metrics: Deployment frequency, change lead time, change fail rate and recovery time: measures of how fast and how safely a team delivers. Your team's engineering playbook
double extortion: A ransomware tactic where attackers copy data out before encrypting it, then threaten to publish it if the victim doesn't pay.In How breaches really happen
draft pull request: A pull request marked as work in progress. It cannot be merged, and code owners are not asked to review it until it is marked ready.In Pull requests and code review
drift: A change over time in what users ask or in how the system behaves, so a test set built earlier stops matching real use.In Evals that never stop, Infrastructure as code
drill-down: Moving one chart to a finer level of the same dimension, for example from city to store, in place.In Designing a dashboard for decisions
drill-through: Opening a separate detail page that is already filtered to the item you clicked, such as one city or one store.In Designing a dashboard for decisions
driver tree: A diagram that splits a metric into the smaller numbers that multiply or add up to it, so you can see which piece moved.In Asking a question data can answer
dry run: Running an action in preview mode: the system works out what would happen, such as the source store and the value, without changing anything.In Designing tools and permissions, Toil and automation
DTO: Data transfer object: a class that describes the shape of data coming into or going out of an API.In APIs done right, From script to service: a small API in Node, Your first NestJS service
DTO (data transfer object): A class describing the shape of a request or response, used for validation. NestJS service
dtype: The data type of a column in pandas, for example int64 for whole numbers, float64 for decimals, datetime64 for dates and object for text.In The same analysis in Python
dual-write problem: Writing to two systems, such as a database and a message queue, in two separate steps, so that a failure between them leaves one updated and the other not.In Events, the outbox pattern and sagas, Background jobs and idempotency
dummy variable: A 0/1 column that marks whether a row belongs to a group, such as rain = 1 for orders placed while it was raining.In Regression and correlation
duplicate: Two or more rows that describe the same observation, for example the same store, date and category, usually because data was loaded twice.In Cleaning data in a spreadsheet, Application portfolio
dwell time: The length of time an attacker is present in an environment before they are detected. Shorter is better; hunting is one way to cut it.In Threat hunting
Dynamic array: An Excel formula result that spills into several cells, such as UNIQUE or FILTER. Data cleaning
dynamic route: A route file with a name in square brackets, such as [id].tsx, that matches any value in that part of the URL and passes it to the screen as a param.In Navigation and app structure
Dynamo paper: A 2007 Amazon paper describing an internal key-value store built so writes always succeed, with replicas that may disagree and are merged later.In Amazon: the store that never closes
DynamoDB: AWS's managed NoSQL database. Items are found by a partition key and an optional sort key; AWS handles servers, replication, backups and scaling.In Amazon: the store that never closes

E

e-money: A digital balance issued by a wallet operator that is backed one-for-one by real money the operator holds in a bank.In Teardown: sending money with a mobile wallet
earliest start: The first quarter a package can begin: the latest finish among all the packages it needs, or quarter 0 if it needs none.In Target state and roadmap
Early stopping: Stopping training when the score on held-out data has not improved for a set number of rounds.In Training and comparing models
EAS: Expo Application Services: cloud services for building (EAS Build), submitting to the stores (EAS Submit) and sending over-the-air JavaScript updates (EAS Update).In How React Native works
EBITDA: Earnings before interest, tax, depreciation and amortisation: gross profit minus the day-to-day running costs.In The three statements, Budgets, capex and opex
edge case: An unusual but real situation at the boundary of normal use, such as an empty list, a missing item or a lost connection.In Reading and fixing what the AI wrote
edge locations: One CDN location, a data centre with cache servers close to a group of users. Also called a point of presence (PoP).In Files, object storage and CDNs
editor: A program for changing text files, with helpers such as colours for code and search.In How code becomes a running program
EDR: Endpoint detection and response: software on each computer and server that records what programs do, flags attacker behaviour and lets the security team isolate a machine remotely.In How breaches really happen, Networks, endpoints and the cloud, Inside a SOC
effect: Code passed to useEffect that runs after React has updated the screen, to connect the component to something outside React, such as a timer, a subscription or a non-React widget.In State, effects and where data should live
effect size: How big the difference is, in units people care about: points of conversion, dollars, minutes.In Hypothesis tests without the fog
effective_cache_size: A hint telling the planner how much memory is available for caching data in total, including the operating system cache. It allocates nothing.In Running Postgres in production
egress: Data sent out of a cloud provider to the internet. Many providers charge per GB for it.In Files, object storage and CDNs, Cloud and SaaS costs with FinOps
element: A plain JavaScript object that describes what to show: a type (a tag name or a component function), props and a key. Cheap to create and thrown away after each render.In Components, props and how React renders
Elicitation: Drawing out information from stakeholders and other sources: what they need, how work happens today, and why.In Stakeholders and elicitation
ELT: Extract, load, transform: raw data is copied into the warehouse as it is, and reshaped there afterwards with SQL.In Transforming data in the warehouse
embedding: A list of numbers, such as 1,024 of them, that a model produces from text. Texts with similar meaning get vectors that point in similar directions.In Retrieval-augmented generation, Designing AI features: LLMs, RAG and vector search
employee assistance programme: A confidential counselling and advice service some employers pay for, which staff can use for personal problems.In Performance and growth
empty state: What a screen shows when there is nothing in it yet, such as an empty cart or no order history.In Onboarding and the first-time experience
enabling teams: A small team of specialists that helps other teams learn a skill for a limited time, then moves on.In Organisation and operating model
encoding ladder: The list of versions of one video, each a pair of resolution and bitrate, from small and low quality to large and high quality. Each version is a rung.In Netflix: streaming to every screen
end-to-end (E2E) test: A test that drives the whole system through its real user interface, from the browser down to the API and database, the way a user would.In End-to-end tests with Playwright
End-to-end encryption: Encryption where only the sender's and recipients' devices hold the keys, so servers in between pass along ciphertext they cannot read.In WhatsApp: billions of messages with a tiny team
end-to-end test: A test that drives the real app the way a user does, by tapping and typing on screen, against a running back end.In Testing what matters, A test strategy, not a pile of scripts
endpoint: One address on the API that the app can call, such as GET /v1/orders, with a fixed set of inputs and answers.In Data model and API, Networks, endpoints and the cloud
engagement: How often and how deeply people use a product, measured in a way that matches the value they get from it.In Engagement loops, habits and notifications without dark patterns
engagement plan: A short table saying, for each stakeholder, what you need from them, how and how often you will work with them, and who owns the contact.In Stakeholders and elicitation
engineering manager: The manager responsible for the people on a team: hiring, performance reviews, pay, careers and team health.In What a tech lead is (and is not)
Enrichment: Adding facts about the users, machines and addresses in an alert from other systems, so the alert can be judged.In Triage and investigation
enrolment: The one-time step where a device registers with the company's management server and accepts its control.In Looking after 1,400 devices
enterprise architect: An architect who keeps the whole company's systems coherent: the portfolio of systems, shared standards and the long-term technology roadmap.In What architects actually do
Enterprise architecture: The practice of describing how a whole organisation fits together, across business and technology, and planning the changes that move it toward its strategy.In Why enterprise architecture exists
Entry criteria: Conditions that must be true before a test activity starts, so the time is not wasted.In Bugs, reports and acceptance, Value streams and processes, From change control to acceptance
entry point: The first file a program starts from, such as main.ts in NestJS or index.js in a script.In How code becomes a running program
enum: A custom column type with a fixed list of allowed values, such as CASH, CARD, WALLET.In Modeling data in tables, Designing an API people can use, Prompts and structured output
Enumeration: Carefully reading what a running service tells you about itself: its type, version, settings and what it will let an unauthenticated user do.In Attacking networks and services
Environment (GitHub): A named deployment target such as staging or production, with its own secrets and protection rules like required approvals. Deploying and securing with GitHub Actions
environment variable: A named text value that every program can read from its process, such as HOME, PATH or STORE_ID. Children inherit their parent's variables.In The terminal without fear, Continuous delivery, Data and security basics, Ship it and know your limits
environments: A separate copy of the system with its own config and data, such as dev, staging or production.In Shipping safely: containers, CI/CD and releases, Branching strategies, releases and versions, Deploying and securing with GitHub Actions, Continuous delivery
epics: A large piece of work made of several user stories, often one feature or one outcome.In User stories and acceptance criteria
equity: The owners' share of the company: money they put in plus profits kept in the business over the years.In The three statements
equivalence partition: A group of inputs that the software should treat in the same way, so one value from the group stands for all of them.In Designing tests that find bugs
Erlang: A programming language made by Ericsson for telephone switches. Code runs as many tiny isolated processes that talk by sending messages.In WhatsApp: billions of messages with a tiny team
ERP: Enterprise resource planning: a commercial software package that runs core business records such as finance, invoicing and accounts receivable.In Why enterprise architecture exists
error: A human mistake, such as misreading a rule or typing the wrong comparison.In What testing is for, Bugs, reports and acceptance
Error analysis: Breaking a model's validation errors down by groups of rows, finding where it fails most, and deciding what to do about each failure.In Explaining and choosing models
error boundary: A component that catches errors thrown while rendering the components inside it and shows a fallback instead of a broken or blank screen.In Loading server data
error budget: The amount of failure an SLO allows: 100% minus the target. A 99.9% SLO allows 0.1% of requests (or minutes) to be bad.In Measuring speed and reliability: p99, SLOs and nines, Observability: logs, metrics and traces, Running products in production
error budget policy: A short written agreement that says what a team does at each level of remaining error budget, who decides, and which exceptions are allowed.In How reliable is reliable enough?
error tracking: A service that catches crashes in the browser and on the server, groups the same error together and alerts you.In Ship it and know your limits
ES modules: The standard JavaScript module system using import and export; Node uses it when "type": "module" is set. Prototypes, classes and modules
Escalation: Passing a ticket to a team with more skill (functional) or to a manager with more authority (hierarchical).In Incidents, problems and changes, Leading with product and management
escaped defect: A bug found by customers or in production that the tests should have caught before release.In A test strategy, not a pile of scripts
ETA: Estimated time of arrival: the time the app predicts the ride or order will arrive.In Teardown: ride-hailing and food delivery
ETag: A version tag for a resource, used for conditional GETs and safe concurrent updates. Caching
ETL: Extract, transform, load: data is reshaped on a separate server before it reaches the warehouse, so only the finished result is stored.In Transforming data in the warehouse
ETL and ELT: Extract, transform, load: cleaning data before loading it, or loading raw data first and transforming it in the warehouse. Warehouse models
eval: A repeatable test of an AI feature's output quality: a set of cases, a way to grade each one, and a result you can compare across versions.In Evals: measuring AI quality, Designing AI features: LLMs, RAG and vector search, Leading AI adoption
eval case: One test for an AI feature: an input with its context, the facts or rules a good output must meet, a category, and the list of checks to apply.In Evals: knowing if it got better
eval set: A fixed set of test inputs with expected behaviour, run against an AI feature after every prompt, model or tool change to catch regressions.In Security in the pipeline and in AI features
Evaluator–optimizer: A pattern where one model call produces a draft and another grades it against criteria, looping with feedback until it passes or a try limit is reached.In What makes an agent
EVCache: Ephemeral Volatile Cache: Netflix's caching layer built on memcached, with copies in several AWS regions.In Netflix: streaming to every screen
event: A message that states a fact that has already happened, named in the past tense, such as order.placed. The sender does not know or care who listens.In Events, the outbox pattern and sagas, GitHub Actions: continuous integration, JavaScript in the browser, Instrumentation: tracking plans and data you can trust, Observability with OpenTelemetry, Risk, compliance and the first hour of an incident
event broker: A system that receives events from producers, stores them and delivers them to every subscribed consumer, such as Kafka, RabbitMQ or a cloud pub/sub service.In Choosing an architecture style
event delegation: Putting one listener on a parent element and using event.target to work out which child was clicked, instead of a listener on every child.In JavaScript in the browser
event log: A table where each row records one step of one case and when it happened, such as order 42 was keyed at 17:30.In Modelling processes
event loop: The loop that keeps the single JavaScript thread busy: when the call stack is empty it drains the microtask queue, then (in a browser) may render, then takes the next task.In The event loop
Event time: The moment the event actually happened on the machine that produced it.In Logs that matter
event-driven architecture: A design where parts of a system communicate by publishing facts about what happened, and other parts react to them.In Events, the outbox pattern and sagas
Eventual consistency: Copies of data may disagree for a short time but will agree once updates stop arriving. CAP and consistency
Eviction: Removing items from a full cache to make room for new ones, following a rule such as least recently used.In Caching
exception: A written, approved permission to miss one rule for a limited time, with a reason, an end date and a compensating control.In Risk, compliance and the first hour of an incident
Exception filters: A class that catches exceptions thrown anywhere in the request and turns them into the HTTP response.In Your first NestJS service
excessive agency: OWASP's name for an AI system that has more tools, more permissions or more autonomy than its tasks need, so a bad model output can do real damage.In Designing tools and permissions
exclusive gateway: A BPMN diamond, often drawn with an X, where exactly one of the outgoing paths is taken, chosen by a condition.In Modelling processes
exemplar: A sample request id attached to one bucket of a metric, usually a trace id, so you can jump from a latency spike to a real trace in that bucket.In Observability with OpenTelemetry
exit assistance: A contract duty for the outgoing vendor to help move the service to you or a new supplier: documents, data, knowledge transfer, time from its people.In Build, buy or outsource
exit code: A number a program returns when it ends. 0 means success; anything else means some kind of failure.In The terminal without fear, Reading code like a detective, Shell scripting for daily jobs
exit criteria: Conditions that must be true before you can say a test activity is finished, such as no open critical bugs.In Bugs, reports and acceptance, Value streams and processes, From change control to acceptance
exit run rate: The yearly cost of everything you are paying for at the end of the period, as if it continued for a full year. It shows next year's starting cost.In Money: budgets, headcount and vendors
expand and contract: A way to make a breaking change in safe steps: add the new form alongside the old one, move every consumer to the new form, and remove the old form last. Also called parallel change.In Contract tests
experience-based techniques: Test techniques that use the tester's knowledge and experience, such as error guessing, exploratory testing and checklist-based testing.In Exploratory testing
Experiment tracking: Recording every training attempt, with its settings, results, code version, data and output files, in one searchable place.In Experiments, versions and the model registry
Exploratory testing: Testing where you learn about the product, design tests and run them at the same time, letting each result guide the next test.In Exploratory testing
Expo: An open-source set of tools and services for building React Native apps, including a CLI, ready-made libraries and build services.In Building the app, How React Native works
Expo Go: A free app from Expo for Android and iPhone that loads your project's bundle from your laptop, so you can test without building an app file.In Building the app, How React Native works
Expo Router: Expo's navigation library. Each file in the app folder becomes a screen, and the file's path becomes the screen's address.In Building the app
exponential backoff: Waiting longer after each failed attempt, usually doubling the wait: 2 s, 4 s, 8 s, 16 s.In Queues, background jobs and pub/sub, Failure is normal: timeouts, retries, backoff and circuit breakers, From notebook to pipeline, Offline first
Exponential backoff with jitter: Waiting longer after each failed retry, with a random part, so many clients do not retry at the same moment. Timeouts and retries
Exponential smoothing (ETS): A family of forecasting methods that weight recent observations more than older ones, with separate terms for level, trend and seasonality.In Forecasting demand
export: A file copied out of a system, such as a CSV file from the sales database, so you can work on the data elsewhere.In Cleaning data in a spreadsheet
Expression index: An index on the result of an expression, such as lower(invoice_no), instead of on a plain column.In Indexes from the inside
extended statistics: Statistics about several columns together, created with CREATE STATISTICS, so the planner knows when columns depend on each other.In Finding and fixing slow queries
extensions: An add-on package that gives VS Code a new skill, such as ESLint checks or Python support. It runs with the same permissions as VS Code.In How code becomes a running program
External Secrets Operator: An open-source Kubernetes add-on that reads secrets from an outside store such as Vault or a cloud secrets manager and keeps matching Kubernetes Secrets up to date.In Kubernetes for application teams
extrapolates: Predict outside the range of values seen in training. A linear model follows its straight line there; a tree model cannot go past the values in its leaves.In Training and comparing models
extrapolation: Using a fitted model to predict for x values outside the range of the data it was fitted on.In Regression and correlation

F

Fabric: React Native's current renderer, written in C++ and shared by Android and iOS, which turns React trees into native views.In How React Native works
facilitator: The person who runs the process of a meeting: keeps time, asks questions, makes sure everyone is heard, and stays neutral on the content.In Stakeholders and elicitation
Fact table: A table of events with numbers, at one declared grain. Warehouse models
facts: A numeric measurement of a business event, such as units sold or an amount in cents, stored in a fact table and added up in reports.In From questions to a reporting model
Fail closed: When the limiter cannot check, reject the request. Safe for sensitive endpoints, but an outage of the limiter becomes an outage of the endpoint.In Rate limiting and protecting your system
Fail open: When the limiter cannot check, allow the request. The service stays up but unprotected for a while.In Rate limiting and protecting your system
Failover: Making a replica the new primary after the old primary fails, and pointing the applications at it.In Replication and read replicas
failure: An event where the system does something it should not do, or does not do something it should, while it runs.In What testing is for, Bugs, reports and acceptance
fairness check: Comparing a model's errors and outcomes across groups of people or places, to see whether some groups get a worse result.In Explaining and choosing models
fake door test: A test that shows the entry point to a feature that does not exist yet, counts who tries to use it, and then shows an honest message.In Testing ideas cheaply
fallacies of distributed computing: Eight false beliefs about networks, collected at Sun Microsystems, that lead to fragile distributed systems.In Failure is normal: timeouts, retries, backoff and circuit breakers
fallback: What your code does instead when a dependency fails or the breaker is open, such as using a cached value, queuing the work, or hiding a feature.In Failure is normal: timeouts, retries, backoff and circuit breakers, Prompts and structured output, Device features
fallback chain: An ordered list of answers to use when the best one is missing, each one safer and less accurate than the one before.In Serving: batch and online
false discovery rate: The expected share of false alarms among all the results you call discoveries.In Traps that fool smart people
false positives: A finding a tool reports that turns out not to be true or not to apply: a false alarm.In Attacking networks and services
falsy: A value that counts as false in a condition. There are exactly eight: false, 0, -0, 0n, the empty string, null, undefined and NaN.In JavaScript in one sitting
family-wise error rate: The chance of at least one false alarm across a whole set of tests.In Traps that fool smart people
fan-out: Sending one request out to many backends at once and waiting for all of them to answer.In Measuring speed and reliability: p99, SLOs and nines, Indexes from the inside, Self-service without chaos
Fan-out on read: Store posts only once; when a user opens the feed, look up everyone they follow and merge their newest posts at read time.In Instagram: from 3 engineers on Postgres to a billion users
Fan-out on write: When a user posts, copy the post id into a feed list for every follower at write time, so reading a feed is one quick lookup.In Instagram: from 3 engineers on Postgres to a billion users
Fast-forward merge: A merge where the branch pointer simply moves ahead because no new commits exist on the target; no merge commit is created. Branches and merging
FCM: Firebase Cloud Messaging: Google's service for sending push notifications to Android apps (and to iOS apps through Apple's service).In Real-time updates: polling, WebSockets, SSE and push, Device features
feature flag: A switch in config or the database that turns a feature on or off at run time, for everyone or for some users, without a new deploy.In Shipping safely: containers, CI/CD and releases, Branching strategies, releases and versions, Launching: pilots, waves and the review after, Writing for decisions, Shipping an agent safely, Releasing safely
feature parity: Rebuilding everything the old system does, exactly as it does it today, on new technology.In Evolving a legacy system
feature scaling: Rescaling numeric features so they have similar ranges, for example to mean 0 and standard deviation 1.In Training and comparing models
feature store: A system that keeps feature definitions in one registry, builds point-in-time training sets from an offline store and copies the latest values to an online store.In Data and feature pipelines
feature table: A table of model inputs with one row per entity and point in time, for example one row per store, product and night.In Data and feature pipelines
feature version: A new definition of a feature, stored next to the old one under a new name or version, so models trained on the old one keep working.In Data and feature pipelines
features: An input column the model uses to make a prediction, such as last week's sales or days to the next festival.In Framing a problem as a model, Features and leakage
feedback loop: A loop where a model's output changes the data it is later trained on, so an early error grows over time.In Explaining and choosing models
fence: A rule that keeps each customer on the price meant for them, such as a volume limit, a feature, a commitment period or an eligibility check.In Pricing and packaging
Fencing: Making sure a failed or isolated node cannot take writes any more, for example by powering it off, cutting its network, or having it stop itself when it loses its lease.In Replication and failover
Fetch vs pull: Fetch downloads new commits and updates origin/main; pull is fetch plus merge or rebase into your branch. GitHub: remotes, clone, push, pull and authentication
Few-shot examples: Putting a few worked input and output pairs in the prompt so the model copies their pattern.In Prompts and structured output
Few-shot prompting: Putting a few worked examples in the prompt so the model copies their pattern. LLM APIs
fidelity: How close a prototype is to the real product in look, behaviour and data, from a paper sketch to a working app.In Prototyping with AI
fields: One piece of information in a record, such as a product's name or price. In a table it is a column.In A spec the AI can follow
filesystem: One disk or disk partition with its own space, attached somewhere in the folder tree. The root filesystem is attached at /.In Linux from the command line
finance business partner: A finance person, often from FP&A (financial planning and analysis), assigned to help a business team plan and track its money.In Budgets, capex and opex
Fine-tuning: Training an existing model further on your own examples so it picks up a habit, format or narrow skill.In How AI features work, just enough for PMs, Prompting, retrieval or fine-tuning?
FinOps: A practice, described by the FinOps Foundation, in which engineering, finance and product share responsibility for cloud spending and its business value.In Cloud architecture and cost, Cloud and SaaS costs with FinOps, Money: budgets, headcount and vendors
firewall: A set of rules that decides which network traffic may enter or leave a machine or network, by address, port and protocol.In Networking you actually need, Networks from the till to the data centre, Networks, endpoints and the cloud
first-contact resolution: The percentage of tickets solved during the first contact, without passing them to another team or calling the user back.In Incidents, problems and changes
first-time-right: The share of cases that go through a step or a whole process correctly the first time, with no correction or rework.In Modelling processes
fitness function: Any mechanism that gives an objective, repeatable check of an architectural characteristic, such as a test that fails when a forbidden dependency appears.In Decisions, diagrams and reviews
five whys: A way of finding causes by asking why a problem happened, then why that happened, repeating until you reach a cause you can act on.In Problems before solutions
five-second test: A quick test where a person sees a screen for five seconds, it is hidden, and they say what they remember and understood.In Designing a dashboard for decisions
fix forward: Recovering from a bad release by shipping a new, fixed version with a higher version number, instead of going back to the old one.In Looking after 1,400 devices
Fixed costs: A cost that stays the same over the period and range of volume you are deciding about, such as salaries, rent and licences.In Unit economics
fixture: In Playwright Test, a named piece of setup a test asks for in its arguments, such as page or a custom cart. Playwright builds it before the test and tears it down after.In End-to-end tests with Playwright
flag debt: The growing cost of old feature flags left in the code: more paths to test, more confusion, more chances that a forgotten switch does something unexpected.In Releasing safely
flags: An option that changes how a command behaves, written with one dash and a letter (-n) or two dashes and a word (--oneline).In The terminal without fear
flaky test: A test that sometimes passes and sometimes fails on the same code, usually because of timing, order or network.In GitHub Actions: continuous integration, Testing and documenting the API, Testing what matters, Continuous integration that people trust, A test strategy, not a pile of scripts, Stable suites in CI
flat config: ESLint's configuration format since version 9: one JavaScript file that exports an array of config objects. The old .eslintrc files are no longer supported in ESLint 10.In VS Code for teams: formatting, linting and shared settings
flipped cases: Cases whose result changed between two versions: fixed (failed before, passes now) or broke (passed before, fails now).In Evals: knowing if it got better
float: The e-money balance an agent holds so it can give customers balance in exchange for their cash.In Teardown: sending money with a mobile wallet
floating point: The way computers store numbers with a decimal point. Most decimals, such as 0.1, can only be stored approximately, so tiny errors appear in sums.In Reading and fixing what the AI wrote
floating-point: A number type that stores a binary approximation, like double precision or real. Fast for science, wrong for money, because most decimal fractions cannot be stored exactly.In Modeling data in tables, Working with data: arrays, Maps, JSON, dates and money
Flow efficiency: Total process time divided by total lead time, as a percent. Martin and Osterling call it the activity ratio.In Value streams and processes
focus: The one element that receives key presses right now. Browsers show it with an outline called the focus indicator.In Accessibility, devices and the real world, Cloud and SaaS costs with FinOps
focus factor: The share of available time that turns into finished planned work after meetings, code reviews, support questions and interruptions.In Slicing and planning work
follow-the-sun: An on-call model where two or more teams in time zones far apart hand the pager to each other, so each team covers its own daytime and nobody is paged at night.In Alerting that does not burn people out
--force-with-lease: A push that replaces the remote branch only if it still points where your last fetch saw it; otherwise it is refused as stale info.In Rewriting history safely: rebase, squash, cherry-pick, stash
foreign key: A column that holds the primary key of a row in another table. The database refuses values that do not exist in that table.In Modeling data in tables, Data model and API, Data and security basics
fork: A copy of someone else's repository under your own GitHub account, used to propose changes when you cannot push to the original.In GitHub: remotes, clone, push, pull and authentication
formatter: A tool that rewrites code layout (spaces, line breaks, quotes, commas) into one consistent style, without changing what the code does.In VS Code for teams: formatting, linting and shared settings
FormData: A built-in browser and server object that holds a form's fields as name and value pairs. You read a field with formData.get('name').In Forms that people can trust
FP&A: Financial planning and analysis: the finance team that builds budgets and forecasts and checks business cases.In Build, buy or outsource
frame budget: The time you have to produce one frame before the screen refreshes: about 16.7 ms at 60 Hz, 11.1 ms at 90 Hz.In Fast on a cheap phone
framework: An organised list of security outcomes or controls that many organisations use, so nobody has to invent their own and miss whole areas.In Risk, compliance and the first hour of an incident
free tier: A way to use a product without paying, with limits on how much you can use it each day or month.In What AI coding tools can and cannot do, Ship it and know your limits
freezes: VACUUM marking very old rows as visible to everyone forever, so their transaction ID no longer matters and the ID can be reused safely.In Running Postgres in production, Inside a PostgreSQL server
freshness: How recent the newest data from a source is. For a POS: how long ago HQ last received a new sale from it.In Case study: syncing POS sales from hundreds of stores, Fresh, tested, trusted, How reliable is reliable enough?
freshness SLA: Service level agreement: a written promise about a service, here what data will be ready, by when, how often the promise is kept and what happens when it is missed.In Fresh, tested, trusted
Frontline: Staff whose work happens away from a desk: in stores, warehouses, vehicles, clinics or on the street, usually on a handheld device or a shared terminal.In Designing for frontline work
Fulcio: Sigstore's certificate authority. It checks an OIDC token and issues a code-signing certificate that names that identity and is valid for about 10 minutes.In Secure supply chain and delivery metrics
fully loaded: The full cost of employing someone: salary, benefits, payroll taxes and equipment. Mangrove uses about $5,000 a month for an engineer.In Build, buy or outsource
fully loaded cost: The full cost of employing someone: salary plus benefits, payroll taxes, equipment and other costs that come with the role.In Budgets, capex and opex, Money: budgets, headcount and vendors
function: A named piece of code that does one job. You give it some inputs and it gives back a result.In Reading and fixing what the AI wrote, Shell scripting for daily jobs
function declaration: A statement that starts with the word function and a name, such as function lineTotal(qty, price) { … }. It is ready to call from the top of its scope.In Functions, scope and closures
function expression: A function created as a value inside an expression, usually assigned to a variable: const f = function (…) { … }.In Functions, scope and closures
function-level access control: A check that the caller's role is allowed to use a route or action at all, such as an admin-only endpoint.In Broken access control
functional requirement: A statement of what the system does for a user, such as: a cashier can look up the stock of a product.In What system design really is, Requirements that work
funding gate: A decision point where the funders release the next tranche of money only if the previous step delivered what it promised.In Target state and roadmap
funnel: A count of how many users reached each step of a flow, in order, so you can see where they drop off.In Launch and learn, Funnels, activation and adoption

G

game day: A planned session where a team injects failures into a system on purpose and practises detecting, handling and learning from them.In Capacity, resilience and game days
gap: Days left out between the end of training and the start of validation, so no training target overlaps the validation period.In Features and leakage, Application portfolio
gap analysis: A structured comparison of baseline and target that lists what must be added, changed or removed, and checks nothing is lost by accident.In Target state and roadmap
Garbage collection: The engine freeing memory for objects that can no longer be reached from any root. Performance and memory in JavaScript
garbage collector: Garbage collector: the part of the engine that finds memory your program can no longer use and gives it back.In Performance and memory in JavaScript
gate: A check in the pipeline that must pass before the next step runs. A failed gate stops the release.In Shipping safely: containers, CI/CD and releases
Gateway: The router a device sends traffic to when the destination is outside its own network. Store networks
Gateway API: A newer set of Kubernetes APIs for traffic into the cluster (Gateway, HTTPRoute and others). The platform team owns the Gateway; app teams attach routes to it.In Kubernetes for application teams
gauge: A metric that can go up and down, such as current queue length or memory in use.In Observability: logs, metrics and traces
generic tests: A reusable test with parameters, written once and applied to any column by naming it in a YAML file.In Transforming data in the warehouse
Generics: A type with a parameter, written with angle brackets, such as Array of T, so one function or type works for many types while keeping each one exact.In TypeScript for a JavaScript team
geospatial index: An index that finds things by where they are on a map, such as all drivers within a kilometre of a point.In Uber: matching riders and drivers in real time
Gherkin: The Given, When, Then format for writing examples that can also become automated tests. Stories and criteria
GIN: Generalized Inverted Index: maps each element inside a value (a JSON key, an array item, a word) to the rows that contain it.In Indexes from the inside
Git: A version control tool that records snapshots of a project's files on your own computer, so you can see and undo any change.In How code becomes a running program, What Git really is: snapshots and the three areas, Building in small steps
Git flow: A branching model from 2010 with two long-lived branches, main and develop, plus feature, release and hotfix branches.In Branching strategies, releases and versions
Git hook: A script Git runs automatically at a certain moment, such as just before a commit is created. If it exits with a non-zero code, Git stops.In VS Code for teams: formatting, linting and shared settings
GitHub: A website and service that hosts Git repositories so a team can share them, review changes in pull requests and run automation.In How code becomes a running program, What AI coding tools can and cannot do
GitHub Actions: GitHub's automation service. It runs the commands you list (lint, test, build, deploy) on fresh machines whenever something happens in the repository.In How code becomes a running program
GitHub flow: A branching strategy with one long-lived branch, main. Every change goes through a short feature branch and a pull request, and main is always deployable.In Branching strategies, releases and versions
GitHub Release: A GitHub page built on a tag, with a title, release notes and optional files such as an APK, so people can see and download what shipped.In Branching strategies, releases and versions
GITHUB_TOKEN: A token GitHub creates for each job so it can call the GitHub API on the repository. It expires when the job ends, and its rights are set by the permissions key.In Deploying and securing with GitHub Actions
glob: A pattern with * or ? that the shell turns into the list of matching file names, like *.log.In Shell scripting for daily jobs
global importance: A single ranking of how much a model relies on each feature across all its predictions.In Explaining and choosing models
Global load balancing: Routing users across regions or data centres, usually by DNS or anycast, to the nearest healthy one.In Load balancers
global model: One model trained on many series at once, with the series ids and their attributes as features, so patterns learned on one series help the others.In Forecasting demand
globals: Names a file can use without defining them, such as process and console in Node.js or window and document in a browser.In VS Code for teams: formatting, linting and shared settings
glue work: Tanya Reilly's term for the less visible work that makes a team succeed, such as onboarding, reviewing designs, unblocking people and writing things down.In Delegation and mentoring
Go-to-market: The plan for how a product reaches its first customers in a market: which customers, in which places, through which channels, in what order.In Growth loops and local go-to-market
Go-to-market (GTM): The plan for how a product reaches and wins its users: channels, pricing, messaging, sales. Growth loops and GTM
golden set: A fixed set of real test cases, each with the expected result or a pass/fail rule written by an expert, used to grade every new version.In Evals: measuring AI quality
golden tasks: A test case for an agent: a fixed starting situation, a user request, and checks on the end state and on the steps taken.In Evaluating agents
Goodhart's law: The idea that when a measure becomes a target, people change their behaviour to hit it, and it stops measuring what it used to.In Team design and delivery health
goods received note: A record that a store or DC received a delivery: which supplier, which products, how many of each. Saving it adds the stock.In APIs done right, Designing AI features: LLMs, RAG and vector search, What makes an agent, Retrieval over procedures, Exploratory testing, Leading AI adoption
Googlebot: Google's web crawler: a program that downloads pages and follows the links on them to find more pages.In Google Search: crawling, indexing and answering in a fraction of a second
governed dataset: A curated, documented dataset on the semantic layer with certified metrics, defined joins and access rules, offered to people who build their own reports.In Self-service without chaos
GPUs: Graphics processing unit: a chip with thousands of small cores, now the most common hardware for running AI models.In Claude: how Anthropic serves Claude at scale
graceful degradation: Designing a system so that when part of it fails, the rest keeps working with fewer features instead of failing completely.In Failure is normal: timeouts, retries, backoff and circuit breakers, Capacity, resilience and game days
Gradient boosting: An ensemble that builds small trees one after another, each fitted to the errors the previous trees still make, and adds their predictions.In Training and comparing models
Grain: The level of detail of one row, for example one row per store per day, or one row per sale.In Asking a question data can answer, Cleaning data in a spreadsheet, Pivot tables and summaries, Analysing with an AI assistant, From questions to a reporting model, Metrics everyone agrees on
Grandfathering: Letting existing customers keep their old price for a set time, or until renewal, after a price rise.In Pricing and packaging
Grid search: Trying every combination of a small list of hyperparameter values and keeping the best by cross-validated score.In Training and comparing models
GRN: Goods received note: the record a store makes when a delivery arrives, listing what was received.In CAP, PACELC and consistency, Evaluating agents
gross margin: Sales minus the cost of the goods sold, before delivery, payment, packaging and discounts.In Unit economics: does each order make money?, Telling the story
Grounding: Making answers rest on supplied sources, ideally with citations. RAG
Group sequential designs: A test plan with a fixed number of interim looks, each with its own stricter threshold, chosen so the overall false-alarm rate stays at alpha.In Designing and reading an A/B test
groupby: A pandas method that splits rows into groups by the values of a column, runs a calculation such as sum on each group and puts the results into a new table.In The same analysis in Python
growth loop: A closed cycle in which the output of one round, such as new customers or revenue, becomes the input to the next round, so growth compounds.In Growth loops and local go-to-market
GSI: Global secondary index: a second copy of the table with different keys, kept up to date automatically and read with eventual consistency.In Amazon: the store that never closes
guard: In NestJS, a class that decides whether a request may reach its route at all, usually by checking a token, an API key or a role.In From script to service: a small API in Node, Your first NestJS service
guardrail: A metric that must not get worse while you chase the goal, such as error rate, complaints or orders sent twice.In From idea to one-page spec, Launch and learn, What a product manager actually does, Designing AI features people can trust, Designing AI features: LLMs, RAG and vector search, Guardrails and red-teaming
guardrail metric: A metric that must stay inside a limit while the team pushes its goals, so a target cannot be hit by damaging something else. Also called a counter-metric or health metric.In OKRs and outcome goals, Writing specs people read

H

H3: Uber's open-source system that covers the earth with hexagons at 16 sizes. Every point maps to one cell ID at each size.In Uber: matching riders and drivers in real time
haircut: A deliberate cut to an estimated benefit, to allow for the part that won't turn into cash or won't happen as planned.In Business cases a managing director will sign
hallucinated: When an AI model states something that sounds right but is not true or not in its input, such as a column, a number or a source that does not exist.In Analysing with an AI assistant
hallucination: When a language model produces something that sounds right but is not true, such as a function, option or package that does not exist.In Working with AI coding assistants, How AI features work, just enough for PMs, What AI coding tools can and cannot do, Designing AI features: LLMs, RAG and vector search
Halo effect: When one strong impression, good or bad, colours how you judge everything else about a person.In Hiring well
handoff: A point where work or data passes from one person, team or system to another.In Watching the work
happy path: The path through a flow where everything works and the user makes no mistakes. You design it first, then add the branches.In Screens and flows, Reading and fixing what the AI wrote
hard links: A second directory entry pointing at the same file on disk. Nothing is copied, so it is instant, but both names share one file.In Upgrades, maintenance and on-call
hash: A one-way function: easy to compute from the password, but practically impossible to reverse back into the password.In Security essentials, Inside the .git folder: objects, refs and HEAD
hash chain: A list where each entry stores a hash of the entry before it, so changing any entry breaks every hash after it.In Shipping an agent safely
hash function: A function that turns any input, such as a string, into a number that looks random but is always the same for the same input.In Sharding and consistent hashing, Cryptography without the maths
HCL: HashiCorp Configuration Language: the block-based text format that Terraform and OpenTofu files use.In Infrastructure as code
HEAD: Git's marker for where you are now. Usually it names the current branch, for example ref: refs/heads/main.In Branches and merging
head branch: The branch that contains the proposed changes in a pull request, for example feature/festival-offer.In Pull requests and code review
Head sampling: Deciding whether to keep a trace when it starts, before you know how it went, for example a random 10%.In Observability for LLM systems, Observability with OpenTelemetry
head-of-line blocking: When one lost or slow item at the front of a line holds up everything behind it.In Networking you actually need
headroom: Capacity you keep unused at the expected peak, so failures, forecast errors and sudden waves do not push the system past its limit.In Capacity, resilience and game days
health check: A regular test the load balancer runs to decide whether a backend server should receive traffic.In Load balancers
health gate: A short list of measurable conditions that must all be true before an update moves to the next wave.In Looking after 1,400 devices
heap: Postgres's name for the main storage of a table: its rows, in no particular order, spread over 8 KB pages.In Indexes from the inside, Objects, arrays and references
Heap snapshot: A file listing every live object in a JavaScript process and what holds it, used to find memory leaks. Performance and memory in JavaScript
heartbeat: A small message sent on a timer just to show a connection is still alive, so proxies do not close it and both sides notice a dead peer.In Real-time updates: polling, WebSockets, SSE and push
heat map: A capability map coloured by a score, so the capabilities that most need investment stand out.In Capability maps
Hermes: The JavaScript engine built for React Native. Release builds compile JavaScript to Hermes bytecode ahead of time.In How React Native works
Hermes bytecode: A compact, precompiled form of your JavaScript that the Hermes engine can run directly, without parsing and compiling the code on the phone.In Fast on a cheap phone
heuristic: A rule of thumb that helps you think of test ideas quickly. It can be wrong, and it does not replace judgement.In Exploratory testing
heuristic evaluation: A review where a few people check a design against usability heuristics on their own, then merge and rate the problems they found.In Design for PMs
Hexagonal architecture: Business logic in the centre with ports, and adapters that connect it to databases, APIs and UIs. Architecture styles
hidden class: An internal description V8 keeps for each object layout: which properties exist and in what order. Objects with the same layout share one hidden class. Also called a shape or map.In Performance and memory in JavaScript
Hidden state: Variables a notebook still holds in memory from cells run earlier, even if those cells were later changed or deleted.In From notebook to pipeline
high-cardinality: A feature with many distinct values, such as a store-product ID or a timestamp.In Explaining and choosing models, Observability with OpenTelemetry
higher-order function: A function that takes another function as an argument, returns a function, or both. map, filter and a discount-rule factory are examples.In Functions, scope and closures
Hindsight bias: The tendency, once you know how events turned out, to believe the outcome was obvious and predictable at the time.In Postmortems and learning
histogram: A metric that counts values into ranges called buckets, such as how many requests took under 100 ms, under 300 ms, and so on. Percentiles are computed from it.In Observability: logs, metrics and traces, Charts that answer the question, Describing data honestly
hit ratio: Hits divided by all lookups. A 95% hit ratio means 95 of every 100 requests were answered from the cache.In Caching
HLS: HTTP Live Streaming: a format from Apple where a video is cut into small files plus text playlists that list them. Players on phones, TVs and browsers support it.In Netflix: streaming to every screen
HMAC: Hash-based message authentication code: a hash computed with a shared secret key, proving a message came from someone who holds that key and was not changed.In Cryptography without the maths
HMAC signature: A hash of the message with a shared secret, proving it came from the sender and was not changed. No overselling
HNSW: Hierarchical navigable small world: an approximate nearest-neighbour index that links vectors into a layered graph, so a search visits a small part of it instead of every row.In Designing AI features: LLMs, RAG and vector search
hoisting: JavaScript creating all the names declared in a scope before running its first line, so they exist (in some state) from the top of the scope.In Functions, scope and closures
holdout: A random part of the target group that deliberately gets no treatment, so you can measure what the treatment changed.In Classification and churn
Hook: A function starting with use that lets a component use React features such as state or effects. State and effects
Horizontal scaling: Running more copies of the same app on several machines and spreading requests across them. Also called scaling out.In Scaling up vs scaling out
host: To run an app on a server connected to the internet, so people can open it with a link.In What AI coding tools can and cannot do, Ship it and know your limits
host components: A built-in React Native component, such as View or Text, that maps to a real platform view instead of to other components.In How React Native works
host discovery: Finding which addresses in an in-scope range are live and responding, so you test real hosts instead of empty addresses.In Reconnaissance
hot: Code that runs very often, such as the body of a loop over a million rows. Speeding it up has a big effect.In Performance and memory in JavaScript
hot shard: A shard that gets much more traffic than the others, because a few very busy key values live on it.In Sharding and consistent hashing
hot standby conflict: A clash on a standby between replaying a WAL record and a query that still needs the data that record changes or removes.In Replication and failover
HOT update: Heap-only tuple: a new row version on the same page that needs no new index entries, because no indexed column changed.In Inside a PostgreSQL server
Hotfix: A small, urgent fix released outside the normal schedule, through the same checks, usually as a patch version. Branching strategies, releases and versions, Project: ship a small app end to end
HSMs: Hardware security module: tamper-resistant hardware that generates, stores and uses keys without ever exporting them.In Cryptography without the maths
HTTP: HyperText Transfer Protocol: the agreed text format for asking a server for something and getting an answer.In What happens when you open an app or website, Ship it and know your limits
hunk: One block of changed lines in a diff, with a few unchanged lines around it for context. A file can have several hunks.In Everyday Git: status, diff, log and good commits, Merge conflicts without panic
hurdle rate: The yearly return a company requires from a project before it will invest; future cash is discounted at this rate. Also called the discount rate.In Business cases
Hybrid search: Combining keyword search and vector search, then merging the results. RAG
hydrates: Attaching React's event handlers and state to HTML that was already rendered on the server, so the page becomes interactive without redrawing it.In Loading server data
hydration: The moment React's JavaScript has loaded in the browser and attached itself to the HTML the server sent, so buttons and forms start using React code.In Forms that people can trust
Hypercare: A period of extra-close support right after a launch or wave, with daily check-ins and fast fixes.In Launching: pilots, waves and the review after
hyperparameters: Settings chosen before training that control how a model learns, such as tree depth, number of trees or learning rate. They are tuned on validation data.In Training and comparing models
hypotheses: A possible explanation, written down before the analysis, with the evidence you would expect to see if it were true.In Asking a question data can answer
hypothesis: A specific, testable statement about attacker behaviour that you can confirm or refute with data you already collect.In Threat hunting

I

IaaS: Infrastructure as a service: renting virtual machines, networks and disks; you manage the operating system and everything above it.In Networks, endpoints and the cloud
IaC: Infrastructure as code: servers, buckets, networks and permissions described in files such as Terraform or Kubernetes manifests, so they can be reviewed and versioned like code.In Security in the pipeline and in AI features
ICE: Individual conditional expectation: one line per row showing how that row's prediction changes as one feature changes.In Explaining and choosing models
IDE: Integrated development environment: an editor plus built-in tools to run, debug and test code, such as Android Studio or IntelliJ.In How code becomes a running program
ideal days: A day of focused work on planned tasks with no meetings, support questions or interruptions.In Slicing and planning work
idempotency: A property of an operation that gives the same result whether it runs once or several times.In Technical vision and design docs
idempotency key: A unique value the client sends with a request, such as a UUID. The server remembers it, and if the same key arrives again it returns the first result instead of doing the work again.In APIs done right, Failure is normal: timeouts, retries, backoff and circuit breakers, Forms that people can trust, Background jobs and idempotency, Screens and flows, Data model and API
idempotent: An operation is idempotent if doing it twice has the same effect as doing it once. Setting a price to 120 is idempotent; adding 5 to stock is not.In APIs done right, Queues, background jobs and pub/sub, Case study: a URL shortener, Background jobs and idempotency, User stories and acceptance criteria, Platforms, integrations and systems of record
identity provider: The one service that holds staff accounts and sign-in rules and tells other apps who a user is, using standards such as SAML or OpenID Connect.In Identity: passwords, MFA and passkeys
identity provider (IdP): An IdP: the service that checks who you are (password, MFA) and then tells other apps, in a signed message, that you are you.In Accounts, SSO and MFA
IDOR: Insecure direct object reference: the API returns or changes a record only because the caller knows its id, without checking that the caller may access that record.In Security essentials
IDS: Intrusion detection system: a tool that watches network traffic or host activity and alerts on patterns that match known attacks.In Attacking networks and services
image digest: A hash of a container image's exact contents, such as sha256:4b1f…. Unlike a tag, it can never point to a different image.In Experiments, versions and the model registry
IMMUTABLE: A Postgres label for a function that always returns the same output for the same input, regardless of time zone or settings. Only immutable functions can be used in an index.In Indexes from the inside
immutable update: Changing data by building a new object or array that contains the change, instead of editing the existing one. Unchanged parts are reused, not copied.In Objects, arrays and references
Impact: The business result that the outcome leads to, such as fewer cut orders or lower write-offs. It usually has many causes and shows up late.In OKRs and outcome goals, Incidents, problems and changes
impact analysis: Following the relationships in a model to find everything that depends on a given element, before you change or remove it.In Frameworks and governance
implementation detail: Anything about how a component works inside that neither the person at the screen nor the code using the component can see: state names, hooks, CSS classes, internal functions, render counts.In Testing React like a user
improper output handling: Passing model output to a browser, shell, database or other system without validating, escaping or limiting it, so the output can act as code there.In Cost, latency and safety
impurity-based importance: Feature importance computed from how much each feature reduced the error inside the trees during training (mean decrease in impurity).In Explaining and choosing models
incident: An unplanned interruption or loss of quality in a service that affects users or the business, such as tills that can't take card payments.In Running products in production, Incidents, problems and changes, Risk, compliance and the first hour of an incident
Incident Command System: A way of organising an emergency response, first used by fire services, where one commander delegates clear roles so the team can grow without chaos.In Leading an incident
Incident commander: The person in charge of an incident: holds the overall picture, assigns roles and tasks, and makes the decisions. Usually shortened to IC.In Leading an incident, Incident response
incident lead: The one person who runs an incident: sets severity, assigns tasks, decides, and owns communication until the incident is closed.In Risk, compliance and the first hour of an incident
incident manager: The person who runs a major incident: decides who does what, keeps time and makes sure updates go out. They usually do not fix things themselves.In Incidents, problems and changes
Income statement (P&L): Revenue, costs and profit over a period. Three statements
incremental: Cash that changes because of the decision: money you would spend or save only if you go ahead, compared with the do-nothing baseline.In Business cases a managing director will sign
incremental cash flows: Cash that changes because of the decision: money you would spend or save only if you go ahead, compared with the baseline.In Business cases
incremental extract: An extract that reads only the rows that are new or changed since the last successful run, instead of the whole table.In Data and feature pipelines
incremental IRR: The IRR of the difference between two options' cash flows; if it is above the hurdle rate, the bigger option is worth its extra cost.In Business cases
index: A series rescaled so that a chosen base period equals 100; every other value shows the percentage of the base, so groups of different sizes can share one chart.In Charts that answer the question
Index (staging area): The list of changes that will go into the next commit; git add puts changes there. What Git really is: snapshots and the three areas
indicator of compromise: Indicator of compromise: a specific observable, such as a file hash, domain or IP address, that is known to be linked to an attack.In Writing detections
Indicator of compromise (IOC): A trace of known bad activity, such as a file hash, domain or IP address. Threat hunting
inference: TypeScript working out a type by itself from the value, such as number from const qty = 2.In TypeScript for a JavaScript team
Information hierarchy: The order in which a screen's content is noticed and understood, set by size, weight, contrast, position and grouping.In Design for PMs
infrastructure as code: Describing servers, networks, databases and DNS in versioned files, such as Terraform, and applying them with a tool instead of clicking in a console.In Shipping safely: containers, CI/CD and releases, Infrastructure as code
Infrastructure as code (IaC): Servers, networks and permissions described in files, reviewed and applied like code. IaC
Ingest time: The moment the log reached the SIEM and became searchable.In Logs that matter
Ingress: A Kubernetes object that maps outside HTTP hosts and paths to Services. It needs an ingress controller, such as an Nginx- or Envoy-based one, to do anything.In Kubernetes for application teams
Inhibition: An Alertmanager rule that mutes some alerts while another, related alert is firing, so one problem sends one notification.In Alerting that does not burn people out
Injection: A flaw where input meant as data reaches an interpreter as part of its instructions, so the input can change what the interpreter does.In Injection and unsafe input
inline cache: A small memory at one spot in the code that records which object shapes it has seen, so the next access with the same shape is fast.In Performance and memory in JavaScript
INNER JOIN: A join that returns only pairs of rows that match the ON condition. Rows without a partner on either side are dropped.In SQL you'll actually use
innovation tokens: McKinley's idea that a company can afford only a few new, unproven technologies at once, so each one should be spent where it gives an edge.In Technology strategy
inodes: The small record Linux keeps for each file. A filesystem has a fixed number, so millions of tiny files can use them all up while space is left.In Linux from the command line
input metric: A metric a team can change directly through its own work, chosen because changing it is expected to move the north star.In North star, input and guardrail metrics
insecure direct object reference: A flaw where the server loads an object using an identifier the caller supplies, without checking the caller may access that object. Often shortened to IDOR.In Broken access control, Attacking web applications
instance: One running copy of your app, such as one container or one VM. Also called a replica.In Scaling up vs scaling out
Instrumentation: Code that creates telemetry. Auto instrumentation patches common libraries at startup; manual instrumentation is code you write yourself.In Observability with OpenTelemetry
integer cents: Money kept as a whole number of the smallest unit, so $6.40 is stored as 640. Whole numbers add up exactly.In Reading and fixing what the AI wrote
integration: A connection that moves data or requests between two systems, such as the Store app sending a GRN to the ERP.In Platforms, integrations and systems of record
integration spaghetti: A tangle of point-to-point connections between systems, each built for one project, that nobody can fully draw or safely change.In Why enterprise architecture exists
integration test: A test that runs several real pieces together, such as an API endpoint with its validation, service code and a real database.In A test strategy, not a pile of scripts
integrity hashes: A hash of the package file recorded in package-lock.json. npm checks the downloaded file against it, so a changed file is rejected.In Security and quality in JavaScript projects
intercept: The predicted value of y when every x is zero: where the fitted line crosses the vertical axis.In Regression and correlation
intercepting proxy: A tool that sits between your browser and the server so you can see, pause and change each request. Used by testers on systems they are allowed to test.In Broken access control, Attacking web applications
interceptor: A NestJS class that runs code before and after every route handler, for example to log or time requests.In Measuring speed and reliability: p99, SLOs and nines, Your first NestJS service
interest: The need, worry or goal behind a position, such as not adding staff hours in the busiest weeks.In Stakeholders, negotiation and leading without authority
interference: When one unit's version changes the outcome of another unit, for example two groups competing for the same riders or delivery slots.In Designing and reading an A/B test
intermittent demand: Demand with many zero days and occasional sales, typical of slow-moving products.In Forecasting demand
interpreter: A program that runs code directly, statement by statement, without producing a separate machine-code file first.In How code becomes a running program, Injection and unsafe input
interview loop: The full set of interviews and exercises a candidate goes through for one role.In Hiring well
interview snapshot: A one-page summary of a single interview: quick facts, a memorable quote, the story as a simple map, the opportunities and other insights.In User interviews that don't lie
intraclass correlation: The share of all variation in a metric that comes from differences between groups (here, stores) rather than within them; 0 means groups are no more alike inside than outside.In Designing and reading an A/B test
Invalidation: Removing or updating a cached copy because the real data changed.In Caching, Files, object storage and CDNs
inverse Conway manoeuvre: Designing team boundaries on purpose so that the system architecture you want becomes the natural one to build.In Team design and delivery health
inverted index: A map from each word to the sorted list of documents that contain it, like the index at the back of a book.In Google Search: crawling, indexing and answering in a fraction of a second
IOPS: Input and output operations per second: how many disk reads and writes the storage can do each second. Cloud volumes have a fixed limit.In Upgrades, maintenance and on-call
IP address: A number that identifies a machine on a network, like 203.0.113.10. The next lesson explains public and private addresses.In What happens when you open an app or website, Networking you actually need
IPsec: A standard set of protocols for encrypting IP packets between two gateways. IKE is the part that agrees on the keys.In Networks from the till to the data centre
IQR: Interquartile range: the third quartile (p75) minus the first quartile (p25). It is the width of the middle half of the data.In Describing data honestly
IRR: Internal rate of return: the discount rate at which an investment's NPV is exactly zero.In Business cases a managing director will sign, Business cases
ISMS: Information security management system: the policies, risk process, roles and reviews an organisation uses to run security as a repeating cycle.In Risk, compliance and the first hour of an incident
ISO 8601: The international date format year-month-day, such as 2026-10-18, which every tool reads the same way.In Cleaning data in a spreadsheet
ISO/IEC 25010: An international standard that defines a product quality model: the characteristics, such as reliability and security, that together make up software quality.In What testing is for, Requirements that work
isolation level: A setting that decides how much one transaction can see of other transactions running at the same time. Stricter levels prevent more anomalies but cause more retries.In Transactions, ACID, isolation and locks
ISPs: Internet service provider: the company that brings the internet to your home or phone, like a broadband or mobile operator.In Netflix: streaming to every screen
issue: A numbered item in a GitHub repository for one piece of work: a bug, a feature, a task or a question, with a discussion thread.In Planning on GitHub: issues, labels and Projects
issue form: A YAML file in .github/ISSUE_TEMPLATE that turns new issues into a web form with required fields, dropdowns and checkboxes.In Planning on GitHub: issues, labels and Projects
ITIL: IT Infrastructure Library: a widely used collection of good practices for running IT services. Version 4 is the current one.In Incidents, problems and changes
ivory tower: An architecture team that works apart from delivery teams, producing models and standards that nobody uses.In Why enterprise architecture exists
IXP: Internet exchange point: a building where many networks connect to each other directly to swap traffic.In Netflix: streaming to every screen

J

jank: Visible stutter: scrolling, animations or taps that skip or lag because frames were not ready in time.In Fast on a cheap phone
JavaScript engine: The program that reads JavaScript and executes it, such as V8 in Chrome and Node.js, or Hermes in React Native.In How code becomes a running program
JIT: Just-in-time compilation: compiling the busy parts of a program to machine code while it runs.In How code becomes a running program
JIT compiler: Just-in-time compiler: it compiles code to machine code while the program is running, using what it has seen so far to make good guesses.In Performance and memory in JavaScript
jitter: Adding randomness to retry waits so that many clients do not retry at the same instant.In Failure is normal: timeouts, retries, backoff and circuit breakers, How LLM APIs work, Offline first
JML process: Joiner, mover, leaver: the three moments when a person's access must change, tied to HR events.In Accounts, SSO and MFA
job: One unit of background work stored in a queue: a name plus the data the worker needs, such as a customer id and an SMS text.In Queues, background jobs and pub/sub, GitHub Actions: continuous integration
job scheduler: A BullMQ feature that stores a repeating schedule in Redis under a fixed id and creates one job each time it is due.In Scaling up vs scaling out
job shadowing: Spending a working day or shift next to someone, watching and timing their real work and asking about it at natural pauses.In Stakeholders and elicitation
job story: A one-sentence description of a need in the form: When situation, I want to motivation, so I can expected outcome.In Problems before solutions
Jobs to be done: The idea that people hire a product to make progress in a particular situation, and that the situation, not the person's demographics, explains what they choose.In Problems before solutions
join: Putting two tables side by side by matching rows on a shared key column, such as store_id.In The same analysis in Python
joiner-mover-leaver: The process that creates, changes and removes access when people join, change roles or leave, ideally driven by the HR system.In Identity: passwords, MFA and passkeys
Joiner, mover, leaver (JML): The three moments when someone's access must change: they start, change role or leave. Accounts and MFA
journal: The systemd log store. It keeps messages from every service with a timestamp, the service name and a priority.In Linux from the command line
journey map: A picture of one person's steps over time in one scenario, with what they do, use, think and feel, and where it hurts.In Watching the work
JS thread: The thread where your JavaScript runs: React renders, state updates, event handlers and most of your app logic.In Fast on a cheap phone
JSI: JavaScript Interface: a C++ API that lets JavaScript hold references to C++ objects and call their methods directly, without turning data into JSON.In How React Native works
JSON Merge Patch: A simple PATCH format (RFC 7396): send the fields to change; a field set to null is removed. Media type application/merge-patch+json.In Designing an API people can use
JSON Pointer: A standard string (RFC 6901) that points to one place inside a JSON document, such as /lines/2/qty for the qty of the third line.In Designing an API people can use
JSON Schema: A standard way to describe the shape of JSON data: which fields exist, their types, which are required, and which values are allowed.In Prompts and structured output, Tools and the Model Context Protocol
JSON-RPC: A small standard for remote calls written as JSON: each request has an id, a method name and parameters, and each response carries the same id with a result or an error.In Tools and the Model Context Protocol
JSONL: JSON Lines: a text file with one complete JSON object on each line, easy to append to and to diff in a pull request.In Evals: knowing if it got better
JSX: A syntax extension that lets you write HTML-like tags inside JavaScript. A build tool turns each tag into a function call before the browser runs it.In Components, props and how React renders
junction table: A table whose rows each link one row of table A to one row of table B, used for many-to-many relationships.In Modeling data in tables
JUnit XML: A common XML format for test results, read by CI tools to show counts and failures.In Stable suites in CI
JWKS: JSON Web Key Set: a URL where a login server publishes its public signing keys, each with an id (kid).In Who are you and what may you do?
JWT: JSON Web Token: a small signed JSON object, such as user id, role and expiry, that the server can verify with a key without a database lookup.In Security essentials, Deploying and securing with GitHub Actions, Who are you and what may you do?, Authentication, sessions and tokens
JWT (JSON Web Token): A signed, base64url-encoded set of claims used as a self-contained access token. Auth and limits

K

K-fold cross-validation: A validation method that shuffles rows into K groups and trains K times, each time holding out one group.In Features and leakage
Kafka: A distributed log where producers append messages and consumers read them in order, at their own pace. Used as a durable queue between systems.In Netflix: streaming to every screen
keep-alive: Keeping a TCP connection open after a response, so the next request can use it without a new handshake.In What happens when you open an app or website
Keep-the-lights-on (KTLO): Keep the lights on: the work needed to keep a product running, such as bug fixes, support tickets, upgrades, security patches and small urgent changes.In Outcome roadmaps
kernel: The three-part core of a strategy in Richard Rumelt's model: a diagnosis of the challenge, a guiding policy for dealing with it, and coherent actions that carry it out.In Vision, strategy and why now, The same analysis in Python, Containers from the inside, Technology strategy
key: The full name of an object inside a bucket, such as articles/1001/9f2c.jpg. Slashes only look like folders.In Files, object storage and CDNs, Components, props and how React renders, State, effects and where data should live
Key exchange: A method, such as Diffie-Hellman, that lets two parties agree a shared secret key over an open network without ever sending the key itself.In Cryptography without the maths
key pair: Two matching files: a private key that never leaves your laptop, and a public key you copy to servers. A server lets you in when you prove you hold the private key.In Linux from the command line
Kill criteria: Conditions, agreed before a test starts, that will make you stop or change the idea, written as a measurable state and a date.In Idea lab: judging and pitching ideas
kill switch: One switch that turns off a dangerous capability everywhere at once, used in an incident.In Shipping an agent safely, Cost and latency at scale, Toil and automation
kiosk mode: A locked-down mode where a device runs only one app, or a short list of apps, and staff cannot reach settings or other apps.In Looking after 1,400 devices
KMS: Key management service: a cloud or on-premises service that stores master keys, performs encrypt and decrypt operations with them, enforces who may use each key and logs every use.In Cryptography without the maths
knock-out criterion: A must-have condition. Any option that fails it is removed before scoring, however well it scores on everything else.In Choosing an architecture style
knowledge areas: A group of related business analysis tasks in the BABOK Guide, such as Elicitation and Collaboration.In What business analysts do
Knowledge article: A short, searchable fix for a known issue, written so the first line can solve it. Service desk
known error: A problem whose cause is understood and written down, usually with a workaround, but not yet permanently fixed.In Incidents, problems and changes
Known Exploited Vulnerabilities: A public list, kept by the US agency CISA, of software flaws that attackers are known to be using in real attacks. A good order for patching.In How breaches really happen
KPI: Key performance indicator: one of the few measures a team has agreed tells it whether it is succeeding.In Designing a dashboard for decisions
KV cache: Key-value cache: the numbers a model keeps for every earlier token in a request, so it does not recompute the whole text for each new token. It lives in GPU memory and grows with every token.In ChatGPT: serving a large language model to hundreds of millions

L

L4 load balancer: A load balancer that works at the transport layer. It sees IP addresses and ports, and forwards whole TCP connections without reading what is inside.In Load balancers
L7 load balancer: A load balancer that works at the application layer. It decrypts HTTPS and reads the HTTP request, so it can route by path, host or header and balance each request separately.In Load balancers
Label (target): The outcome the model learns to predict. Framing
label delay: The time between making a prediction and learning the true outcome. For a next-day sales forecast it is about one day.In Monitoring, drift and retraining
label window: The period after the as-of date that decides the label, here 30 days.In Classification and churn
labels: Key-value pairs attached to a metric, such as route or status code, that let you split it. Each unique combination is stored as its own time series.In Observability: logs, metrics and traces, Framing a problem as a model, Features and leakage, Kubernetes for application teams
lag feature: A feature that copies the label from an earlier day, such as sales 7 days before the day being predicted.In Features and leakage
lagging indicator: A metric that tells you whether you succeeded, but only after the fact, such as quarterly write-offs or revenue.In OKRs and outcome goals
LAN IP address: The address your laptop has on the local Wi-Fi network, such as 192.168.1.20. Other devices on the same network use it to reach the laptop.In Building the app
landing zone: A prepared cloud environment with an account structure, identity, guardrails, networking and logging already set up, so new workloads start safe.In Cloud architecture and cost
Lanes: A row inside a pool for one role or department, such as sales rep or finance.In Modelling processes
large language model: Large language model: a program trained on a very large amount of text and code that predicts the most likely next piece of text, given everything before it.In Working with AI coding assistants, What AI coding tools can and cannot do, A spec the AI can follow, Designing AI features: LLMs, RAG and vector search, Analysing with an AI assistant
large language model (LLM): A program trained on huge amounts of text to predict the next piece of text; it reads text in and writes text out.In How AI features work, just enough for PMs, Security in the pipeline and in AI features
last complete week: The most recent full week that has ended, for example Monday to Sunday, as opposed to the current week so far.In Designing a dashboard for decisions
Latency: How long one request takes from start to finish, for example 200 ms.In Speed numbers and back-of-the-envelope math
latency budget: The maximum time a request may take, agreed in advance and split between the steps that serve it.In Serving: batch and online
lateral movement: Moving from the first machine an attacker controls to other machines and accounts inside the same organisation.In How breaches really happen, Networks, endpoints and the cloud, Privilege escalation and lateral movement
launch tier: A level of launch effort, from a full plan with training and leadership communication for big changes to a changelog line for small ones.In Launching: pilots, waves and the review after
layers: A level where a setting can be set. Lower layers are overridden by higher ones: default, then user, then workspace, then folder.In VS Code essentials, Containers from the inside
layout: The browser step that calculates the size and position of every element on the page.In JavaScript in the browser
lead time: The time between placing an order and the goods being ready to sell.In Framing a problem as a model, Value streams and processes
Lead time for changes: The time from a change being committed to version control until it runs in production.In Team design and delivery health
leading indicator: A metric that moves early and predicts a lagging one, so a team can steer by it week to week.In OKRs and outcome goals
leading question: A question that hints at the answer the interviewer expects, so people tend to agree with it.In User interviews that don't lie
leak: Memory that the app keeps holding after it no longer needs it, usually because something still references it, such as a listener that was never removed.In Fast on a cheap phone
Leakage: When buyers and sellers who met on a platform move their repeat transactions off it, so the platform stops earning from them.In Marketplaces, platforms and moats, Features and leakage
learning rate: In gradient boosting, the fraction of each new tree's prediction that is added, such as 0.1. Smaller values need more trees but usually generalise better.In Training and comparing models
learning review: A review of an event that went well or nearly went wrong, held to learn how people succeeded under pressure, not only why things failed.In Postmortems and learning
lease: A token the cache gives to the first client that misses a key. Only the holder of a valid lease may fill that key, and a delete cancels the lease.In Facebook: the social graph, TAO and memcache at planet scale, Networks from the till to the data centre
least frequency of occurrence: Grouping events by a value and sorting by how rarely each appears, so the rare values that an attacker often produces float to the bottom of the list.In Threat hunting
least privilege: Giving each person or service only the permissions it needs for its job, and nothing more.In Running Postgres in production, Accounts, SSO and MFA, Thinking like an attacker, Injection and unsafe input
Least squares: The method that picks the line with the smallest possible sum of squared residuals.In Regression and correlation
LEFT JOIN: A join that keeps every row from the left table. Where no right-side row matches, the right-side columns are filled with NULL.In SQL you'll actually use
legacy system: A system that the business depends on today and that has become slow, risky or expensive to change.In Evolving a legacy system
Level: The typical amount a series sits around right now, ignoring weekly and festival ups and downs.In Forecasting demand
leverage: How much output a task creates for each hour you spend on it, counting the effect on everyone it touches.In The manager's job
LFU: Least frequently used: when full, remove the item that is read least often.In Caching
libpq: The official PostgreSQL client library in C. psql and many drivers in other languages are built on it.In Replication and failover
library: A package of ready-made code that you load into your program with import, so you do not write everything yourself.In The same analysis in Python
libuv: The C library inside Node that provides the event loop, timers, network and file I/O. Some work, such as file reads, it runs on a small pool of background threads.In The event loop
licence: The legal terms that say how a piece of code may be used, changed and shared.In What AI coding tools can and cannot do
lie-fi: A connection that looks available (the phone shows bars or Wi-Fi) but is so slow or lossy that requests hang or time out.In Offline first
lifecycle rule: A storage rule that moves or deletes objects automatically after a set age, for example to a cheaper archive tier after 90 days.In Cloud and SaaS costs with FinOps
lifting state up: Moving state from a child component up to the closest parent that all the components using it share, then passing it down as props.In State, effects and where data should live
like-for-like: A comparison that uses only the stores trading for the whole of both periods, over matching calendar weeks, so that openings, closures and the calendar do not distort the change. Also called same-store or comparable sales.In Asking a question data can answer, Pivot tables and summaries, Telling the story
line of visibility: The line in a service blueprint that separates what the customer can see from work done out of their sight.In Watching the work
lineage: The chain of models and sources a table is built from, and the models that depend on it, shown as a graph.In Transforming data in the warehouse, Experiments, versions and the model registry
linearizable: Linearizable: once a write is confirmed, every later read on any machine returns that value or a newer one. The system behaves as if it had one copy.In CAP, PACELC and consistency
link-local address: An address a device gives itself (from 169.254.0.0/16) when it asked for an address and got no answer. It only works on the local cable or Wi-Fi.In Networks from the till to the data centre
linter: A tool that reads code without running it and reports likely bugs and risky patterns, such as an undefined variable or == instead of ===.In VS Code for teams: formatting, linting and shared settings
liquidity: How reliably a marketplace matches a request with the other side within an acceptable time, such as an order delivered on time or a listing sold within a week.In Marketplaces, platforms and moats
liquidity management: Making sure each agent has enough cash and enough e-money at the right time to serve customers.In Teardown: sending money with a mobile wallet
listens: Waits for incoming connections on a port. A program that listens is a server for that port.In Networking you actually need
literal types: A type that is one exact value, such as 'wallet', instead of any string.In TypeScript for a JavaScript team
Little's law: A rule from queueing theory: the average number of items inside a system equals the arrival rate times the average time each item spends inside.In Speed numbers and back-of-the-envelope math, Capacity, resilience and game days, Performance and load testing
live region: An element marked with aria-live, or a role such as status or alert, whose text changes are read out by screen readers without moving focus.In Accessible, fast pages, Testing React like a user
liveness: A health check that only answers: is the process alive and able to respond? Used to decide whether to restart it.In Load balancers
Liveness and readiness probes: Kubernetes checks: liveness restarts a stuck container, readiness keeps traffic away until it is ready. Kubernetes
LLM as judge: Using a model with a rubric to score another model's output. Evals
LLM judge: A separate model call that grades another model's output against a written rubric and returns pass or fail with a reason.In Evals: measuring AI quality, Evals that never stop
load average: The average number of tasks running, ready to run, or stuck waiting for the disk, over the last 1, 5 and 15 minutes.In Linux from the command line
load balancer: A server or service that receives all incoming requests and forwards each one to one of several backend servers.In Load balancers
Load shedding: Deliberately rejecting some requests when a system is overloaded, so the requests it does accept still finish quickly.In Rate limiting and protecting your system, Capacity, resilience and game days
local explanation: An explanation of one single prediction: how much each feature pushed this forecast up or down.In Explaining and choosing models
Local privilege escalation: Gaining higher rights on a single machine than your account was given, for example going from an ordinary user to administrator or SYSTEM on that host.In Privilege escalation and lateral movement
local rationality: The idea that people act in ways that seem reasonable from where they stand, given what they know and the goals they are juggling.In Postmortems and learning
locale: The regional settings of a computer or spreadsheet: language, date order, decimal mark and thousands separator.In Cleaning data in a spreadsheet
locator: A Playwright object that describes how to find an element. It finds the element again every time you use it, so it always sees the current page.In End-to-end tests with Playwright
lock file: A file that lists the exact version of every package, including the packages your packages need, so every install gets the same set.In From notebook to pipeline, Experiments, versions and the model registry
lock task mode: An Android setting for dedicated devices that runs only allowlisted apps in a kiosk-like way, hiding the home screen and notifications.In Designing for frontline work
Lock timeout: How long a statement waits for a lock before giving up, which stops a migration from queueing everyone behind it. Safe migrations
lock_timeout: A Postgres setting: the longest a statement will wait to get a lock. If it cannot get the lock in time, it fails with an error instead of waiting forever.In Changing the schema without downtime
lock-in: The cost and difficulty of moving away from a vendor or system once you depend on it.In Build, buy or partner, Build, buy or outsource
lockfile: A file that records the exact version of every installed package, so every machine installs the same set.In Continuous integration that people trust
log: An append-only list of events kept for a set time (days or weeks). Each reader tracks its own position and can re-read old events. Kafka and Redis Streams are logs.In Queues, background jobs and pub/sub
log scale: An axis where each step multiplies the value by the same factor, so 5, 10, 20 and 40 are evenly spaced.In Describing data honestly
logical backup: A backup that saves the database as SQL statements or table data, made with pg_dump. It can be restored on another Postgres version or into another database.In Backups and disaster recovery
Logical replication: Replication that sends row changes (insert, update, delete) for chosen tables, so the receiving database can be a normal, writable database with its own schema extras.In Replication and read replicas, Upgrades, maintenance and on-call
logical shards: One slice of the data, small enough to move between servers. Many logical shards live on each physical database server.In Facebook: the social graph, TAO and memcache at planet scale
Long polling: The client sends a request and the server holds it open until there is news or a timeout (often 20 to 30 seconds), then the client immediately asks again.In Real-time updates: polling, WebSockets, SSE and push
long task: A piece of work on the browser's main thread that runs longer than 50 milliseconds. While it runs, the page cannot respond to input.In Accessible, fast pages
long-term containment: Durable changes that keep the attacker out while you rebuild, such as resetting credentials and blocking their route back in.In Incident response
longest prefix match: When several routes match a destination, the router uses the most specific one, the one with the biggest number after the slash.In Networks from the till to the data centre
longest-first greedy: A scheduling rule: sort the jobs from longest to shortest, then give each one to the machine with the least work so far.In Stable suites in CI
lookback window: A margin of extra time an incremental model reprocesses on every run, so rows that arrive late or are corrected are picked up.In Transforming data in the warehouse
loose object: An object stored as its own zlib-compressed file under .git/objects, before Git packs it.In Inside the .git folder: objects, refs and HEAD
LoRA: Low-Rank Adaptation: a fine-tuning method that freezes the original model and trains small added matrices, so only a small fraction of parameters change and adapters can be swapped per task.In Prompting, retrieval or fine-tuning?
loss function: The number a model tries to make small during training, computed from its errors on the training rows, such as the average squared error.In Training and comparing models
lost update: When two transactions read the same value, each computes a new value from it, and the second write silently replaces the first.In Data with PostgreSQL and Prisma
low fidelity: A rough version of a design with grey boxes and real words, without colours, images or polish, made to test structure and flow quickly.In Screens and flows
LPT: Longest Processing Time first: a greedy scheduling rule that hands out the biggest jobs first, each to the least busy worker.In Continuous integration that people trust
LRU: Least recently used: when full, remove the item that has not been read for the longest time.In Caching, Performance and memory in JavaScript
LRU cache: A cache with a size limit that removes the least recently used entry when full. Performance and memory in JavaScript
LSN: Log sequence number: the byte position of a record in the WAL, shown like 0/3DDD4A0. A bigger LSN means a later change.In Replication and read replicas
LTS: Long-Term Support: a Node release line that gets bug and security fixes for about 30 months from its first release. Up to Node 26, even numbers become LTS and odd numbers are short-lived; from Node 27 in 2027, every yearly release becomes LTS.In Node.js and npm
LTV: Lifetime value: the total contribution margin you expect from a customer over the whole time they keep buying.In Unit economics: does each order make money?, Unit economics
Lua script: A small program written in the Lua language that Redis runs on the server. While it runs, no other command executes, so the whole script is atomic.In Rate limiting and protecting your system

M

MAC address: Media Access Control address: a hardware ID burned into each network card, written like 3c:52:82:1a:9f:07.In Networks from the till to the data centre
MADR: Markdown Architectural Decision Records: a widely used ADR template with sections for the problem, decision drivers, considered options, the outcome, its consequences and how it will be confirmed.In Decisions, diagrams and reviews
MAE: Mean absolute error: the average size of the misses, in units.In Forecasting demand
Maestro: An open-source tool that drives Android and iOS apps on an emulator, simulator or device using test flows written in YAML.In Testing what matters
magic link: A sign-in method where the user types their email and gets a one-time link; clicking it signs them in, so there is no password to forget or steal.In Data and security basics
major incident: An incident with such high impact that it gets a special process: one coordinator, a shared call, and regular updates to everyone affected.In Incidents, problems and changes
major version: A PostgreSQL release with new features, such as 16, 17 or 18. Its internal catalog format changes, so moving to it needs an upgrade procedure.In Upgrades, maintenance and on-call
Managing up: Keeping your own manager informed and aligned so they can help you and your team. Managing up
manifest: A small JSON document that lists an image's config and layers by their hashes. Its own SHA-256 hash is the image digest.In Containers from the inside
Map: A built-in key-value collection whose keys can be any type, that remembers insertion order, and that has a size property and get, set, has and delete methods.In Working with data: arrays, Maps, JSON, dates and money
MAPE: Mean absolute percentage error: the average of each row's miss divided by its actual value.In Forecasting demand
mapping table: A two-column table that maps every raw label you found to one standard label.In Cleaning data in a spreadsheet
margin: A profit line divided by revenue, written as a percentage. It shows how much of each dollar of sales is left at that level.In The three statements
marginal cost: The extra cost of producing or serving one more unit, holding everything else the same.In Unit economics
Markdown: A simple way to format plain text: a line starting with # is a heading, a line starting with - is a list item. AI tools read and write it easily.In A spec the AI can follow
marketplace: A platform that matches two or more groups, such as buyers and sellers, and usually takes a share of each transaction.In Marketplaces, platforms and moats
MASE: Mean absolute scaled error: the MAE divided by the MAE of a naive forecast on the training data. Below 1 means better than naive.In Forecasting demand
masking: When outliers inflate the mean and standard deviation so much that a z-score rule no longer flags them.In Describing data honestly
Master data: The core reference data a business runs on, such as stores, products, suppliers, customers and prices, which changes slowly and is used by many systems and transactions.In Platforms, integrations and systems of record
master table: A reference table that holds the one agreed description of each store, product or customer, such as the site table with each store's city.In Cleaning data in a spreadsheet
materialisation: How dbt stores a model in the warehouse: as a view, a table rebuilt each run, an incremental table, or not at all (ephemeral).In Transforming data in the warehouse
materialized view: A query whose result is saved as a table and refreshed on demand, so reading it is fast even if computing it is slow.In Beyond SQL: choosing the right data store
matrix: A strategy that runs the same job several times with different values, such as Node 22 and Node 24, each as its own job.In GitHub Actions: continuous integration, Contract tests
max output tokens: A request parameter that caps how many tokens the model may write in its answer. The field name differs by provider.In How LLM APIs work
max_slot_wal_keep_size: A primary setting (PostgreSQL 13 and later) that limits how much WAL replication slots may hold back. A slot that needs more is invalidated.In Replication and failover
MCP (Model Context Protocol): An open standard for exposing tools, resources and prompts to AI apps through a server. Tools and MCP
mcrouter: Facebook's open-source proxy that routes memcached requests to the right server and pool, with failover.In Facebook: the social graph, TAO and memcache at planet scale
MDM: Mobile device management: software that lets IT configure, update, lock and wipe phones and handheld devices remotely.In Looking after 1,400 devices
mean: The sum of all values divided by how many there are. Also called the average.In Describing data honestly
measurable objectives: A goal with a metric, a baseline, a target and a date, so everyone can tell later whether it was reached.In What business analysts do
Measure: A calculation evaluated for whatever filters are in place, such as net sales. Agreed metrics
MECE: Mutually exclusive, collectively exhaustive: a set of items with no overlaps and no gaps.In Executive updates and presentations, Asking a question data can answer, Capability maps
median: The middle value when the values are sorted. Half the values are below it and half above.In Describing data honestly, Team design and delivery health
Memoization: Reusing a previous result when inputs have not changed (memo, useMemo, useCallback). Accessible and fast
Memory leak: Memory that stays in use because something still references objects the program no longer needs. Performance and memory in JavaScript
mentoring plan: A one-page note per engineer: where they are, where they want to go, one or two focus areas, the work that will grow them and how you will both know.In Delegation and mentoring
merge base: The newest commit that two branches both contain. Git compares each branch against it to see what each side changed. git merge-base prints it.In Branches and merging, Pull requests and code review
merge commit: A commit with two parents, which joins two lines of history. Its snapshot contains the work of both branches.In Branches and merging
Merge conflict: When two branches changed the same lines and Git needs a person to decide the result. Merge conflicts without panic
merge queue: A GitHub feature that lines up approved pull requests and tests each one on top of main plus the PRs ahead of it before merging.In Continuous integration that people trust
Message Batches API: An API for sending many model requests at once to be processed asynchronously, at a lower price, with results collected later.In Claude: how Anthropic serves Claude at scale
metadata: Small key-value facts stored with an object, such as Content-Type, Cache-Control and size.In Files, object storage and CDNs
method: An action that belongs to an object, written after a dot, for example df.drop_duplicates() on a table.In The same analysis in Python
methodology: A documented, repeatable process for carrying out and reporting a security test, so results are consistent and defensible.In Ethics, law and scoping
metric: A number measured over time, stored as a time series, such as requests per second or memory used.In Observability: logs, metrics and traces, Asking a question data can answer
Metric definition: A written rule for a number: what counts, what is excluded, which time zone, who owns it. Agreed metrics
metric dictionary: A shared catalogue with one entry per metric: its name, plain definition, formula, grain, filters, owner, source and caveats.In Metrics everyone agrees on
metric tree: A diagram that breaks a top metric into the smaller metrics that produce it, down to inputs that individual teams can move.In North star, input and guardrail metrics
Metro: The JavaScript bundler React Native uses. It packs your code into one bundle and serves it to the phone while you develop.In Building the app, How React Native works
MFA: Multi-factor authentication: a second proof besides the password. Accounts and MFA
Microservices: An architecture where each business capability runs as its own separately deployed service with its own database, talking to others over the network.In Monolith, modular monolith, microservices
Microtask: A high-priority callback, such as a promise reaction, that runs before the event loop picks the next task. The event loop
microtask queue: A higher-priority queue for promise callbacks (then, catch, finally, the rest of an async function after await) and queueMicrotask. It is fully emptied after each task, before anything else.In The event loop
middleware: A function that runs on every request before your route handler, for example to parse JSON, log the request or check a header. It can pass the request on or answer it.In From script to service: a small API in Node, Your first NestJS service
migration: A numbered SQL file that changes the database structure one step, kept in Git and applied in order to every environment.In Data with PostgreSQL and Prisma, Continuous delivery, Platforms, integrations and systems of record
minimum detectable effect: The smallest true effect a test is designed to detect reliably; smaller effects may be missed. Choose it as the smallest effect worth acting on.In Experiments: A/B tests, holdouts and store pilots, Designing and reading an A/B test
minor version: A bug-fix and security release within one major version, such as 16.4 to 16.6. It never changes the on-disk format.In Upgrades, maintenance and on-call
misfit: An application used for a job it was not designed for, such as a chat app used as a customer database.In Application portfolio
mitigation: An action that reduces or stops the customer impact without necessarily fixing the underlying cause, such as rolling back or turning off a feature.In Leading an incident
MITRE ATT&CK: A free public knowledge base, run by the non-profit MITRE, that describes how real attackers behave, grouped into tactics (goals) and techniques (methods), each with an ID.In How breaches really happen, Inside a SOC, Writing detections, Threat hunting, Incident response
mix-adjusted rate: A rate recalculated with the group shares held fixed at one period's mix, so a change in the mix cannot move it.In Traps that fool smart people
Mnesia: A distributed database built into Erlang. Tables can live in RAM, on disk, or both, and be copied across nodes.In WhatsApp: billions of messages with a tiny team
moat: A lasting advantage that lets a business keep better results than rivals even when they try to copy it.In Marketplaces, platforms and moats
mobile wallet: A service that stores money as a balance linked to a phone number, so people can pay and send money from a phone without a bank account.In Teardown: sending money with a mobile wallet
mock: A fake version of a function or service used in a test. It records how it was called and returns whatever you tell it to.In Errors, debugging and testing
mock server: A small fake HTTP server started by the test that answers only the requests you described, so you can run the app's real API code without the real service.In Contract tests
Mock, stub, fake: Stand-ins for real dependencies: a stub returns fixed answers, a mock also checks how it was called, a fake is a simple working version. Unit and API tests
modal: A screen for one short, self-contained task that appears on top of the current screen and has a clear way out, such as Done or Cancel.In Navigation and app structure
model: A single SQL file containing one SELECT statement. The tool turns it into a view or table in the warehouse.In Transforming data in the warehouse
model card: A short document that travels with a trained model and says what it is for, what it is not for, how well it works for which groups, its limits and who owns it.In Responsible ML in a business
model choice memo: A one-page record of which model was chosen, why, what it costs, how it fails, the fallback and who owns it.In Explaining and choosing models
Model Context Protocol: An open protocol, built on JSON-RPC, for connecting AI applications to tools and data through a standard server interface.In Tools and the Model Context Protocol
model registry: A catalogue of models that are candidates for production: each named model has numbered versions, each linked to the run that made it, with aliases and tags.In Experiments, versions and the model registry
modular monolith: One deployable application whose code is split into modules with clear public interfaces, where each module owns its own data.In Monolith, modular monolith, microservices
module: A file whose top-level names are private to it, except the ones it explicitly exports. Other files get them by importing.In Prototypes, classes and modules, Infrastructure as code
Monitoring: Watching a fixed set of known numbers and checks, such as error rate and disk space, and alerting when they cross a line.In Observability: logs, metrics and traces
monolith: One application, built and deployed as a single unit, usually with one database.In Monolith, modular monolith, microservices
Monthly business review: A monthly meeting where each director reviews the numbers, risks and progress of their area against the plan with their leader.In Organisation and operating model
MoSCoW: Prioritising as Must have, Should have, Could have, Won't have this time. Requirements
MTBF: Mean time between failures: the average time a system runs before something breaks.In Measuring speed and reliability: p99, SLOs and nines
MTTR: Mean time to recovery: the average time from a failure starting to the service working again. It includes noticing, finding the cause and fixing it.In Measuring speed and reliability: p99, SLOs and nines
muda: The Japanese word for waste: activity that uses resources but creates no value for the customer.In Modelling processes
Multi-cursor: Several cursors in the editor at once, so what you type goes into many places at the same time.In VS Code essentials
Multi-factor authentication: Signing in with two or more different kinds of proof, such as something you know (a password) and something you have (a phone or security key).In Identity: passwords, MFA and passkeys
Multi-factor authentication (MFA): Multi-factor authentication: logging in with two or more kinds of proof, such as a password plus a phone or a security key.In Accounts, SSO and MFA
multi-stage build: A Dockerfile with several FROM stages. Early stages install tools and compile; the final stage copies only the results, so the shipped image stays small.In Shipping safely: containers, CI/CD and releases, Containers from the inside
Multiple regression: A regression with two or more x variables; each coefficient is the change in y for one unit of that x with the other x variables held fixed.In Regression and correlation
musl: A small C standard library used by Alpine Linux instead of glibc. Programs compiled against glibc usually do not run on musl without rebuilding.In Containers from the inside
must-pass category: A category where any drop against the current version blocks the change, because a wrong answer there can hurt someone.In Evals that never stop
mutant: A copy of the code with one small deliberate bug, such as greater-than in place of greater-than-or-equal, used to check whether the tests notice.In Designing tests that find bugs
Mutation: Changing the inside of an existing object or array, such as setting a property or pushing an item. Everyone holding a reference sees the change.In Objects, arrays and references
mutation check: Changing the code on purpose, in a small way, to see whether any test notices. A change that no test catches shows a gap in the tests.In Stable suites in CI
mutation testing: Checking a test suite by making small deliberate bugs (mutants) in the code and seeing whether the tests fail.In Unit and API tests
MVCC: Multi-version concurrency control: an UPDATE writes a new version of the row and keeps the old one, so each transaction can see the version that was current when it started.In Running Postgres in production, ChatGPT: serving a large language model to hundreds of millions, Inside a PostgreSQL server
MVP: Minimum viable product: the smallest thing that tests whether the idea delivers value. Cheap tests
MyRocks: Facebook's MySQL storage engine built on RocksDB, a log-structured store that writes sequentially and compresses well.In Facebook: the social graph, TAO and memcache at planet scale

N

N+1: A pattern where code runs one query for a list and then one more query for each item in the list, so the number of round trips grows with the list.In Data with PostgreSQL and Prisma, Capacity, resilience and game days
N+1 query: One query for a list, then one more query per item, which turns a single page into hundreds of round trips. Query plans
namespaces: A Linux kernel feature that gives a process its own private view of one kind of resource, such as process IDs, network interfaces or mounted files.In Containers from the inside, Kubernetes for application teams
NaN: Not a Number: a special number value that comes out of a failed calculation, such as Number('2 pcs'). It is not equal to anything, not even itself.In JavaScript in one sitting
narrowing: TypeScript making a type more specific inside a branch, after a check such as typeof, === undefined, in, or a switch on a tag field.In TypeScript for a JavaScript team
NAT: Network Address Translation: the router replaces the private source address of outgoing traffic with its own public address, and reverses this for replies.In Networking you actually need
natural keys: A key taken from the real world, such as a store code, a barcode or a phone number.In Modeling data in tables
Near misses: An event that could have caused an incident but did not, often because of luck or a last-minute catch.In Postmortems and learning
nearest-rank: A way to compute a percentile: sort the values, take rank = ceil(p/100 × n), and return the value at that rank. The answer is always a value that was really measured.In Performance and load testing
negative caching: Storing the fact that something does not exist, for a short time, so repeated lookups for it do not reach the database.In Case study: a URL shortener
Net revenue retention: This period's revenue from a group of existing customers divided by the same group's revenue a year earlier; above 100% means growth from existing customers outweighs losses.In Retention and cohorts
Net sales: The value of completed sales at the price customers paid, after discounts and returns, excluding VAT or sales tax.In Metrics everyone agrees on
network effect: When each additional user makes a product more valuable to other users, directly or through another group of users.In Marketplaces, platforms and moats
network partition: A network partition: some machines can no longer reach other machines, while each side keeps running and keeps getting requests.In CAP, PACELC and consistency
Network segmentation: Splitting one network into smaller zones, so devices in one zone can only reach other zones through a firewall that allows named traffic.In Networks, endpoints and the cloud, Attacking networks and services
NIST AI Risk Management Framework: A voluntary framework from the US National Institute of Standards and Technology for managing AI risks, organised in four functions: Govern, Map, Measure and Manage.In Leading AI adoption
NIST Cybersecurity Framework 2.0: A framework from the US National Institute of Standards and Technology, version 2.0 published in 2024, that organises cybersecurity outcomes into six functions: Govern, Identify, Protect, Detect, Respond and Recover.In Risk, security and compliance
NIST SP 800-61: A US National Institute of Standards and Technology guide on computer security incident handling, widely used worldwide.In Incident response
nodes: One machine (virtual or physical) in a Kubernetes cluster that runs pods. An agent called the kubelet on each node starts and watches the containers.In Kubernetes for application teams
noise floor: The error you would get even with a perfect forecast of the average, because daily sales vary by chance.In Explaining and choosing models
Non-additive: A measure that can't be summed along any dimension, such as a percentage, an average, a distinct count or a percentile.In Metrics everyone agrees on
non-functional requirement: A statement of how well the system must work, with a number: how fast, how often available, how much data, how secure.In What system design really is, User stories and acceptance criteria, Requirements that work
non-goals: Something a reasonable person could expect the project to do, which you have decided it will not do in this version, with the reason.In From idea to one-page spec
non-inferiority margin: The largest harm you are willing to accept on a guardrail; the guardrail passes only if the confidence interval rules out harm bigger than this.In Designing and reading an A/B test
Normalization: The process of splitting data into tables so that each fact is stored once, which prevents copies that disagree.In Modeling data in tables
north star metric: The single metric that best captures the value customers get from a product, and that moves before the business results it leads to.In North star, input and guardrail metrics
notebook: A document that mixes blocks of code, their results and notes, run one block at a time in a web browser.In The same analysis in Python
notice: How long before the end date one side must tell the other that it wants to end or change the contract.In Build, buy or outsource
notification budget: An agreed cap on how many optional messages, such as offers and reminders, one person receives in a period, shared by every team that wants to send them.In Engagement loops, habits and notifications without dark patterns
notification channel: A named category of notifications on Android 8 and later, with its own importance, sound and vibration, that the user can mute separately.In Device features
novelty effect: A short-lived lift because something is new and people try it, which fades as they get used to it.In Designing and reading an A/B test
npm ci: Installs exactly what the lockfile says, from scratch, and fails if package.json and the lockfile disagree; use it in CI. Node.js and npm
NPV: Net present value: the sum of all future cash flows, each discounted back to today at the company's required rate of return, minus the investment. Positive means the investment beats that rate.In Business cases a managing director will sign, Business cases
NPV profile: A chart of a project's NPV at every discount rate. It crosses zero at the IRR.In Business cases
NTP: Network Time Protocol, which keeps device clocks right; wrong clocks break certificates and logs. Store networks
NULL: The marker SQL uses for a missing or unknown value. It is not zero and not an empty string.In SQL you'll actually use
null hypothesis: The default claim that nothing changed: here, that both groups have the same true conversion rate. A test asks whether the data are surprising under this claim.In Hypothesis tests without the fog
null prototype: An object created without the usual Object.prototype, so it has no inherited methods such as hasOwnProperty or toString.In Working with data: arrays, Maps, JSON, dates and money
nullable: Allowed to hold null, which means there is deliberately no value.In Prompts and structured output
nullish coalescing: The ?? operator: returns the right side only when the left side is null or undefined, so 0 and empty strings are kept.In JavaScript in one sitting
NXDOMAIN: Non-existent domain: the DNS reply that says this name does not exist at all.In Networks from the till to the data centre

O

OAuth 2.0: An industry standard that lets an app get a token to act for a user, issued by a separate login server, without the app ever seeing the user's password.In Security essentials
object database: The part of .git that stores every file version, folder listing and commit as an object, each named by a hash of its own content.In Inside the .git folder: objects, refs and HEAD
Object storage: A storage service that keeps whole files (objects) under a name (key) in a container (bucket), read and written over HTTP. Examples: Amazon S3, Cloudflare R2, MinIO.In Files, object storage and CDNs
object-level access control: A check that the caller may act on the specific record requested, usually by comparing its owner or tenant with the caller.In Broken access control
objects: One stored file plus its metadata, addressed by a key inside a bucket.In Files, object storage and CDNs, Facebook: the social graph, TAO and memcache at planet scale
Observability: How well you can understand what is happening inside a running system from the data it sends out, including questions nobody planned for.In Observability: logs, metrics and traces
observation: One recorded fact about one thing at one time, such as one store's fresh produce sales on one day.In Cleaning data in a spreadsheet
observation window: The period before the as-of date that features are computed from, here 60 days.In Classification and churn
OCAs: Open Connect Appliance: a Netflix server full of disks that stores video files and sends them to nearby viewers over plain HTTP.In Netflix: streaming to every screen
octet: A group of 8 bits. In an IPv4 address each of the four numbers is one octet, so it can only be 0 to 255.In Networks from the till to the data centre
off-peak fill window: A daily period of low traffic, such as the middle of the night, when Open Connect servers download new and updated video files.In Netflix: streaming to every screen
offline evaluation: Testing a model on historical data, without changing anything in the real business.In Framing a problem as a model
offline store: The store of feature history, usually warehouse tables, used to build training sets and run batch predictions.In Data and feature pipelines
offline-first: An app design where the app reads and writes a local database on the device first, and syncs with the server in the background whenever a connection is available.In Offline-first apps and data sync, Offline first
offset: In a log like Kafka, the position of the next event a consumer group will read.In Queues, background jobs and pub/sub
Offset pagination: Paging by position: skip the first N rows, then return the next page. ?page=3 usually becomes OFFSET 40 LIMIT 20.In APIs done right
OIDC: OpenID Connect: a layer on top of OAuth 2.0 that adds a standard way to learn who the user is, through an ID token.In Security essentials, Deploying and securing with GitHub Actions, Who are you and what may you do?, Supply chain and secrets
OIDC (in Actions): A way for a workflow to get a short-lived cloud credential by proving who it is, so no long-lived secret is stored. Deploying and securing with GitHub Actions
OIDC (in CI): CI proves its identity to the cloud with a short-lived token, so no long-lived keys are stored. Supply chain
OKRs: Objectives and key results: a short goal-setting format with one qualitative objective and a few measurable results that show it was reached.In Vision, strategy and why now
OLA: Operational level agreement: an internal promise between the desk and another IT team, such as the network team answering escalations within 30 minutes.In Incidents, problems and changes
OLAP: Online analytical processing: few, large queries that scan and aggregate millions of rows, like sales by category for the last year.In Beyond SQL: choosing the right data store
old generation: The large area of the heap for objects that have survived several young collections. It is collected less often, with mark-and-compact.In Performance and memory in JavaScript
OLTP: Online transaction processing: many small, fast reads and writes of a few rows each, like recording a sale or reserving stock.In Beyond SQL: choosing the right data store
omitted variable bias: The bias in a coefficient when a variable that affects y and is related to x is left out of the model.In Regression and correlation
on-demand: Paying the list price per hour or second with no promise of future use. Flexible, and the most expensive rate.In Cloud and SaaS costs with FinOps
Onboarding: The path a new user takes from first contact with a product to getting its core value for the first time, and then coming back.In Onboarding and the first-time experience
one-hot encoding: Turning a category column into one 0 or 1 column per category, so a model does not treat category codes as amounts.In Training and comparing models
one-way door: A decision that is expensive or impossible to reverse, such as a public API, a core data model or a data store. It deserves slow, careful thought.In Running a design review, Writing for decisions
Online evals: Grading a sample of real production traffic, usually every day, to see how the live system is doing.In Evals that never stop
online store: A fast key-value store that holds only the latest feature values per entity, for low-latency lookups during serving.In Data and feature pipelines
OODA loop: Observe, orient, decide, act: a fast decision loop. The defender who cycles through it faster than the attacker tends to win.In Inside a SOC
OOM killer: The Linux kernel feature that kills a process to free memory when the machine runs out of it.In Inside a PostgreSQL server
opaque: A value the client stores and sends back but must not read or build. The server can change what is inside without breaking clients.In Designing an API people can use
opaque token: A random token with no meaning of its own; the server looks it up to find the login. A server session in a header instead of a cookie.In Who are you and what may you do?
Open Connect: Netflix's own content delivery network, launched in 2011 and announced to the public in 2012. It serves all Netflix video from servers placed inside or near internet providers.In Netflix: streaming to every screen
open model: A load test where new requests start at a fixed rate whatever the response time, like real customers arriving, so a slow system builds a queue instead of getting less load.In Capacity, resilience and game days, Performance and load testing
open redirect: A service that forwards visitors to any address given to it. Attackers use it to make a phishing link look like it comes from a trusted domain.In Case study: a URL shortener
open-weight model: A model whose trained weights are published, so you can download it and run it on hardware you control.In How AI features work, just enough for PMs
OpenAPI: A standard file format (YAML or JSON) that describes every endpoint, parameter, request body and response of an HTTP API.In APIs done right, Testing and documenting the API
OpenID Connect: An identity layer on OAuth 2.0 that adds an ID token saying who the user is. Identity
OpenID Connect (OIDC): OIDC: a newer SSO standard built on OAuth 2.0; the IdP sends a signed ID token in JSON.In Accounts, SSO and MFA
OpenTelemetry: The open standard and libraries for producing traces, metrics and logs. OpenTelemetry
OpenTelemetry Collector: A separate OpenTelemetry program that receives telemetry, processes it (filter, sample, batch, enrich) and exports it to one or more backends.In Observability with OpenTelemetry
operating agreement: A one-page agreement between you and your manager about goals, how news travels and which decisions you make alone.In Managing up and across
operating and capital spending: Operating expenditure (opex) is spending that counts as a cost in the year it happens, like salaries and cloud bills. Capital expenditure (capex) buys an asset that lasts several years and is spread over them in the accounts.In Money: budgets, headcount and vendors
operating cadence: The fixed calendar of planning and review meetings, and the documents that feed them, through which an organisation sets goals, tracks progress and makes decisions.In Organisation and operating model
Operating model: How a department is organised, plans, decides and measures itself. Operating model
Operational: A dashboard for people running daily work, such as a store manager, that shows what needs action now or today.In Designing a dashboard for decisions
opex: Operating expenditure: day-to-day running costs such as licences, support and subscriptions, charged in the year they happen.In Business cases a managing director will sign, The three statements, Budgets, capex and opex
opportunity: An unmet customer need, pain point or desire that, if addressed, would move your outcome.In From research to opportunities
opportunity cost: The value of the best alternative you give up when you choose one option, such as other work the same engineers could have done.In Business cases
opportunity solution tree: A diagram with an outcome at the root, opportunities below it, solutions below a chosen opportunity, and assumption tests at the bottom.In From research to opportunities
optimistic update: Showing the expected result of a change on screen before the server confirms it, and undoing it if the server refuses.In Forms that people can trust
Optional chaining: The ?. operator: reads a property only if the thing before it is not null or undefined; otherwise the whole expression gives undefined instead of throwing an error.In JavaScript in one sitting
oracle: Any way of deciding whether a result is right or wrong, such as a spec, an older version, a comparable product or a user's reasonable expectation.In Exploratory testing
orchestration: A saga where one coordinator, the orchestrator, keeps the state of each process and tells each service what to do next.In Events, the outbox pattern and sagas
orchestrator: A tool that runs data jobs in the right order, starts each job when the jobs it depends on have succeeded, retries failures and keeps a record of every run.In Fresh, tested, trusted, From notebook to pipeline
Orchestrator–workers: A pattern where one model call splits a job into sub-tasks at run time, other calls do each sub-task, and the results are combined.In What makes an agent
origin: The real source a CDN fetches from on a miss, such as your S3 bucket or your web server.In Files, object storage and CDNs, GitHub: remotes, clone, push, pull and authentication, JavaScript in the browser, Identity: passwords, MFA and passkeys
ORM: Object-relational mapper: a library that turns rows in database tables into objects in your programming language, and turns method calls into SQL.In Data with PostgreSQL and Prisma
orphaned volumes: A storage volume that is no longer attached to any running machine but is still billed every month.In Cloud and SaaS costs with FinOps
OSINT: Open-source intelligence: information gathered from public sources. For people it means names, roles and public posts, used only when the engagement authorises social engineering.In Reconnaissance
OTA updates: Over-the-air update: replacing an app's JavaScript bundle and assets without a new store release, as with Expo Updates in React Native.In Shipping safely: containers, CI/CD and releases, Continuous delivery
OTLP: The OpenTelemetry Protocol: the standard format and transport for sending traces, metrics and logs.In Observability for LLM systems, Observability with OpenTelemetry
ours: In a merge, your side: the branch you are on when you run git merge. Git labels it HEAD or current change.In Merge conflicts without panic
out-of-band channel: A communication channel the attacker cannot see, used when normal email or chat might be compromised.In Incident response
outbox: A table on the device that lists actions the server has not confirmed yet, in the order they happened. A sync loop sends them and marks each one done.In Offline first
Outbox pattern: Writing an event to a table in the same transaction as the data, then publishing it. Outbox and sagas
outcome: A change in what people do, or in a business result, that happens because of what the team made.In What a product manager actually does, From research to opportunities, OKRs and outcome goals
outcome checks: A check on what the world looks like when the agent has finished: which records exist, what changed and what the final answer cites.In Evaluating agents
outcome roadmap: A roadmap organised around problems to solve and the metrics that show they are solved, with features listed as current bets rather than promises.In Outcome roadmaps
Outcome vs output: An outcome is a change in behaviour or results; an output is a thing shipped. Problems first
outlier: A value far from the rest of the data. It can be an error or a real but unusual case.In Describing data honestly
output: Something the team produces, such as a feature, a release or a document.In What a product manager actually does, OKRs and outcome goals
output encoding: Converting special characters into a safe form for the exact place the text will appear, such as HTML text, an attribute or a URL.In Injection and unsafe input
over-the-air (OTA) update: Replacing an installed app's JavaScript bundle and assets by download, without a new store build and without store review.In Testing and releasing
Overfitting: When a model learns noise in its training data, so its training error keeps falling while its error on new data rises.In Training and comparing models
Overhead: Administrative work not tied directly to running a service, such as meetings, planning, hiring, training and peer reviews.In Toil and automation
Overlap: Repeating the end of one chunk at the start of the next, so a sentence or rule cut at a boundary still appears whole in one chunk.In Retrieval-augmented generation
overlay file system: A Linux file system that stacks several directories and shows them as one merged tree. Upper directories hide files of the same name in lower ones.In Containers from the inside
override: A person changing a model's suggestion before it is used, here a store manager changing a suggested order quantity.In Responsible ML in a business
OWASP Top 10 for LLM Applications: A list of the ten most important security risks for applications built on large language models, published and updated by the OWASP GenAI Security Project. The 2025 edition starts with prompt injection.In Guardrails and red-teaming
OWASP Web Security Testing Guide: The OWASP Web Security Testing Guide: a free, structured catalogue of what to test in a web application and how, organised into categories such as authentication, authorisation and business logic.In Attacking web applications
own properties: A property stored directly on an object, as opposed to one it inherits from its prototype. Object.hasOwn(obj, 'x') checks for one.In Prototypes, classes and modules

P

p-value: If the change had no effect at all, the chance of seeing a difference at least this large by luck alone.In Experiments: A/B tests, holdouts and store pilots, Hypothesis tests without the fog
P&L: Profit and loss statement, also called the income statement or statement of profit or loss: revenue and costs over a period, ending in profit.In The three statements, Budgets, capex and opex
p95: The 95th percentile: 95 out of 100 requests are faster than this value. The last lesson of this course explains percentiles.In What system design really is, Observability: logs, metrics and traces, How AI features work, just enough for PMs, Shipping an agent safely, Cost and latency at scale, Serving: batch and online
p95 latency: The response time that 95 out of 100 requests are faster than; the usual load-test target. Load testing
p99: The latency that 99 out of 100 requests beat. The slow tail that users remember.In Google Search: crawling, indexing and answering in a fraction of a second, Serving: batch and online
PaaS: Platform as a service: the provider runs the operating system and runtime, such as a managed database; you manage your code, settings and data.In Networks, endpoints and the cloud
PACELC: If there is a Partition, choose Availability or Consistency; Else, when the network is fine, choose Latency or Consistency.In CAP, PACELC and consistency
Packaging: Deciding which features, limits and benefits go together into each plan or tier a customer can buy.In Pricing and packaging
packets: One small piece of data travelling across a network, usually up to about 1,500 bytes.In Networking you actually need
packfile: A single file that stores many objects together, with similar objects stored as small deltas against each other. It comes with an .idx index for fast lookup.In Inside the .git folder: objects, refs and HEAD
Pact Broker: A service that stores pact files and verification results for every version of every consumer and provider, and answers whether two versions are compatible.In Contract tests
pact file: A JSON file that Pact writes after a passing consumer test. It lists every interaction: the provider state, the request and the expected response with its matching rules.In Contract tests
page cache: Memory the operating system uses to keep recently read file blocks, so the next read does not need the disk.In Inside a PostgreSQL server
page object: A class that wraps one screen of the app and offers methods named after user intent, such as payWith or placeOrder, so tests do not repeat locators.In End-to-end tests with Playwright
pager: A program that shows a long file one screen at a time, so you can scroll and search without loading it all in an editor.In Linux from the command line
pages: A fixed-size block of 8 KB. Postgres reads and writes tables and indexes one page at a time, never one row at a time.In Indexes from the inside, Alerting that does not burn people out, Inside a PostgreSQL server
paint: The browser step that fills in pixels: text, colours, borders and images.In JavaScript in the browser
pair programming: Two engineers working on the same code at the same time, one typing and one reviewing and thinking ahead, swapping roles often.In Delegation and mentoring
Pairwise grading: Showing the judge two answers to the same question and asking which is better, or whether they tie.In Evals that never stop
Pairwise testing: A way to choose test combinations so that every pair of values of any two parameters appears in at least one test, using far fewer tests than all combinations.In Designing tests that find bugs
pandas: A free Python library for tables of data: reading files, cleaning, joining, grouping and summarising.In The same analysis in Python
paper prototype: A set of paper screens, one card per screen or state, that a person taps with a finger while someone on the team plays the computer and puts down the next card.In Screens and flows
parallel gateway: A BPMN diamond marked with a plus sign that starts all outgoing paths at the same time, or waits for all incoming paths to finish.In Modelling processes
parallel run: Running the old and new implementations on the same real inputs, using the old answer, and comparing the two results to find differences before the new one takes over.In Evolving a legacy system, Build, buy or outsource
parameterised query: A query where the SQL text has numbered slots such as $1 and the values are sent separately, so the database never parses a value as SQL.In Injection and unsafe input
parameters: Values a model learns from the training data, such as the weights of a linear model or the split points of a tree.In Training and comparing models
parking lot: A visible list where the group parks topics that are real but off-track, each with an owner, so the meeting can move on without losing them.In Running a design review
partial dependence: A plot of the model's average prediction as one feature changes while every other feature keeps its real values.In Explaining and choosing models
Partial index: An index with a WHERE clause, which stores only the rows that match it. Small and cheap when the queries you care about touch a small subset.In Indexes from the inside
partition: One of the smaller tables that holds part of the rows of a partitioned table, chosen by a rule on one column.In Partitioning big tables, Queues, background jobs and pub/sub, From notebook to pipeline, Serving: batch and online
partition key: The column (or columns) whose value decides which partition a row goes into, for example sale_date.In Partitioning big tables
partition pruning: The planner skipping partitions whose value range cannot match the query's filter, so they are never read.In Partitioning big tables
partitioned: Split into smaller child tables by a key such as month, so old months can be dropped in one quick step.In Case study: a URL shortener
Partitioning: Splitting one big table into child tables by a key such as month, on one server; queries with the key scan only the right parts. Partitioning
pass@1: The share of all runs that passed: the chance that one run of a typical task succeeds. With 36 tasks and 5 runs each, passed runs divided by 180.In Evaluating agents
Passive reconnaissance: Reconnaissance that uses only public or third-party sources and sends no traffic to the target's own systems, so the target cannot see it.In Reconnaissance
Passkeys: A login credential based on a key pair: the private key stays on your phone, laptop or security key, and the site keeps only the public key.In Accounts, SSO and MFA, Identity: passwords, MFA and passkeys, Authentication, sessions and tokens
password hashing function: A hash function built to be slow and tunable on purpose, such as Argon2id, scrypt or bcrypt, so each guess costs an attacker real time and memory.In Authentication, sessions and tokens
password manager: An app that creates a long random password for every site, stores them encrypted, and fills them in for you.In Identity: passwords, MFA and passkeys
Password spraying: Trying one or two very common passwords against many accounts, so each account sees only a failure or two.In How breaches really happen
patch policy: A written rule for how quickly updates must be installed after release, usually by severity.In Looking after 1,400 devices
Patch ring: A group that gets updates at a set stage, so problems show up on a few devices first. Device fleet
PATH: The list of folders the shell searches, in order, when you type a program's name; the first match runs. The terminal without fear
paved road: A supported, pre-approved way to build and run a common kind of service, with templates and tooling, that teams can use without asking.In Frameworks and governance
Payback: The point in time when the cumulative cash from an investment is back to zero: everything spent has been earned back.In Business cases a managing director will sign, Business cases
payback period: How long an investment takes to earn back its cost; for automation, the build hours divided by the hours saved per week.In Toil and automation, Quality, debt and AI tools
PCI DSS: Payment Card Industry Data Security Standard: the rules every business that takes card payments must follow, checked by yearly assessments.In Networks, endpoints and the cloud, Risk, security and compliance
peak factor: How many times busier the busiest period is than the average. Peak QPS = average QPS × peak factor.In Speed numbers and back-of-the-envelope math, Capacity, resilience and game days
Pearson correlation: A number from -1 to +1 that measures how closely two variables follow a straight line; +1 is a perfect upward line, 0 means no straight-line relationship.In Regression and correlation
Peeking: Checking a running test repeatedly and stopping early on a good result, which inflates false wins. A/B test design
pepper: A secret key kept outside the database, for example in a secrets manager, mixed into every password hash so a database-only leak cannot be cracked.In Authentication, sessions and tokens
percentile: The value below which a given share of measurements fall. p95 = 250 ms means 95% of requests took 250 ms or less.In Measuring speed and reliability: p99, SLOs and nines, Describing data honestly
Percentile (p95, p99): The value that 95 or 99 out of 100 requests are faster than; it shows the slow tail that averages hide. p99 and SLOs
Performance improvement plan: A written, time-bound plan with clear expectations and support for someone below the bar. Performance
performance improvement plan (PIP): A formal written plan with specific goals, support and a deadline, used when informal steps have not fixed a performance problem; often called a PIP.In Performance and growth
performance test: Any test that measures how fast and how stable a system is under a given amount of traffic, as opposed to whether it gives the right answer.In Performance and load testing
periodic snapshot: A fact table with one row per thing per period, such as stock per store per article per day, written whether or not anything happened.In From questions to a reporting model
Permutation importance: How much a fitted model's score gets worse when one feature's values are shuffled, which breaks its link to the target.In Explaining and choosing models
persona: A short description of a typical user, based on research, that helps a team keep real people and their situations in mind.In Users, customers and stakeholders
personal access token: A token you create on GitHub that acts like a password for Git and the API, with an expiry date and limited permissions. Fine-grained tokens can be limited to specific repositories.In GitHub: remotes, clone, push, pull and authentication
pg_hba.conf: The PostgreSQL host-based authentication file. Each line matches a connection type, database, user and client address, and names the login method.In Security and access
pg_repack: An open-source PostgreSQL extension and command-line tool that rebuilds a bloated table online, with only short exclusive locks at the start and end.In Upgrades, maintenance and on-call
pg_rewind: A PostgreSQL tool that rewinds a stopped former primary to the point where its history split from the new primary, copying only changed blocks, so it can rejoin as a replica.In Replication and failover
pg_stat_archiver: A PostgreSQL view with one row of archiver statistics: files archived, failures, and the names and times of the last success and failure.In Backups you have actually restored
pg_stat_statements: A Postgres extension that records statistics for every distinct query shape: how many times it ran, total and average time, rows, and pages read.In Reading query plans and fixing slow queries, Finding and fixing slow queries
pg_upgrade: The PostgreSQL tool that upgrades a data directory to a new major version by rebuilding the system catalogs and reusing the existing data files.In Upgrades, maintenance and on-call
pgaudit: An open-source PostgreSQL extension that writes detailed audit records of statements to the server log, by class or by table.In Security and access
PgBouncer: A lightweight program that sits between applications and PostgreSQL and lets many app connections share a small number of real database connections.In ChatGPT: serving a large language model to hundreds of millions, Finding and fixing slow queries
Phishing: A message that pretends to come from someone you trust so that you click a link, open a file, type a password or send money.In How breaches really happen
phishing-resistant: Phishing-resistant: an authenticator a fake website cannot trick you into using for the real site, because it checks the site's domain itself.In Accounts, SSO and MFA
physical backup: A backup that copies the database's data files themselves, made with pg_basebackup or a tool like pgBackRest. It restores the whole server, fast, on the same major version.In Backups and disaster recovery
pickaxe: git log -S text: finds the commits that added or removed the given text, so you can see when a value first appeared or disappeared.In Everyday Git: status, diff, log and good commits, Reading code like a detective
PII: Personally identifiable information: data that points to a real person, such as a phone number, card number or national ID number.In Designing AI features: LLMs, RAG and vector search
pilot: A small, time-boxed release to real users you chose, with a written plan for what you will learn and when you will decide.In Launch and learn, Launching: pilots, waves and the review after
pilot list: A table of invited shops that the API checks on every request to a pilot feature.In Launch and learn
pinball loss: The loss used to train a quantile forecast: misses below the actual cost q per unit, misses above cost 1 minus q per unit.In Forecasting demand
pipe: The | symbol. It sends the output of the command on its left into the input of the command on its right.In The terminal without fear, From script to service: a small API in Node, Your first NestJS service
Pipeline: A scikit-learn object that chains preprocessing steps and a model, so that fit learns every step from training data only and predict applies them in order.In Features and leakage, From notebook to pipeline, Shell scripting for daily jobs
PITR: Point-in-time recovery: restore a base backup, then replay WAL up to a chosen moment. Restored backups
PITR (point-in-time recovery): Restoring a database to any chosen moment, using a base backup plus archived WAL. Backups
pivot table: A summary table that groups the rows of a data table by the values of one or more fields and adds up, counts or averages another field for each group.In Pivot tables and summaries
pivoting: Following one thing, such as an account, a host or an IP address, from event to event and source to source to build the story of what happened.In Logs that matter, Triage and investigation
PKCE: Proof Key for Code Exchange: the app creates a random secret, sends only its hash at the start, and sends the secret itself when it trades the code for tokens, so a stolen code is useless.In Security essentials
plan-driven: A way of running projects where requirements are gathered and signed off up front, then built, tested and released in stages.In What business analysts do
planner: The part of Postgres that looks at a query, estimates the cost of each possible way to run it, and picks the cheapest. The next lesson shows how to read its choices.In Indexes from the inside, Reading query plans and fixing slow queries
planning fallacy: The habit of making plans and forecasts that are close to the best case, even when similar work took longer before.In Slicing and planning work
planogram: A drawing that shows exactly which shelves and products go where in a store.In Infrastructure as code
plateau: In ArchiMate, a relatively stable state of the architecture that exists for a limited period of time.In Target state and roadmap
platform teams: A team that builds and runs internal services, such as deployment pipelines or shared APIs, that make other teams faster.In Organisation and operating model
pod: The smallest unit Kubernetes schedules: one or more containers that share a network address and run together on one node.In Kubernetes for application teams
point of no return: The moment in a migration after which going back costs more than fixing forward, usually when the old system stops receiving changes.In Evolving a legacy system
point-in-time join: A join that matches each row to the latest value whose time is on or before the row's own time, so no value from the future is used.In Data and feature pipelines
Point-in-time recovery: Restoring a base backup and then replaying archived WAL up to a chosen moment, such as one minute before a mistake.In Backups and disaster recovery
poison message: A job that fails every time it runs because of its own data, so retrying never helps.In Queues, background jobs and pub/sub
policy: A named rule on one table that says which rows a user may read, add, change or delete.In Data and security basics
policy layer: Code that sits between the model and the tools and decides, for each tool call, whether it is allowed, needs a person to approve it, or is denied.In Designing tools and permissions
polling: Asking the server for fresh data on a timer, for example every 30 seconds, instead of the server pushing changes.In Loading server data
pool: The list of backend servers a load balancer can send requests to. Nginx calls it an upstream; AWS calls it a target group.In Load balancers, Modelling processes
pooled rate: The rate you get by adding up both groups: all successes divided by all units. Used in the z-test because the null says both groups share one rate.In Hypothesis tests without the fog
population: Every unit you want a conclusion about, such as all 320 stores or every app order this month.In Samples, uncertainty and confidence intervals
population stability index (PSI): A score of how far a variable's distribution has moved between two samples, added up over buckets. Zero means no change.In Monitoring, drift and retraining
port: A number from 0 to 65535 that picks which program on a machine receives the traffic, for example 443 for HTTPS or 5432 for Postgres.In Networking you actually need, Reconnaissance
position: What someone says they want, such as no new counting before the festival.In Stakeholders, negotiation and leading without authority
Positioning: How you want an audience to understand what a product is, who it is for and why it is better than what they do today.In Launching: pilots, waves and the review after
Postel's law: Be conservative in what you send and liberal in what you accept; a rule from the early internet standards for building systems that keep working together.In Contract tests
posting date: The date an amount is booked in the accounting ledger. It can be later than the day the sale happened.In Metrics everyone agrees on
posting list: The list of document ids that contain one word, usually sorted, often with counts and positions.In Google Search: crawling, indexing and answering in a fraction of a second
postmaster: The parent PostgreSQL process. It listens for connections, starts a new process for each one, and restarts the others if one crashes.In Inside a PostgreSQL server
postmortem: A written review after an incident that sets out what happened, its impact, the contributing causes and the actions that will prevent, detect or limit it next time.In Running products in production, Leading an incident, Postmortems and learning
power: The chance a test detects an effect of at least the minimum detectable effect when it really exists; usually set at 80%.In Experiments: A/B tests, holdouts and store pilots, Hypothesis tests without the fog
power and interest grid: A two-by-two chart with power to change the outcome on one axis and interest in it on the other. Each corner suggests how closely to engage.In Stakeholders and elicitation
Power Query: Excel's and Power BI's recorded, repeatable cleaning steps. Data cleaning
PR AUC: Area under the precision-recall curve, reported by scikit-learn as average precision. A random model scores about the base rate.In Classification and churn
PR/FAQ: Press release and frequently asked questions: a short launch announcement written before the product exists, followed by the hard questions customers and leaders will ask.In Writing specs people read
PRD: Product requirements document: the problem, goals, scope and requirements for one feature or product, written by the product manager.In Writing specs people read
Pre-commit hook: A script Git runs just before creating a commit; if it exits non-zero, there is no commit. husky installs it and lint-staged keeps it fast. VS Code for teams: formatting, linting and shared settings
pre-mortem: A meeting held before a project starts, where the team imagines it has already failed and lists the reasons why.In Writing for decisions
preattentive attributes: Visual features such as colour, size and position that the eye notices in a fraction of a second, before reading.In Charts that answer the question
Precision: TP divided by (TP + FP): of the customers you flagged, the share who really churn.In Classification and churn, Alerting that does not burn people out, Writing detections
Precision and recall: Share of flagged cases that were right, and share of real cases that were caught. Churn
predefined roles: Roles that come with PostgreSQL and carry a fixed set of system rights, such as reading statistics or cancelling queries.In Security and access
Prefill: The first phase of answering: the model reads the whole prompt in one parallel pass and fills the KV cache. It keeps the GPU's math units busy.In ChatGPT: serving a large language model to hundreds of millions
preflight: An automatic OPTIONS request the browser sends before a non-simple cross-origin request, asking the server whether the real request is allowed.In JavaScript in the browser
prepayment: Cash paid before the period it covers. It is an asset that is expensed month by month as the period passes.In The three statements, Budgets, capex and opex
present value: The value today of money paid in the future, found by dividing each amount by (1 + rate) for each period it is away.In Build, buy or outsource
presigned URL: A URL that carries a signature in its query string, allowing one operation (such as PUT) on one object key until it expires. Anyone holding the URL can use it.In Files, object storage and CDNs
Preventive guardrails: A rule that blocks an action before it happens, such as creating a server without a team tag.In Cloud architecture and cost
preview: A working copy of the app shown next to the chat, so you can click through it as you build.In What AI coding tools can and cannot do
preview deployment: A deployment of one branch or pull request at its own temporary web address, for testing a change before real users see it.In Ship it and know your limits
preview mode: A temporary editor tab, shown in italics, that the next file you open replaces. Double-click or edit the file to keep it open.In VS Code essentials
price-rise cap: A clause that limits how much the vendor can raise prices each year or at renewal, such as no more than 3% or the inflation rate.In Build, buy or outsource
primacy effect: A short-lived drop because people are used to the old way and stumble on the new one, which fades as they learn.In Designing and reading an A/B test
primary: The one database server that accepts writes. Every change starts here and is then copied to the replicas.In Replication and read replicas
primary key: The column (or columns) that uniquely identifies each row. It must be unique, never NULL, and should never change.In Modeling data in tables, Data model and API, Data and security basics
primary metric: The one metric chosen before the test that decides whether the change ships.In Designing and reading an A/B test
primary user: The person who will actually tap through the app to do the job. Other people may be affected, pay for it or decide on it.In From idea to one-page spec
primary_conninfo: A connection string the standby uses to reach its primary and stream WAL, for example host, port, user and application_name.In Replication and failover
primitives: A simple value that cannot be changed in place, such as a number or a string. It is copied when you assign it to another variable.In JavaScript in one sitting
principal: The authenticated caller as the API sees it after login: user id, role and the shops they may act for.In Who are you and what may you do?
principal engineer: A very senior engineer who shapes technical direction across a whole area or department, usually working closely with directors or the CTO.In Organisation and operating model
principle of proportional ink: The rule that the size of the shaded area or length in a chart must be proportional to the value it shows.In Charts that answer the question
Prioritisation: Working out, service by service, what to test first, based on what it guards, who can reach it and how out of date it is.In Attacking networks and services
Priority: How soon a defect should be fixed, compared with everything else the team could work on.In Bugs, reports and acceptance
private address: An address from a reserved range (10.x.x.x, 172.16-31.x.x, 192.168.x.x) that only works inside one local network.In Networking you actually need
privilege creep: Access that builds up over time because people gain new rights when they change jobs but never lose the old ones.In Accounts, SSO and MFA
privilege escalation: Gaining more permissions than the account you started with, for example going from a normal user to an administrator.In How breaches really happen, Broken access control
problem: The unknown underlying cause of one or more incidents. A problem record exists to find and remove that cause.In Incidents, problems and changes
problem details: A standard JSON error body for HTTP APIs (RFC 9457) with the members type, title, status, detail and instance, sent as application/problem+json.In Designing an API people can use, Your first NestJS service
problem matcher: A pattern that reads a task's terminal output and turns error lines into entries in the Problems panel, with clickable file and line.In VS Code for teams: formatting, linting and shared settings
problem statement: Two to four sentences that say who has a problem, what they are trying to do, what goes wrong today and what it costs, with the evidence for it.In From idea to one-page spec, Problems before solutions, What business analysts do
process: A program that is currently running, with its own memory, ID number and environment variables.In How code becomes a running program, Linux from the command line
process cycle efficiency: Touch time divided by cycle time. A low value means the case spends most of its life waiting.In Modelling processes
process mining: Analysing the event logs of IT systems to discover how a process really runs, how long each step takes and where cases loop back.In Modelling processes
process model: A diagram of the steps, decisions, people and hand-offs that turn a trigger into a result, such as a phone call into a delivered order.In Modelling processes
process owner: The named person on the requesting side who will change the procedure, brief the people affected and accept the result of a workflow change.In Product management for internal tools
Process time: The hands-on time someone actually spends working on one item in a stage, without waiting.In Value streams and processes
producer: The code that creates jobs or messages and puts them on a queue, usually your API.In Queues, background jobs and pub/sub
product: Something people use to get a job done, such as an app, a tool or a service, that a team keeps improving over time.In What a product manager actually does
product review: A recurring meeting where a product team brings work to leaders and stakeholders for a decision: a strategy, a spec, a launch or a metric result.In Stakeholders, negotiation and leading without authority
product risk: A risk to the quality of the product, such as a wrong fee, a double charge or a slow screen. Testing helps reduce it.In What testing is for
product trio: The product manager, designer and tech lead who work together to decide what to build.In What a product manager actually does
product-market fit: The state where a clear group of users gets enough value from a product that they keep using it without being pushed, and tell others about it.In Product-market fit before growth
production: The live version of your app that real users reach at your main web address.In Ship it and know your limits
profile: A named set of VS Code settings, extensions, shortcuts and snippets you can switch to, such as one for Node work and one for writing.In VS Code for teams: formatting, linting and shared settings
Progressive delivery: Releasing in automated steps, where each step is checked against metrics and the release stops or reverses by itself if a check fails.In Releasing safely
progressive enhancement: Building a page so its core job works with plain HTML, then adding JavaScript features for browsers that load them.In Forms that people can trust
Project: A GitHub planning space that collects issues, pull requests and draft notes from many repositories, with custom fields and table, board and roadmap views.In Planning on GitHub: issues, labels and Projects, What a product manager actually does, End-to-end tests with Playwright
project risk: A risk to the work itself, such as late deliveries from a supplier, missing people or a poor test environment. It is managed with plans and people.In What testing is for
promise: An object that stands for a result that is not ready yet. It ends exactly once: fulfilled with a value, or rejected with an error.In Async JavaScript: callbacks, promises and async/await
promote: Telling a replica to stop replaying WAL and start accepting writes as a normal primary.In Replication and read replicas
prompt: The short text the shell prints when it is ready for your next command, often showing the user and current folder.In The terminal without fear, What AI coding tools can and cannot do, Building in small steps
Prompt caching: An API feature that stores the processed form of the beginning of a prompt for a few minutes, so later requests that start with exactly the same text skip that work and are billed at a lower rate.In Claude: how Anthropic serves Claude at scale, How AI features work, just enough for PMs, Cost, latency and safety, What makes an agent, Memory, state and multi-step plans, Cost and latency at scale
Prompt injection: An attack where text the model reads, from a user or a document, contains instructions that try to override its rules.In Designing AI features people can trust, Prompts and structured output, Tools and the Model Context Protocol, Cost, latency and safety, Designing AI features: LLMs, RAG and vector search, What makes an agent
properties: The details attached to an event, such as the store, the app version or the number of items, each with a fixed name and type.In Instrumentation: tracking plans and data you can trust
props: The inputs a parent passes to a child component, as one object of named values. The child reads them and must not change them.In Components, props and how React renders
prototype: The object that another object falls back to when it does not have a property itself. Each object has at most one prototype, and prototypes link into a chain.In Prototypes, classes and modules, Prototyping with AI, What AI coding tools can and cannot do
Prototype chain: The chain of objects JavaScript searches when a property is not found on the object itself. Prototypes, classes and modules
Prototype pollution: An attack where input with a key like __proto__ makes code write onto Object.prototype, so every object in the program suddenly has the attacker's property.In Security and quality in JavaScript projects
provenance: A signed, public record that links a published npm package to the exact source repository, commit and CI workflow that built it.In Security and quality in JavaScript projects, Secure supply chain and delivery metrics, Supply chain and secrets
provider: A plugin that knows one platform's API, such as AWS, GitHub or Cloudflare, and turns resource blocks into API calls.In Infrastructure as code, Contract tests
provider state: A named setup step, such as an order exists, that the provider runs before an interaction is replayed so the right test data is in the database.In Contract tests
pseudonymous key: An ID that stands in for a person, such as a random customer key, so the data does not show who they are, while a separate, protected table can still link back.In Responsible ML in a business
Psychological safety: A shared belief that people can speak up, ask and admit mistakes without being punished. Team health
Pub/sub: Publish/subscribe: a publisher sends an event to a topic, and every subscriber of that topic receives its own copy.In Queues, background jobs and pub/sub
PUBLIC: A pseudo-role that every role belongs to automatically. Rights granted to PUBLIC are granted to everybody.In Security and access
public address: An address that is reachable from anywhere on the internet. Only one machine in the world uses it at a time.In Networking you actually need
public key: A key that identifies your project to the backend service. It is meant to sit in browser code, so anyone can read it. Supabase calls it the publishable key (older projects: the anon key).In Data and security basics
pull request: A request on GitHub to merge the commits of one branch into another, with a page for discussion, review and automated checks.In Pull requests and code review
Pull request (PR): A request to merge one branch into another, with the diff, CI checks, review and discussion in one place. Pull requests and code review
pulled forward: Demand that moves to an earlier day, such as shopping done before a festival, leaving a dip on the days after.In Forecasting demand
purchase order: The order a Mangrove buyer sent to the supplier: which articles, how many cases, at what price.In Exploratory testing
pure: A pure component returns the same JSX for the same props and state, and changes nothing that existed before it was called.In Components, props and how React renders
pure function: A function whose result depends only on its arguments and that changes nothing outside itself, so the same input always gives the same output.In Functions, scope and closures, Testing what matters, Unit and API tests
purl: Package URL: a standard string that names a package in any ecosystem, for example pkg:npm/express@4.21.2.In Supply chain and secrets
Purpose limitation: Using personal data only for the purposes people were told about when it was collected, or compatible ones.In Responsible ML in a business
push bombing: Sending a victim prompt after prompt, often late at night, until they tap Approve to make it stop. Also called MFA fatigue.In Identity: passwords, MFA and passkeys
push notification: A message your server sends to an app on a phone through the platform's push service, shown even when the app is closed.In Device features
Push protection: A secret scanning feature that rejects a git push when it contains a recognised secret, before the secret reaches the repository.In Supply chain and secrets
push token: A string that identifies one installed copy of an app on one device, so a push service knows where to deliver a message.In Device features
pyramid principle: A way to structure writing with the main answer at the top, a few grouped reasons below it, and evidence below each reason.In Executive updates and presentations, Telling the story

Q

QoS class: Quality of Service class: Guaranteed, Burstable or BestEffort, worked out from requests and limits. It decides which pods are evicted first when a node runs short of memory.In Kubernetes for application teams
QPS: Queries per second: how many requests the system receives each second.In What system design really is, Speed numbers and back-of-the-envelope math
quality attribute: A measurable property of a system that the business cares about, such as speed, availability, security or how easy it is to change.In What architects actually do, Quality attributes and constraints
quality attribute scenario: A short, testable statement of how the system must respond to one event in one situation, written in six parts from source to response measure.In Quality attributes and constraints
Quality Attribute Workshop: A facilitated meeting, defined by the Software Engineering Institute, where stakeholders brainstorm, merge, vote on and refine quality attribute scenarios before the architecture is designed.In Quality attributes and constraints
quality characteristics: One side of software quality, such as performance efficiency or security, that you can specify, test and measure.In What testing is for
quantile: A value that demand stays at or below on a given share of days. The 80th percentile is the level demand stays under on 8 days in 10.In Forecasting demand
quarantine: Setting aside data that fails checks in a separate place, instead of rejecting or dropping it, so a person can inspect and fix it later.In Case study: syncing POS sales from hundreds of stores, Continuous integration that people trust
Query planner: The part of the database that picks how to run a query, using table statistics. Query plans
queryid: A number Postgres computes from the structure of a normalised query. The same query shape gets the same queryid.In Finding and fixing slow queries
queue: A list of jobs waiting to be processed, stored durably so they survive restarts, from which workers take jobs in order.In Queues, background jobs and pub/sub
Quick Open: A search box that opens any file in the project by typing part of its name. Ctrl+P (Cmd+P on macOS).In VS Code essentials
quorum: The minimum number of copies that must take part in a read or write for it to count, chosen so that reads and writes overlap.In CAP, PACELC and consistency
quorum commit: Synchronous replication where a commit waits for any N of the listed standbys, whichever answer first, instead of named ones.In Replication and failover

R

R²: The share of the variation in y that the model explains, from 0 to 1; with one x variable it equals r squared.In Regression and correlation
race condition: A bug where the result depends on which of two things finishes first, such as two requests answering in a different order than they were sent.In State, effects and where data should live
RACI: A table that lists, for each activity, who is Responsible (does the work), Accountable (answers for it and signs off), Consulted (asked before) and Informed (told after).In Responsible ML in a business, Stakeholders and elicitation
RAG: Retrieval-augmented generation: find relevant text from your own documents first, put it in the prompt, and let the model answer from it.In Retrieval-augmented generation
RAG (retrieval-augmented generation): Finding relevant documents first and putting them in the prompt so the model answers from your data. RAG
random forest: Many decision trees, each trained on a different random sample of the rows, whose predictions are averaged.In Training and comparing models
random sample: A sample in which every unit in the population has a known, usually equal, chance of being chosen, decided by chance and not by a person.In Samples, uncertainty and confidence intervals
Randomisation unit: What gets assigned to A or B: a customer, a store, a session. A/B test design
Randomized search: Trying a fixed number of random combinations drawn from lists or distributions of hyperparameter values.In Training and comparing models
range: A rule in package.json that says which versions of a package are acceptable, such as ^5.5.0.In Node.js and npm
Ransomware: Malicious software that encrypts an organisation's files and demands payment to unlock them; many groups also copy the data first and threaten to publish it.In Thinking like an attacker, How breaches really happen
rate limit: A rule that caps how many requests a client may make in a period of time, and rejects or delays the rest.In Rate limiting and protecting your system, How LLM APIs work, Capacity, resilience and game days
Rate limiting: Capping how many requests a client may make in a period, answering 429 when it is exceeded. Rate limiting
RBAC: Role-based access control: users get roles, such as cashier or zonal manager, and each role allows a set of actions.In Security essentials, Who are you and what may you do?, Accounts, SSO and MFA, Kubernetes for application teams
Re-render: React calling a component again to work out the new UI, then updating only what changed. State and effects
React Compiler: A build-time tool from the React team, version 1.0 released in 2025, that adds memoization to components and hooks automatically.In Accessible, fast pages
read replicas: A copy of the database that receives every change from the primary and serves read-only queries.In ChatGPT: serving a large language model to hundreds of millions
read-after-write: After a user writes something, that same user always sees the write in later reads, even if others may see it later.In Facebook: the social graph, TAO and memcache at planet scale
read-your-own-writes: A guarantee that after you save something, your next read shows your own change, even if other people may still see an older copy for a while.In Loading server data
readiness: A health check that answers: can this copy serve real requests right now, for example is its database connection working? Used to decide whether to send it traffic.In Load balancers
readiness check: A health check that tells the load balancer a container can take traffic. It should pass only after the model and connections are loaded.In Serving: batch and online
Real-time inference: Running the model on each request because its inputs only exist at request time. Also called online inference.In Serving: batch and online
reap: Collecting the exit status of a finished child process so the kernel can free its entry. Unreaped children stay as zombie processes.In Containers from the inside
Reassignment: Pointing a variable name at a different value. Only that one name changes; any object it used to point to is untouched.In Objects, arrays and references
Rebase: Replaying your commits on top of another commit to make history linear; it rewrites those commits. Rewriting history safely: rebase, squash, cherry-pick, stash
rebuffer: A pause in playback because the buffer ran empty and the next segment has not arrived yet. Also called a stall.In Netflix: streaming to every screen
Recall: TP divided by (TP + FN): of all real churners, the share you flagged.In Classification and churn, Alerting that does not burn people out, Writing detections
Recall@k: For one question, the share of its relevant chunks that appear in the top k results. Averaged over a labelled set, it measures how well retrieval finds what the answer needs.In Retrieval-augmented generation
receivable: Money a customer owes for goods already delivered. It is an asset until the customer pays.In The three statements, Unit economics
recency bias: The habit of giving recent events more weight than earlier ones when judging a longer period.In Performance and growth
Reciprocal rank fusion: Reciprocal rank fusion: each list gives an item 1/(k + rank) points, with k often 60, and items are sorted by their total.In Retrieval-augmented generation, Designing AI features: LLMs, RAG and vector search
reconcile: To compare a total from your data with the same total from an independent source, and explain any difference.In Cleaning data in a spreadsheet
reconciler: A background job that compares your records with another system's, here asking the payment provider about payments that are still unconfirmed, and fixes the difference.In Case study: ordering without overselling stock
reconciliation: React's comparison of the new element tree with the previous one, to work out the smallest set of changes to make to the page.In Components, props and how React renders
Reconnaissance: The phase of a test that maps the target's attack surface before any attempt to exploit it.In Reconnaissance
recycles: Reusing a row's existing native views and component for a different item when it scrolls off screen, instead of unmounting it and mounting a new one.In Fast on a cheap phone
RED: Rate, Errors, Duration: the three numbers to track for every request-serving service.In Observability: logs, metrics and traces
red-team exercise: A planned exercise where a group of people deliberately try to make a system fail or misbehave, the way an attacker or a careless user would, so weaknesses are found and fixed before real users find them.In Guardrails and red-teaming
Red-teaming: Deliberately attacking your own AI feature to find harmful or unsafe behaviour before users do. Red-teaming
Redirection: Sending a command's output into a file (> replaces, >> appends) instead of the screen.In The terminal without fear
Redis adapter: A Socket.IO plug-in that uses Redis pub/sub so that an emit on one server reaches matching sockets on every server.In Real-time updates: polling, WebSockets, SSE and push
Redis Stream: A Redis data type that works like an append-only log. Producers add entries with XADD; consumer groups read them, and each reader acknowledges what it has processed with XACK.In Case study: a URL shortener
ReDoS: Regular expression denial of service: input crafted so a regular expression takes a very long time, blocking the event loop.In Security and quality in JavaScript projects
reducer: A pure function (state, action) => newState that describes how state changes for each kind of action.In State, effects and where data should live
reference: The address of an object in memory. Copying a reference gives you a second way to reach the same object, not a second object.In Objects, arrays and references
reference architecture: A tested, documented design for a common kind of system, which teams can reuse without a separate review.In Frameworks and governance
refinement: A regular team meeting where upcoming stories are discussed, clarified and sized before work starts.In What testing is for
reflected XSS: XSS where the unsafe text comes from the current request, such as a search term in the link, and is echoed straight back in the page.In Injection and unsafe input
reflog: A local log of every position HEAD and each branch has pointed to, with the command that moved it. It lets you find commits that are no longer on any branch.In Undo anything: restore, reset, revert and reflog
reforecast: A new estimate of the full year that combines months already closed with a fresh forecast of the rest, for example 3 months of actuals plus 9 months of forecast.In Budgets, capex and opex
refresh token: A long-lived random token, stored in the database as a hash, used only to get a new access token when the old one expires.In Security essentials, Who are you and what may you do?, Authentication, sessions and tokens
refs: A named pointer to an object, usually a commit: a branch, a tag or a remote-tracking branch, stored under .git/refs or in .git/packed-refs.In Inside the .git folder: objects, refs and HEAD
reftable: A newer binary format for storing refs in a single structure instead of one file per ref. It is planned to become the default for new repositories in Git 3.0.In Inside the .git folder: objects, refs and HEAD
regions: A geographic area where a cloud provider has data centres, such as US East in Virginia. A region contains several availability zones.In Netflix: streaming to every screen
registry: A public server that stores published packages. npm downloads from registry.npmjs.org by default.In How code becomes a running program, Containers from the inside
Regression: A prediction task whose output is a number, such as units sold.In Framing a problem as a model
regression testing: Rerunning earlier tests after a change to check that nothing that used to work is now broken.In What testing is for, Bugs, reports and acceptance
regression to the mean: A group chosen because its measurements were extreme will, on average, measure closer to the average next time, because part of what made it extreme was luck.In Traps that fool smart people
Regularisation: A penalty added to the loss that keeps model parameters small or simple, so the model fits less noise.In Training and comparing models
Rekor: Sigstore's public transparency log. It records every signing event in an append-only log, so anyone can check that a signature was made while the certificate was valid.In Secure supply chain and delivery metrics
relation: In Prisma, a pair of fields that link two models: a scalar field holding the other row's id, and an object field you can load or filter through.In Data with PostgreSQL and Prisma
relative luminance: How bright a colour looks to the eye, on a scale from 0 for black to 1 for white, calculated from its red, green and blue values.In Accessibility, devices and the real world
relative path: A path that starts from the current directory, such as data/sales.csv or ../scripts.In The terminal without fear
relay: A background process that reads unsent rows from the outbox table, publishes them to the broker, and marks them as sent.In Events, the outbox pattern and sagas
release: The moment users start getting the new behaviour.In Releasing safely
release build: A build of the app with development features turned off: no dev warnings, minified JavaScript, and Hermes bytecode compiled ahead of time. It is what users install.In Fast on a cheap phone
remote: A named link from your repository to another copy of it, usually on GitHub, with a URL. git remote -v lists them.In GitHub: remotes, clone, push, pull and authentication
remote-tracking branch: A read-only local ref such as origin/main that records where a branch on the remote pointed the last time you fetched. It only moves when you fetch, pull or push.In GitHub: remotes, clone, push, pull and authentication
Rendering: In React, rendering means calling your component functions to get elements. It does not mean drawing pixels or changing the page.In Components, props and how React renders
replenishment order: The order a store sends to its distribution centre to refill its shelves, usually computed by a system every night.In Framing a problem as a model
replicas: A database server that keeps a continuously updated, read-only copy of the primary by replaying its changes.In Replication and read replicas
ReplicaSet: A Kubernetes object whose only job is to keep a set number of identical pods running. Deployments create one per version of the pod template.In Kubernetes for application teams
replicated: Several identical copies of the same data, on different machines, so any copy can answer and losing one is fine.In Google Search: crawling, indexing and answering in a fraction of a second
replication lag: How far a replica is behind the primary, measured in bytes of WAL or in seconds.In Replication and read replicas
replication slot: A record on the primary of how far one consumer has read the WAL, so the primary keeps every WAL file that consumer still needs.In Replication and read replicas, Replication and failover, Upgrades, maintenance and on-call
repository: The .git folder: every commit ever made in this project, plus branches and settings.In What Git really is: snapshots and the three areas
reprocessing window: A fixed number of recent days that a pipeline recomputes from scratch on every run, so that late-arriving data is included.In Data and feature pipelines
Reproducibility: Being able to rebuild a run later from its records and get the same model, or one within an agreed small tolerance.In Experiments, versions and the model registry
request handlers: In MSW, a function that matches a request by method and URL and returns the response the fake server should send.In Testing React like a user
request hash: A short fingerprint, such as a SHA-256 digest, computed from the request body. Equal bodies give equal hashes, so the server can tell whether a retry asks for the same thing.In Background jobs and idempotency
request id: A unique id given to one incoming request and attached to every log line, every downstream call and the response, so you can find everything about that request. Also called a correlation id.In Observability: logs, metrics and traces, Bugs, reports and acceptance
request intake: The single, agreed way that requests reach a product team, with a form, a regular triage and a promised reply time.In Prioritisation that survives the meeting
request key: A unique code the app creates once per order and sends with every attempt, so the server can tell a retry from a new order and save it only once. Also called an idempotency key.In Reading and fixing what the AI wrote
request waterfall: A chain of requests that run one after another even though some of them could run at the same time, so the total time is the sum instead of the slowest one.In Loading server data
required status check: A CI check that must pass before GitHub allows a pull request to merge, set in a ruleset or branch protection.In Continuous integration that people trust
Requirements traceability: Linking each need to its requirements, stories and tests. Requirements
requirements workshop: A structured meeting where a facilitator guides a group of stakeholders to a defined result, such as a ranked list or an agreed process.In Stakeholders and elicitation
rerank: Rescoring a short list of retrieved candidates with a more accurate, slower model, and keeping the best few.In Retrieval-augmented generation
Reranker: A model that reorders retrieved chunks by how well they answer the query. RAG
Reset vs revert: Reset moves a branch pointer back (rewrites history); revert adds a new commit that undoes an old one (safe after pushing). Undo anything: restore, reset, revert and reflog
residual: The actual value minus the value the model predicts for that row; positive means the row was above the line.In Regression and correlation, Training and comparing models
residual plot: A scatter plot of each row's residual against its x value or its predicted value, used to check whether a straight-line model fits.In Regression and correlation
residual risk: The risk that remains after controls are in place; someone with authority must accept it or reduce it further.In Thinking like an attacker, Risk, compliance and the first hour of an incident
resolver: A DNS server, usually run by your internet provider or a public service, that finds answers for you and caches them.In What happens when you open an app or website
resource: A thing your API exposes, such as a GRN, an order or a store's stock. Named with a noun in the URL.In APIs done right
Resource attributes: Attributes that describe the process sending the telemetry, such as service.name and service.version. They are attached to every span, metric and log it sends.In Observability with OpenTelemetry
Responsible disclosure: Reporting a vulnerability privately to the owner and giving them time to fix it before any public disclosure.In Ethics, law and scoping
REST: A style of HTTP API where URLs name things (resources) and HTTP methods say what to do with them.In APIs done right
rest parameter: A last parameter written with three dots, ...notes, that collects all remaining arguments into a real array.In Functions, scope and closures
restart storm: Many containers restarting together because of a shared cause, often a liveness probe that checks a shared dependency, which adds load and makes recovery slower.In Kubernetes for application teams
restart_lsn: The oldest WAL position a replication slot still needs. WAL after this point cannot be removed while the slot exists.In Replication and failover
restore drill: A planned, timed restore of a real backup onto a spare server, with checks and a written result, repeated on a schedule.In Backups you have actually restored
restore point: A label written into the WAL with pg_create_restore_point(), so a later recovery can stop exactly at that point by name.In Backups you have actually restored
retainer: A fixed monthly fee that buys an agreed amount of service, such as support and a set number of hours.In Build, buy or outsource
Retention: The share of a group of users who are still doing the product's core action a given number of periods after they started.In Retention and cohorts, Product-market fit before growth
retention curve: A line showing the share of a cohort still active at each period after they started.In Retention and cohorts, Product-market fit before growth
Retrieval: Retrieval-augmented generation: finding the most relevant pieces of your documents for each question and adding them to the model's context, so it answers from them and can cite them.In How AI features work, just enough for PMs
retry: Trying a failed operation again, usually after a short wait that grows each time.In Async JavaScript: callbacks, promises and async/await
retry storm: A flood of retries that adds load to a struggling service and stops it from recovering.In Failure is normal: timeouts, retries, backoff and circuit breakers
reverse proxy: A server that receives requests from clients and forwards them to servers behind it, such as Nginx in front of a NestJS app.In Networking you actually need
revert: A Git command that makes a new commit undoing an earlier one, so history is kept.In Ship it and know your limits
review board: A small group of senior architects and business owners that reviews the few high-impact decisions, with a fixed time limit for its answer.In Frameworks and governance
RFC 3339: An internet standard that fixes one exact way to write ISO 8601 dates and times, such as 2026-10-20T08:15:00Z.In Designing an API people can use
RFI: Request for information: a short questionnaire sent to many vendors early on, to see who could meet your needs before asking for full proposals.In Build, buy or partner
RFM: Recency, frequency, monetary: days since the last order, number of orders, and money spent. The oldest and strongest churn features.In Classification and churn
RFP: Request for proposal: a document that describes your needs and asks shortlisted vendors for a detailed proposal and price in a set format.In Build, buy or partner
RICE: A scoring method: reach times impact times confidence, divided by effort. Prioritisation
right to win: What you have that rivals cannot easily copy, which gives you a better chance of winning a particular market or job.In Vision, strategy and why now
right-skewed: Not symmetric: one side of the distribution stretches out much further than the other. Right skew means a long tail of large values.In Describing data honestly
Rightsize: Changing a resource to a smaller or cheaper size that still meets its real load, based on measured usage.In Cloud and SaaS costs with FinOps
Ringpop: Uber's open-source library that gives a group of servers a shared consistent-hash ring, so each key has one owner and requests are forwarded to it.In Uber: matching riders and drivers in real time
rings: Groups of devices that receive an update one after another, smallest and safest first, so a bad update is caught before it reaches everyone.In Networks, endpoints and the cloud
risk: Something that may happen and would cause harm if it did. Its level comes from its likelihood and its impact.In What testing is for, Thinking like an attacker
risk appetite: How much risk leadership is willing to take on to reach its goals, written down so teams can make consistent choices.In Risk, compliance and the first hour of an incident, Risk, security and compliance
risk owner: The person with the authority and the budget to decide what happens about a risk, usually a business leader rather than the security team.In Risk, compliance and the first hour of an incident
risk register: A list of an organisation's risks, each with a statement, likelihood, impact, score, owner, chosen treatment and review date.In Thinking like an attacker, Risk, compliance and the first hour of an incident, Risk, security and compliance
risk tier: A level, such as low, medium or high, given to an AI use according to how much harm it could do to people, and used to decide which tests, reviews and sign-offs it needs.In Governance and accountability
Risk-based testing: A test approach where risk analysis decides what you test, in what order and how thoroughly.In What testing is for
riskiest assumption: The belief that, if it turns out to be false, makes the whole idea fail. Test it first and as cheaply as you can.In From idea to one-page spec
RMSE: Root mean squared error: square the misses, average them, take the square root. Big misses count much more.In Forecasting demand
Roadmap: A shared view of the problems the team will work on and roughly when. Roadmaps
robust standard errors: Standard errors that stay valid when the spread of residuals changes across the data; statsmodels offers them with cov_type such as HC3.In Regression and correlation
ROC AUC: Area under the ROC curve: the chance that a randomly chosen churner gets a higher score than a randomly chosen stayer. 0.5 is a coin flip, 1.0 is perfect.In Classification and churn
ROI: Return on investment: total net benefit over a stated horizon minus the investment, divided by the investment. Usually not discounted, so it ignores timing.In Business cases a managing director will sign
role: A type of user with its own set of permissions, such as shop owner, sales rep or admin.In A spec the AI can follow, Security and access
rollback: Putting back the previous working version after a bad release, for example by deploying the previous container image.In Continuous delivery, Ship it and know your limits
rolling deploy: A deploy that replaces app servers one or a few at a time, so some run the old version and some the new one for a while.In Changing the schema without downtime
rolling mean: The average of the label over a sliding window of recent days, such as the last 7 or 28 days.In Features and leakage
rolling update: A release that replaces old pods with new ones a few at a time, so both versions serve traffic for a while.In Releasing safely
rolling window: A window that always covers the last N days up to now, so it moves forward by one day every day.In How reliable is reliable enough?
rolling-origin backtest: A backtest that repeats the train-then-validate split several times, moving the cut-off date forward each time, usually with a growing training window.In Features and leakage
root cause: The deepest cause that you can actually remove, so the incident cannot happen the same way again.In Incidents, problems and changes, Leading an incident, What testing is for
Roots: The starting points the GC always treats as alive: global variables, module-level variables, the variables of functions that are running now, and callbacks still registered with timers or events.In Performance and memory in JavaScript
rotate: To replace a key or password with a new one and switch off the old one, so anyone who saw the old one can no longer use it.In Ship it and know your limits
rotation: Issuing a new refresh token every time one is used, and marking the old one as used, so a stolen old token can be detected.In Security essentials
round trip (RTT): Round-trip time: how long it takes for a small message to go from the client to the server and for the reply to come back.In What happens when you open an app or website, Networking you actually need
route: The address of one screen in the app, such as /orders/5012, the way a web page has a URL.In Building the app
route group: An Expo Router folder whose name is in parentheses. It groups routes and can have its own layout, but its name never appears in the URL.In Navigation and app structure
routing table: The list of networks a router or computer knows, and for each one, where to send packets heading there.In Networks from the till to the data centre
row: One line in a table: one shop, one order or one product.In Data and security basics
row lock: A lock on one row, taken by UPDATE, DELETE or SELECT ... FOR UPDATE. Other writers of that row wait; plain reads of it do not.In Transactions, ACID, isolation and locks
row-level security: A Postgres feature where a table policy filters which rows each query can see or change, based on the current user or a session setting.In Security essentials, Data and security basics, Self-service without chaos, Security and access, Broken access control
Row-level security (RLS): Policies that filter which rows each role may see or change. DB security
RPO: Recovery point objective: how much recent data you can afford to lose in a disaster, measured in time.In Replication and read replicas, Backups and disaster recovery, Backups you have actually restored
RPO and RTO: Recovery point objective: how much data you may lose. Recovery time objective: how long you may be down. Backups
RSC payload: The React Server Components payload: a compact description of the rendered server tree, with the props for any Client Components, that the browser uses to build and update the page.In Loading server data
RTO: Recovery time objective: the longest you may be down before service is back, from the moment of failure.In Backups and disaster recovery, Backups you have actually restored
rubric: A written list of criteria, each with a clear pass rule, used to grade outputs the same way every time.In Evals: knowing if it got better
ruinous empathy: Kim Scott's name for caring about someone so much that you never tell them the hard thing they need to hear.In 1:1s, feedback and coaching
rules of engagement: The agreed rules for how a test is carried out: timing, intensity, forbidden actions, communication and when to stop.In Ethics, law and scoping, Reconnaissance
ruleset: A named set of rules on GitHub that applies to matching branches or tags, such as requiring pull requests, approvals and passing checks before merging.In Branching strategies, releases and versions
run: One training attempt: one set of code, data and settings producing one set of results and model files.In Experiments, versions and the model registry
run and change: Run is spending that keeps today's systems working (support, licences, cloud, maintenance). Change is spending that builds something new or improves it.In Money: budgets, headcount and vendors
run rate: What an option costs per month once it is running normally, without one-off costs.In Build, buy or outsource
Run versus change: Splitting spend between keeping systems running and building new capability. Budgets
runbook: A short written guide for one alert: what it means, what to check first, and how to fix or escalate.In Observability: logs, metrics and traces, Fresh, tested, trusted, Monitoring, drift and retraining, Alerting that does not burn people out, Upgrades, maintenance and on-call
runc: The reference command-line tool that creates and starts containers according to the OCI runtime specification, by calling Linux kernel features directly.In Containers from the inside
runner: The machine that runs a job. A GitHub-hosted runner is a fresh virtual machine created for that job and deleted afterwards.In GitHub Actions: continuous integration, Continuous integration that people trust
runtime: An engine plus the extra functions a program can call, such as reading files in Node.js or changing the page in a browser.In How code becomes a running program
runtime permissions: A permission the user grants or refuses while the app is running, through a dialog drawn by the operating system, instead of at install time.In Device features
runtime version: A label baked into each build and each update. A build only loads updates whose runtime version matches its own.In Testing and releasing

S

SaaS: Software as a service: software you rent by subscription and use over the internet, run and updated by the vendor.In Build, buy or partner, Networks, endpoints and the cloud, Why enterprise architecture exists, Budgets, capex and opex, Build, buy or outsource
safe: A method that only reads and asks the server to change nothing, such as GET or HEAD.In Designing an API people can use
safe area: The part of the screen not covered by the status bar, notch, rounded corners or the home gesture bar.In How React Native works
saga: A long business process split into a sequence of local transactions, where each step has a compensating action that undoes it if a later step fails.In Events, the outbox pattern and sagas, Case study: ordering without overselling stock
salt: Random data mixed into each password before hashing, so two users with the same password get different hashes.In Security essentials, Cryptography without the maths, Authentication, sessions and tokens
same-origin policy: The browser rule that a page's JavaScript may freely read responses only from its own origin.In JavaScript in the browser
SAML 2.0: Security Assertion Markup Language: an older XML-based standard for SSO, common in business apps.In Accounts, SSO and MFA
sample: The units you actually measure, chosen from the population.In Samples, uncertainty and confidence intervals
Sample ratio mismatch: When the groups in an experiment are not the sizes the split planned, beyond what chance allows; a sign that assignment or logging is broken.In Experiments: A/B tests, holdouts and store pilots, Designing and reading an A/B test
Sampling: Keeping only some traces, such as 10% of requests, to control cost. Head sampling decides at the start; tail sampling decides at the end, so it can keep all slow or failed ones.In Observability: logs, metrics and traces
sampling distribution: The spread of values an estimate would take over many random samples of the same size; you see it by simulation or describe it with a formula.In Samples, uncertainty and confidence intervals
sandbox: A closed copy of the world an agent works in, with fake tools and fake data, so tests can't change anything real and start the same way every time.In Evaluating agents, Analysing with an AI assistant
SAST: Static application security testing: a tool that reads your source code without running it and looks for dangerous patterns, such as user input reaching a SQL query.In Security in the pipeline and in AI features
saturation: The point in a study where new interviews stop revealing new themes, which is a signal that you can stop recruiting.In User interviews that don't lie, Capacity, resilience and game days
savepoint: A marker inside a transaction. You can roll back to it, undoing only the work done after it, without losing the rest of the transaction.In Case study: syncing POS sales from hundreds of stores
SBI: Situation, Behaviour, Impact: a way to give specific, factual feedback. 1:1s and feedback
SBOM: Software bill of materials: a machine-readable list of every component in a piece of software, with names, versions and where each came from.In Supply chain and secrets
SCA: Software composition analysis: a tool that lists the open-source packages in your lockfile and matches their versions against databases of published vulnerabilities.In Security in the pipeline and in AI features
scatter plot: A chart that places one dot per item, with one measure on the horizontal axis and another on the vertical axis, to show whether they move together.In Charts that answer the question, Regression and correlation
scavenge: V8's collector for the young generation. It copies the few live objects out and treats everything left behind as free.In Performance and memory in JavaScript
scenarios: A set of inputs moved together to a consistent story, such as a low case where several things go wrong at once.In Business cases
schema: A namespace inside a Postgres database that groups tables, like a folder. Tables are named schema.table, for example inventory.stock.In Monolith, modular monolith, microservices, Instagram: from 3 engineers on Postgres to a billion users
Schema migration: A versioned script that changes tables, run the same way in every environment. Safe migrations
schema registry: A service that stores every version of each event schema and refuses a new version that breaks the agreed compatibility rule.In Integration and data ownership
schema-constrained output: A model setting that forces the output to match a given JSON schema, so required fields and allowed values are always respected.In Prompting, retrieval or fine-tuning?
SCIM: System for Cross-domain Identity Management: a standard web API the IdP uses to create, update and disable accounts in other apps automatically.In Accounts, SSO and MFA
Scope: The region of code where a variable name can be used. JavaScript has block scope, function scope and module scope, nested inside each other.In Functions, scope and closures, Designing tools and permissions, Risk, compliance and the first hour of an incident, Incident response, Ethics, law and scoping
Scope creep: Uncontrolled growth of what the project must deliver. Change and UAT
scorecard: A short document that says why a role exists, what the person must achieve, and how they should work.In Hiring well, Organisation and operating model
SCR: situation, complication, resolution: A three-part opening for a report or a talk: what the reader already knows, what changed, and what we found and recommend.In Telling the story
SCRAM: Salted Challenge Response Authentication Mechanism: a login method where the password itself never crosses the network and the server stores only a salted verifier.In Security and access
scraping: Prometheus pulling the current metric values from an HTTP endpoint, usually /metrics, on a fixed schedule such as every 15 seconds.In Observability: logs, metrics and traces
screen reader: Software that reads the screen aloud or sends it to a braille display, and lets the user move through it with keys or gestures. NVDA, VoiceOver and TalkBack are examples.In Accessible, fast pages, Testing React like a user, Accessibility, devices and the real world
scribe: The person who keeps the single running timeline of an incident: every observation, action and decision with its time.In Incident response, Stakeholders and elicitation
script injection: An attack where text an outsider controls, such as a PR title or branch name, is pasted into a shell script by the workflow and runs as a command.In Deploying and securing with GitHub Actions
scripted testing: Testing that follows test cases written in advance, with fixed steps and expected results.In Exploratory testing
Scrum: A popular way of organising software work into short, fixed-length cycles called sprints, with set roles and meetings.In What a product manager actually does
SDK: The OpenTelemetry implementation that the application installs at startup. It decides what to sample, batches telemetry and exports it.In Observability with OpenTelemetry
seam: A place in a system where you can change what happens without editing the code at that place, for example by routing a request or swapping an implementation behind an interface.In Evolving a legacy system
search params: The part of a link after the question mark, such as ?q=oil&lowOnly=1. Browsers keep it in history and in shared links.In State, effects and where data should live
Seasonal naive: A forecast that repeats the value from one season ago, such as the same weekday last week.In Framing a problem as a model
Seasonality: A pattern that repeats on a fixed cycle, such as every week or every year.In Forecasting demand
Secret: A Kubernetes object for small sensitive values such as passwords and tokens. Values are base64 encoded, which is not encryption.In Kubernetes for application teams
secret key: A key with full powers over your project. In Supabase it skips row-level security entirely, so it may only be used in server code. Older projects call it the service_role key.In Data and security basics
Secret scanning: Automatic search of code and commits for keys and passwords, ideally blocking the push. Pipeline and AI
secrets and variables: Named values stored on GitHub (Settings, Secrets and variables, Actions). Secrets are encrypted and hidden in logs; variables are plain text.In GitHub Actions: continuous integration
security champion: A developer on a product team who has extra security training and is the first person the team asks about security.In Threat modelling a feature, Security in the pipeline and in AI features
SECURITY DEFINER: A function that runs with the privileges of the role that owns it, not the role that calls it.In Security and access
security group: A cloud provider's firewall rules attached to a server, for example: allow port 443 from anywhere, allow 5432 only from the app servers.In Networking you actually need
seed: A small, fixed data file that sets up a fake store before a test: its stock, deliveries, goods received notes, procedures and the date.In Evaluating agents
segments: A small piece of a video file, a few seconds long, downloaded with one HTTP request. Players can switch quality at segment boundaries.In Netflix: streaming to every screen
Selection bias: A distortion that appears when the people or things in your data got there through a process related to the outcome you are measuring.In Traps that fool smart people
selectivity: The fraction of rows that match a condition. 'id = 5' is very selective (one row); 'pay_method = CASH' is not (a third of the table).In Indexes from the inside
self-contained token: A token that carries its own facts, such as user id and expiry, plus a signature, so the server can check it without a lookup.In Who are you and what may you do?
Self-service BI: Letting business users build their own reports on top of certified, well-named datasets. Self-service
semantic cache: A cache that returns a stored answer when a new question is close in meaning to an old one, measured by embedding similarity, instead of identical in text.In Cost, latency and safety, Cost and latency at scale
semantic conventions: Agreed names and meanings for span attributes, so different tools and vendors can read the same data.In Shipping an agent safely, Observability with OpenTelemetry
semantic layer: A layer between the warehouse and the BI tools where metrics, dimensions and joins are defined once; tools ask for a metric by name and the layer writes the SQL.In Metrics everyone agrees on
Semantic versioning: A version format MAJOR.MINOR.PATCH, where major means breaking changes, minor means new compatible features and patch means compatible bug fixes.In Branching strategies, releases and versions, Node.js and npm
Semi-additive: A measure that can be summed along some dimensions but not along time, such as stock on hand or an account balance.In Metrics everyone agrees on
Sender Keys: A group encryption method: each member shares one sender key with the group once, then encrypts each message only once with it.In WhatsApp: billions of messages with a tiny team
Sensitivity analysis: Testing how far the result moves when you change one input at a time, holding the others at their base values.In Business cases
sensitivity point: In ATAM, a property of the design that one quality attribute depends on strongly, so a small change to it changes that attribute's response a lot.In Choosing an architecture style
sequential scan: Sequential scan: reading every page of a table from start to end and testing every row against the WHERE clause.In Indexes from the inside
serialisable: Possible to write down as data and rebuild on the other side, the way React writes props into the RSC payload for the browser.In Loading server data
serialization failure: An error that means: this transaction conflicted with another one and was cancelled. Running it again from the start will usually succeed.In Transactions, ACID, isolation and locks
server: A program on a computer that waits for requests and sends back responses. The word also means the computer it runs on.In What happens when you open an app or website
Server Component: A React component that runs only on the server, at build time or per request. It can be async and read data directly, and its code is never sent to the browser.In Loading server data
Server Function: An async function marked with the 'use server' directive. It runs only on the server, and the browser calls it with a POST request.In Forms that people can trust
server session: A login stored on the server, in Redis or Postgres, under a random id. Each request looks the id up.In Who are you and what may you do?
Server-Sent Events: Server-Sent Events: a normal HTTP response that never ends; the server writes small text events into it whenever something happens. Server to client only.In Real-time updates: polling, WebSockets, SSE and push, ChatGPT: serving a large language model to hundreds of millions, How LLM APIs work
Service: A Kubernetes object that gives a changing set of pods one stable name and virtual IP address, and spreads traffic across the pods that match its selector and are ready.In Kubernetes for application teams
service accounts: A user account made for one program, not a person. It cannot log in, and it owns only that program's files.In Linux from the command line, Accounts, SSO and MFA, Triage and investigation
service banner: The short text a service returns on connection, often naming the software and version; useful as a lead but easy to hide or change.In Reconnaissance
service blueprint: A diagram that shows the people, systems and backstage work behind each step of a customer's journey, in lanes.In Watching the work
service container: A Docker container, such as Postgres or Redis, that GitHub Actions starts next to a job and stops when the job ends.In Continuous integration that people trust
service credits: Money or credit a supplier owes you when it misses the service levels in the contract.In Build, buy or partner, Build, buy or outsource
service level: The share of days, or of demand, that is met from stock without running out.In Forecasting demand
service level agreement: Service level agreement: a contract that sets measurable targets, such as response and fix times, and what happens if they are missed.In Working with vendors and agencies
service request: A user asking for something normal and expected, such as a new account, a new scanner or a password reset. Nothing is broken.In Incidents, problems and changes
session: An uninterrupted, time-boxed stretch of exploratory testing against one charter.In Exploratory testing
Session fixation: An attack where the attacker gets a session id they already know into the victim's browser before login, so that after the victim signs in the attacker shares the signed-in session.In Authentication, sessions and tokens, Attacking web applications
session guarantees: Promises that hold for one user's session: you see your own writes, and your view never goes back in time. Other users may still see older data.In CAP, PACELC and consistency
session-based test management: A way to manage exploratory testing in time-boxed, uninterrupted sessions, each with a charter, a session sheet and a debrief.In Exploratory testing
Set: A built-in collection of unique values. Adding a value that is already there does nothing.In Working with data: arrays, Maps, JSON, dates and money
settled: A promise that is no longer pending: it has either fulfilled or rejected, and it can never change again.In Async JavaScript: callbacks, promises and async/await
seven levels of delegation: A scale of seven steps for who makes a decision, from the leader deciding alone to the team deciding alone, from the book Management 3.0.In Delegation and mentoring
severity: A level, usually 1 to 4, that says how serious an incident is and therefore how fast and how widely the response must happen.In Running products in production, Leading an incident, Bugs, reports and acceptance, Incident response, From change control to acceptance
Severity and priority: Severity is how bad the bug's effect is; priority is how soon it should be fixed. Bugs and UAT
Severity level: A label such as SEV1 to SEV3 that sets how urgent an incident is and who must be told. Incident command
SHA-1: A hash algorithm that produces 160 bits, shown as 40 hex characters. Git has used it for object IDs since 2005.In Inside the .git folder: objects, refs and HEAD
shadow database: An empty scratch database that Prisma creates, replays all migrations into, and then compares with your schema to work out what the next migration must change.In Changing the schema without downtime
shadow IT: Software or spreadsheets a team builds or buys for itself, outside the official tools, because the official ones don't meet its needs.In Users, customers and stakeholders, Product management for internal tools
Shadow mode: Running a new model on live traffic without acting on its output, to compare it safely. Serving
Shadow run: Sending real requests to a candidate system in parallel with the live one, without showing its answers to users, so you can compare them.In Evals that never stop
shadow tree: An immutable tree of C++ nodes, one per host component, that the renderer uses to calculate layout before it changes any native view.In How React Native works
shadow week: Running a new model on real inputs alongside the current one, storing its outputs but not using them, to see how it behaves before it decides anything.In Experiments, versions and the model registry
shadows: When an inner scope declares a name that an outer scope also has, so the inner one hides the outer one inside that scope.In Functions, scope and closures, Prototypes, classes and modules, From notebook to pipeline
shallow copy: A copy that duplicates only the top level. Nested objects and arrays are not copied; the copy and the original point to the same nested ones.In Objects, arrays and references
SHAP: A method that splits one prediction into contributions from each feature. Explaining
SHAP values: Per-feature contributions to one prediction, based on Shapley values from game theory; they add up from the average prediction to this prediction.In Explaining and choosing models
shard: One of the database servers in a sharded system. It holds the rows whose shard key maps to it.In Sharding and consistent hashing, Google Search: crawling, indexing and answering in a fraction of a second, Continuous integration that people trust
shard key: The column whose value decides which shard a row lives on, for example site_id.In Sharding and consistent hashing
Sharding: Splitting the rows of a table across several database servers, where each server (a shard) holds only part of the data.In Sharding and consistent hashing
shared buffers: Memory PostgreSQL reserves at startup as its own cache of table and index pages, shared by all processes.In Inside a PostgreSQL server
shared mutable fixture: Test setup data that several tests share and that a test can change, so one test's change silently affects the next.In Unit and API tests
shared_buffers: Memory Postgres reserves at startup as its own cache of table and index pages, shared by all connections.In Running Postgres in production
shared-responsibility model: The split of security duties between a cloud provider and its customer: the provider secures the cloud itself, the customer secures what they build and store in it.In Networks, endpoints and the cloud
shebang: The first line of a script, starting with #!, that tells the system which program should run the file.In The terminal without fear, Shell scripting for daily jobs
shell: The program inside the terminal that reads each command you type, runs it and prints the result. Examples: bash, zsh, PowerShell.In The terminal without fear, Injection and unsafe input
ShellCheck: A free program that reads shell scripts and warns about bugs such as missing quotes, unsafe patterns and code that does not do what it seems to.In Shell scripting for daily jobs
Shift left: Doing testing work earlier in development, for example reviewing stories and writing test cases before the code exists.In What testing is for
Short polling: The client asks the server for new data on a fixed timer, for example every 5 seconds, whether anything changed or not.In Real-time updates: polling, WebSockets, SSE and push
short-lived token: An access token that expires after a short time, such as 15 minutes, so a leaked or forgotten token is soon useless.In Designing tools and permissions
short-term containment: Quick action to stop the immediate damage, such as isolating a host or disabling an account.In Incident response
showback: Showing each team the cost of what it runs, without moving money between budgets.In Cloud architecture and cost, Cloud and SaaS costs with FinOps, Money: budgets, headcount and vendors
shuffle sharding: Giving each customer a random small set of servers from a larger pool, so two customers rarely share the same full set and one bad customer cannot take down others.In Amazon: the store that never closes
SIEM: Security information and event management: a system that collects logs from many sources, stores them, lets you search them and runs correlation rules that raise alerts.In Inside a SOC, Logs that matter
SIGKILL: Signal 9. It cannot be caught: the kernel ends the process at once, mid-request if need be.In Containers from the inside
Sigma: An open, YAML-based format for writing detection rules once and converting them to many different SIEM query languages.In Writing detections
sign-in: Checking who a user is, usually with an email and password or a one-time code, before showing them their data.In What AI coding tools can and cannot do
Signal protocol: An open protocol from Signal for encrypted messaging. It sets up keys between two devices and changes the key for every message.In WhatsApp: billions of messages with a tiny team
signals: One kind of data a system emits about itself. The three main signals are logs, metrics and traces.In Observability: logs, metrics and traces, Linux from the command line
signed: Adding a cryptographic signature to an app file so the phone and the store can prove who made it and that nobody changed it.In Testing and releasing
signed URL: A link to one private file that carries a signature and an expiry time, so whoever holds it can download that file until it expires, without logging in.In Threat modelling a feature
significance level: The chance a test declares a winner when the change really has no effect; usually set at 5%.In Experiments: A/B tests, holdouts and store pilots, Hypothesis tests without the fog
significant figures: The digits in a number that carry meaning, counted from the first non-zero digit. 99,847 to two significant figures is 100,000.In Telling the story
SIGTERM: Signal 15, the polite request to stop. A program can catch it, finish its work and exit.In Containers from the inside
SIM swap: A fraud where an attacker convinces a mobile operator to move your phone number to their SIM card, so your texts go to them.In Accounts, SSO and MFA, Identity: passwords, MFA and passkeys
Simpson's paradox: When a trend in combined data differs from, or reverses, the trend inside every group, because the mix of groups changed.In Funnels, activation and adoption, Traps that fool smart people
single point of failure: A part with no backup. If it stops, the whole system stops.In What system design really is, Scaling up vs scaling out
single sign-on: Signing in once to an identity provider, which then vouches for you to many apps, so each app does not keep its own passwords.In Identity: passwords, MFA and passkeys
Single sign-on (SSO): SSO: logging in once at the identity provider and then using many apps without typing a password into each one.In Accounts, SSO and MFA
singular test: A one-off dbt test: a .sql file in the tests folder whose SELECT returns the rows that break a business rule.In Transforming data in the warehouse
SIPOC: A one-page view of suppliers, inputs, process, outputs and customers. Process models
site-to-site VPN: A tunnel between two networks, such as a store and a data centre, that encrypts every packet and carries it across the internet.In Networks from the till to the data centre
Situation-Behaviour-Impact: A feedback method from the Center for Creative Leadership: describe the situation, the behaviour you saw, and its impact.In Delegation and mentoring
Situation-Behaviour-Impact (SBI): A feedback structure from the Center for Creative Leadership: describe the situation, the observable behaviour, and its impact.In 1:1s, feedback and coaching
SKIP LOCKED: A Postgres option that lets many workers claim different rows at once without waiting on each other. Transactions
skip-level: A conversation between a manager's manager and the people who report to that manager, skipping one level.In 1:1s, feedback and coaching
SLA: Service level agreement: a contract with a customer or partner that says what happens, such as a refund or penalty, if a level is not met.In Measuring speed and reliability: p99, SLOs and nines, From notebook to pipeline, Incidents, problems and changes, How reliable is reliable enough?
slack: How long a package can slip before it delays a fixed date or another package.In Target state and roadmap
SLI: Service level indicator: a measured number that shows how well the service is doing, usually the share of good events, such as requests under 300 ms.In Measuring speed and reliability: p99, SLOs and nines
SLI, SLO, SLA: What you measure, the target you set, and the promise with penalties you sign. SLOs and budgets
slice: A group of rows that share a value, such as all small stores or all dairy products.In Explaining and choosing models
Slicer: A clickable filter on a pivot table or dashboard. Pivot tables
SLO: Service level objective: the target for an SLI over a time window, such as 99% of requests good over 30 days.In Measuring speed and reliability: p99, SLOs and nines, Observability: logs, metrics and traces, Performance and load testing, Technology strategy
slope: How much the predicted y changes for one more unit of x; for delivery data, minutes per extra kilometre.In Regression and correlation
slopsquatting: Registering a package name that AI assistants tend to invent, so that people who install the suggestion without checking get the attacker's code.In Working with AI coding assistants
slowly changing dimension: A dimension whose attribute values change from time to time, with a chosen rule (type 1, type 2 and so on) for what happens to history.In From questions to a reporting model
SLSA: Supply-chain Levels for Software Artifacts: an industry framework of levels that describe how trustworthy a build and its provenance are.In Secure supply chain and delivery metrics, Supply chain and secrets
smoke test: A short, automatic check right after a deploy that the most important paths work, such as the health endpoint and one real request. It takes seconds, not minutes.In Continuous delivery, Performance and load testing
snapshot: A consistent picture of the committed data at one moment. Reads use a snapshot, so they never see half-finished changes.In Transactions, ACID, isolation and locks, What Git really is: snapshots and the three areas, State, effects and where data should live, Data model and API, Inside a PostgreSQL server, Finding and fixing slow queries
snippets: A short trigger word that expands into a code template with fields you fill in by pressing Tab.In VS Code for teams: formatting, linting and shared settings
Snowflake ID: A 64-bit ID made from the current time in milliseconds, a machine number and a counter, so many servers can create unique, time-ordered IDs without talking to each other. First used at Twitter.In Case study: a URL shortener
SOAR: Security orchestration, automation and response: tools that run repeatable response steps automatically, such as enriching an alert or disabling an account.In Inside a SOC
SOC: Security operations centre: the team that watches alerts, triages them and responds to incidents. SOC tour
Socket.IO: A popular library on top of WebSockets that adds automatic reconnection, rooms, acknowledgements and a long-polling fallback. Its client and server must both use Socket.IO.In Real-time updates: polling, WebSockets, SSE and push
software architect: An architect who shapes the inside of one system: its modules, layers, patterns and how the code meets its quality targets.In What architects actually do
software bill of materials: Software bill of materials: a machine-readable list of every package in a piece of software, with name, version and a package URL, like the ingredient list on food.In Secure supply chain and delivery metrics
solely by automated means: A decision made entirely by a system, with no meaningful human review, for example an automatic loan refusal.In Governance and accountability
solution architect: An architect who designs how several systems, teams and vendors work together to deliver one business change.In What architects actually do
Solution architecture: The design of one specific change, such as a new product, feature or system, so that it works, fits the existing landscape and can be built and run.In Why enterprise architecture exists
SOP: Standard operating procedure: the agreed, written way to do a task, so everyone does it the same way.In Change management and adoption, Designing AI features: LLMs, RAG and vector search
source code escrow: An arrangement where a trusted third party holds a software supplier's source code and releases it to the customer if agreed events happen, such as the supplier's insolvency.In Working with vendors and agencies
source freshness: A check that compares the newest load time in a raw table with the current time, and warns or fails when the data is older than an agreed limit.In Transforming data in the warehouse
source of truth: The one place whose copy of a piece of data wins when copies disagree. Caches and replicas are copies of it.In Running a design review, Case study: a URL shortener, Launch and learn, Instrumentation: tracking plans and data you can trust, Looking after 1,400 devices
SPACE: A framework for developer productivity with five dimensions: satisfaction and well-being, performance, activity, communication and collaboration, efficiency and flow.In Team design and delivery health
SPACE framework: Five dimensions of developer productivity: satisfaction, performance, activity, communication and efficiency. Operating model
span link: A reference from one span to a span in another trace, used when work is caused by something else but is not part of the same request.In Observability with OpenTelemetry
span of control: The number of people who report directly to one manager.In Organisation and operating model
spans: One timed step inside a trace, such as one database query or one outgoing HTTP call, with a start time, a duration and a parent.In Observability: logs, metrics and traces, Shipping an agent safely, Observability for LLM systems
sparkline: A tiny line chart without axes, about the size of a word, showing the recent shape of a measure.In Designing a dashboard for decisions
spec: A written description of what a feature should do and why, used by the people who design, build and test it.In Writing specs people read
specification by example: A way of agreeing requirements by collecting concrete examples with the business, then using those examples as the acceptance tests.In User stories and acceptance criteria
spending cap: A setting with your provider that alerts you, or stops extra usage, when spending reaches an amount you choose.In Ship it and know your limits
spike: A short, time-boxed experiment (often one or two days) that answers one technical question, such as a quick load test. The code is usually thrown away.In Running a design review, How products create value, User stories and acceptance criteria, Working with engineers, What a tech lead is (and is not), Slicing and planning work
spike test: A spike test jumps to a high rate in seconds, like a push notification wave. A soak test holds the expected peak for hours to expose slow problems such as leaks.In Capacity, resilience and game days
split brain: Two servers both acting as primary and accepting writes at the same time, so the data splits into two versions that cannot be merged automatically.In Replication and read replicas
split DNS: Running DNS so that internal names are answered only by the company's own resolvers, while public names resolve normally everywhere.In Networks from the till to the data centre
sponsor: The senior person who asked for the project, pays for it and makes the final calls on scope and priorities.In What business analysts do, Performance and growth
Spread: The ... syntax used inside an object or array literal to copy the items or properties of another one into the new one.In Objects, arrays and references
sprints: A short, fixed-length cycle, often two weeks, in which a team builds and shows a working piece of the product.In What business analysts do
SQL injection: An attack where user input is pasted into SQL text and changes what the query does, for example adding OR 1=1 to read every row.In SQL you'll actually use, Data with PostgreSQL and Prisma
Squash merge: Merging a PR as a single new commit on the target branch, combining all its commits. Pull requests and code review
SSH: Secure Shell: a program that gives you an encrypted command line on another computer over the network.In Linux from the command line
SSRF: Server-side request forgery: tricking your server into calling a URL the attacker chooses, often an internal address the attacker cannot reach directly.In Security essentials
stack: A navigator that keeps screens in a pile. Opening a screen pushes it on top; going back pops it off and shows the one below.In Navigation and app structure
stack trace: The list of function calls that were active when the error was created, newest first, each with a file, line and column.In Errors, debugging and testing, Ship it and know your limits
Staff engineer: A senior individual contributor level above senior engineer whose work reaches beyond a single team.In What a tech lead is (and is not), Organisation and operating model
staged rollout: Releasing a new app version to a growing share of users, so a bad version hurts a few people before you notice and stop it.In Testing and releasing
staging: A copy of production with its own database, fake data and its own secrets, where each release runs before real users see it.In Continuous delivery
staging area: A list Git keeps of exactly what the next commit will contain. git add copies a file's current content into it. Also called the index.In What Git really is: snapshots and the three areas
stakeholder: Anyone who can affect a product decision or is affected by its result.In Users, customers and stakeholders, Why enterprise architecture exists, Stakeholders and elicitation
stale-while-revalidate: A caching rule: when an entry is old, answer with it immediately and refresh it in the background, so nobody waits.In Loading server data
staleness budget: The longest a copy of data is allowed to lag behind its owner, written as a measurable target.In Integration and data ownership
Standard deviation: A measure of spread: roughly the typical distance of values from the mean. The square root of the average squared distance from the mean.In Describing data honestly
standard error: How much an estimate would move if you repeated the measurement on a different sample; for a pass rate p over n cases it is about the square root of p times 1 minus p, divided by n.In Evals: knowing if it got better, Samples, uncertainty and confidence intervals, Hypothesis tests without the fog, Regression and correlation
standard operating procedures: Standard operating procedure: a written, numbered set of steps for a routine job, such as receiving a delivery or counting stock.In Retrieval over procedures
standby.signal: An empty file in the data directory. If it exists when the server starts, the server starts as a standby and keeps replaying WAL instead of becoming a primary.In Replication and failover
star schema: A fact table in the middle joined to several dimension tables around it, each by one key, so the diagram looks like a star.In From questions to a reporting model
state: Data a component owns that can change over time, such as a search text. Changing it makes React re-render the component.In JavaScript in the browser, Screens and flows, Building the app
state file: A JSON file in which Terraform records, for every resource address in your code, the real object's ID and its attributes as last seen.In Infrastructure as code
state lock: A marker that one Terraform run holds while it reads and writes the state, so a second run waits instead of overwriting the first one's changes.In Infrastructure as code
state machine: A model where something is always in exactly one of a fixed set of states, and only certain moves between states are allowed.In Case study: ordering without overselling stock
state model: A diagram or table of a system's states, the events it accepts in each state, and the state each event leads to.In Designing tests that find bugs
stateful: A firewall that remembers connections it allowed, so the replies come back in without a rule of their own.In Networks from the till to the data centre, Networks, endpoints and the cloud
stateless: Keeping no memory of past requests inside the server, so any copy of the server can answer any request.In What system design really is, What happens when you open an app or website, Scaling up vs scaling out, How LLM APIs work, What makes an agent, Memory, state and multi-step plans
Stateless service: A server that keeps nothing between requests, so any copy can answer any request. Scale up vs out
statement of work: The contract document that describes the work: scope, deliverables, milestones, acceptance, roles, change control, support and handover.In Working with vendors and agencies
statement_timeout: A Postgres setting: the longest any single statement may run before Postgres cancels it.In Changing the schema without downtime
static testing: Testing a work product without running code, for example by reviewing a story, a design or the code itself.In What testing is for
Statistical significance: A p-value below the chosen alpha; not the same as a result that matters. Hypothesis tests
statistics: Summaries Postgres keeps about each column: number of distinct values, most common values and how often they appear, and a histogram. The planner uses them to guess how many rows a condition will match.In Reading query plans and fixing slow queries
status check: A pass, fail or pending result that a tool reports on a commit. Rules can require certain checks to pass before merging.In Pull requests and code review
status code: A three-digit number in the response that says what happened: 2xx success, 3xx go elsewhere, 4xx the client made a mistake, 5xx the server failed.In What happens when you open an app or website, APIs done right, Data model and API, Testing APIs and data
status page: A public web page where a company tells customers which of its services are working, degraded or down, and what it is doing about it.In Leading an incident
status rule: A written rule that turns a tile's gap against its target into a status such as normal, amber or red, agreed before the dashboard is used.In Designing a dashboard for decisions
stderr: Standard error: a separate stream for error messages, so they do not mix with results sent to a file or pipe.In The terminal without fear
stdout: Standard output: the stream where a program writes its normal results.In The terminal without fear
step: One item in a job: either a shell command (run) or a reusable action (uses). Steps in a job run in order.In GitHub Actions: continuous integration
sticky: Routing where follow-up requests from the same user or conversation go back to the same server, usually to reuse its cache.In Claude: how Anthropic serves Claude at scale
sticky session: A load balancer setting that keeps sending the same user to the same backend server, usually using a cookie or the client IP.In Load balancers
stop reason: A field in the response that says why the model stopped writing: it finished, it hit the output limit, or it wants to call a tool.In How LLM APIs work
Stop rules: A condition, agreed before the pilot starts, that pauses or reverses the rollout as soon as it happens, without a meeting.In Shipping an agent safely
stopping cue: A clear signal in the product that the useful part is done, so the user can stop, such as an end to a feed.In Engagement loops, habits and notifications without dark patterns
store build: The installable app file you send to a store: an Android App Bundle (.aab) for Google Play or an .ipa for the App Store.In Testing and releasing
stored XSS: XSS where the unsafe text is saved first, for example in a profile or a note, and runs later for everyone who views it.In Injection and unsafe input
story map: A two-dimensional backlog: user activities left to right in journey order, with the stories for each activity stacked below, most needed at the top.In User stories and acceptance criteria
Story points: A relative size for a piece of work on a team's own scale; useful for that team's forecasts, meaningless when compared across teams.In Working with engineers
story-based interview: An interview built around one specific past event, told in order and probed for detail, instead of opinions or general habits.In User interviews that don't lie
Strangler fig: Replacing a legacy system piece by piece behind a routing layer until the old one can be switched off. Legacy migration
strangler fig pattern: Replacing an old system gradually: put a proxy in front, move one feature at a time to new code, and remove the old code when nothing uses it.In Monolith, modular monolith, microservices
Strategic: A dashboard for leaders that tracks a few high-level measures against targets, usually reviewed weekly or monthly.In Designing a dashboard for decisions
strategic themes: A broad area of the strategy, such as partner platform or data, with an outcome measure, an owner and a few actions. Teams plan their work under a theme.In Technology strategy
stratified sample: Splitting the population into groups, such as cities, and drawing a random sample inside each group.In Samples, uncertainty and confidence intervals
stratified sampling: Sampling a fixed number from each group separately, so small groups are not drowned out by large ones.In Evals: knowing if it got better
streak: A count of consecutive days or weeks in which a user did something, shown so they want to keep it going.In Engagement loops, habits and notifications without dark patterns
Stream-aligned: A team that owns one flow of work for a part of the business, such as one product or one user journey, end to end.In Team design and delivery health
stream-aligned teams: A team that owns one flow of work for a part of the business end to end, from idea to running it in production.In Organisation and operating model
streaming: Sending the model's answer to the screen piece by piece as it is generated, so the first words appear quickly.In Guardrails and red-teaming, Serving: batch and online
streaming replication: Replication where the primary sends WAL records to each replica over a network connection as soon as they are written, instead of in whole files.In Replication and read replicas
stretch assignment: A piece of real work slightly beyond what a person has done before, given on purpose so they grow, with support around it.In Delegation and mentoring, Performance and growth
strict mode: A stricter mode of JavaScript that turns some silent mistakes into errors. ES modules and class bodies always use it.In Functions, scope and closures, Guardrails and red-teaming
STRIDE: A threat checklist developed at Microsoft in the late 1990s: Spoofing, Tampering, Repudiation, Information disclosure, Denial of service, Elevation of privilege.In Threat modelling a feature
strong-style pairing: A way of pairing where every idea must go through the other person's hands: the navigator explains, the driver types.In Delegation and mentoring
structural sharing: When a new version of a data tree reuses the unchanged branches of the old version instead of copying them, so only the changed path costs memory.In Objects, arrays and references
structured interview: An interview where every candidate for a role gets the same questions, in the same order, scored against the same written scale.In Hiring well
structured log: A log written as data, usually one JSON object per line, with named fields instead of a free-text sentence.In Observability: logs, metrics and traces
Structured output: Model output that follows a given schema, usually JSON, so code can use it safely. Structured output
sub-resource: A resource that lives inside another one and is reached through it, such as the lines of one order.In Designing an API people can use
subject experts: People with deep knowledge of one system, such as the payment service or PostgreSQL, called in to investigate or change it during an incident.In Leading an incident
subnet mask: Another way to write the network part: 1 bits for the network, 0 bits for devices. /24 is 255.255.255.0.In Networks from the till to the data centre
subnets: A smaller block of addresses inside a VPC. A public subnet can be reached from the internet; a private subnet cannot.In Networking you actually need
subquery: A query written inside another query, in brackets. It can return one value, a list, or only whether any row exists.In SQL you'll actually use
success criteria: One testable rule in WCAG, such as 1.4.3 Contrast (Minimum). Each has a number, a name and a level.In Accessibility, devices and the real world
supertest: A Node.js library that sends HTTP requests to an app in the same process and lets you check status, headers and body.In Testing and documenting the API, Testing what matters
supervised learning: Learning from past examples where both the inputs and the correct outcome are known, so the model can predict the outcome for new inputs.In Framing a problem as a model
Supply chain attack: Getting into your systems through something you depend on: a package, a CI action, a vendor. Supply chain
supply-chain attack: An attack that gets into your software through something you depend on, such as a package, a build tool, a CI action or a stolen pipeline credential, instead of attacking your running app directly.In Secure supply chain and delivery metrics
surge pricing: Raising the price in an area when riders far outnumber available drivers, to bring more drivers in and balance demand.In Uber: matching riders and drivers in real time
surrogate key: A key with no business meaning, created only to identify the row, such as an auto-increment number or a UUID.In Modeling data in tables, From questions to a reporting model
Survivorship bias: Drawing conclusions only from the things that made it through some filter, such as shops still open, and ignoring the ones that dropped out.In Traps that fool smart people
Suspense: A React component that shows a fallback, such as a skeleton, while anything inside it is still waiting for data, then swaps in the real content.In Loading server data
Swimlane: A band in a process diagram that holds the steps of one role or system. Process models
swimlane map: A process map with one lane per team or system, showing each step, each handoff between lanes, and how long work waits at each point.In Product management for internal tools
switchback experiments: Experiments that randomise by region and time window instead of by user, so everything in one region for one time slot gets the same version.In Experiments: A/B tests, holdouts and store pilots
switching value: The value an input would need to reach, with everything else at base, for the NPV to fall to zero.In Business cases, Build, buy or outsource
switchover: A planned, controlled change of primary: the old primary stops cleanly, a replica catches up fully and is promoted, and nothing committed is lost.In Replication and failover
symbolic ref: A ref that points to another ref instead of to an object. HEAD is usually symbolic: ref: refs/heads/main.In Inside the .git folder: objects, refs and HEAD
Symmetric encryption: Encryption where the same secret key both locks and unlocks the data. Fast, and used for almost all bulk data.In Cryptography without the maths
symptom: Something users notice directly, such as failed orders, slow answers or stale data.In Alerting that does not burn people out
SYN-SENT: The TCP state of a connection that has sent its opening SYN and is still waiting for an answer.In Networks from the till to the data centre
synchronous: The primary waits for a named replica to confirm it received the WAL before it tells the client the COMMIT succeeded.In Replication and read replicas, Queues, background jobs and pub/sub, Running a design review
Synthesis: Turning many separate research observations into a few patterns you can act on, each with evidence you can trace.In From research to opportunities
synthetic users: AI-generated stand-ins for research participants that answer questions as if they were real users.In From research to opportunities
system catalogs: The tables PostgreSQL uses to describe your database: its tables, columns, types, functions and roles.In Upgrades, maintenance and on-call
system integration testing: System integration testing: the vendor tests the whole system together with the buyer's other systems, such as the ERP, in a test environment.In Working with vendors and agencies
system of record: The system that holds the authoritative version of a piece of data; when systems disagree, this one is right by definition.In Platforms, integrations and systems of record, Integration and data ownership, Evolving a legacy system, Application portfolio, Target state and roadmap
system prompt: The fixed instructions sent at the start of every call that tell the model its job, its rules and how to answer.In How AI features work, just enough for PMs, Prompts and structured output, Security in the pipeline and in AI features
systemd: The first program Linux starts at boot (PID 1). It starts, stops, watches and restarts every service.In Linux from the command line
systemd timer: A systemd unit that starts another unit, usually a service, on a calendar time or after a delay.In Shell scripting for daily jobs
systemd unit: A file that tells systemd how to run and restart a service, timer or mount. Linux CLI

T

t distribution: A bell-shaped distribution with fatter tails than the normal one, used for intervals when the standard deviation is estimated from a small sample; it depends on the degrees of freedom, n minus 1 for a mean.In Samples, uncertainty and confidence intervals
T1071, Application Layer Protocol: MITRE ATT&CK technique T1071: using normal application-layer protocols such as HTTPS to hide command-and-control traffic.In Threat hunting
tables: A grid in a database, like one sheet of a spreadsheet: each row is one record, such as one order, and each column is one field.In A spec the AI can follow, Data and security basics
tabletop exercise: A discussion-based rehearsal where the people who would handle an incident talk through a realistic scenario, step by step, without touching real systems.In Risk, compliance and the first hour of an incident
Tabs: A row of buttons, usually at the bottom of the screen, that switches between top-level sections of an app that live side by side.In Navigation and app structure
Tacit knowledge: Know-how people use without thinking about it, so they rarely mention it when asked to describe their work.In Watching the work
Tactic: A design technique that improves one quality attribute, such as caching for performance. Quality attributes
tag: A Git ref that names one commit permanently, usually a version such as v1.4.0. An annotated tag also stores a message, the tagger and a date.In Branching strategies, releases and versions, Cloud and SaaS costs with FinOps
tagged template: A JavaScript function called with a template literal, as in fn`...${x}...`. The function receives the fixed text and the values separately.In SQL you'll actually use, Data with PostgreSQL and Prisma
tail: The rare, slowest requests at the far right of a latency chart. Also called tail latency.In Measuring speed and reliability: p99, SLOs and nines
Tail sampling: Deciding whether to keep a trace after all its spans have arrived, so the rule can look at errors, latency and attributes.In Observability for LLM systems, Observability with OpenTelemetry
take rate: The share of the money flowing through a marketplace that the platform keeps as revenue: platform revenue divided by the total value of transactions.In Marketplaces, platforms and moats
Tamper-evident: Built so that any later change or deletion can be detected, even if it can't be prevented.In Shipping an agent safely
target architecture: A description of the architecture you want at the end of the planning period, designed to meet the goals.In Target state and roadmap
Target encoding: Replacing a category such as a store id with the average label for that category, for example that store's average daily sales.In Features and leakage
target tracking: An autoscaling rule that adds or removes copies to keep one metric, such as average CPU, close to a target value.In Scaling up vs scaling out
task queue: Also called the macrotask queue: callbacks from timers, I/O, network responses and user events, waiting to run. The event loop runs one task per turn.In The event loop
task-relevant maturity: Andy Grove's term for how ready a person is to do one specific task well on their own, based on their experience with it and their drive to take responsibility.In Delegation and mentoring
TCO: Total cost of ownership: every cost of an option over its life, not just the price. Build or buy
TCP: Transmission Control Protocol: a way to send data that guarantees every byte arrives, in order, or the sender is told it failed.In What happens when you open an app or website, Networking you actually need
TDE: Transparent data encryption: the database encrypts its data files itself, so files copied off the disk cannot be read without the key.In Security and access
team charter: A one-page document a team writes about itself: its purpose, what it owns, who it serves, how to send it work and how it works together.In The manager's job
teardown: A structured breakdown of a real product to understand the choices behind it: users, loop, money, metrics, growth and defences.In How to break down any product
Tech debt: Shortcuts and old design choices in the code that make future changes slower or riskier, and the work to fix them.In Outcome roadmaps
Tech lead: The engineer accountable for a team's technical direction and quality, usually still writing code. Tech lead role
Technical debt: The future cost of shortcuts and outdated design in a codebase: each later change takes longer and breaks more often until the shortcut is fixed.In Working with engineers
Technical fit: How healthy an application is as technology: vendor support, available skills, security, ease of change, reliability and ability to connect to other systems.In Application portfolio
technical vision: A short document that says where a team's technical work is heading over a fixed period, and the rules it will use to make choices on the way.In Technical vision and design docs
Technology Radar: A picture of technology choices placed in four rings (adopt, trial, assess, hold) that shows teams what to use, what to test and what to avoid for new work.In Technology strategy
Temperature: A sampling setting that sharpens or flattens the next-token probabilities. Low values make the likely token win almost every time; high values give unlikely tokens more chances.In How LLM APIs work, Designing AI features: LLMs, RAG and vector search
Template literals: A string written with backticks that can contain ${expression} placeholders and line breaks.In JavaScript in one sitting
temporal coupling: Two parts are temporally coupled when both must be running at the same moment for the work to succeed.In Integration and data ownership
temporal dead zone: Temporal dead zone: the time between the start of a scope and the line that declares a let, const or class. Touching the name in that window throws a ReferenceError.In Functions, scope and closures
tenant: One customer organisation whose data must be kept apart from others in a shared system; here, one partner shop.In Broken access control
terminal: The window that shows text and sends your key presses to a shell, such as Windows Terminal, macOS Terminal or VS Code's terminal panel.In The terminal without fear, What AI coding tools can and cannot do
Terraform state: Terraform's record of which real resources match which code; kept remote, locked and never edited by hand. IaC
test: A small piece of code that runs part of your app with fixed inputs and checks the output is exactly what you expected. It prints pass or fail.In Building in small steps
test basis: Everything you design tests from: stories, acceptance criteria, designs, rules and past bugs.In What testing is for, Designing tests that find bugs, Bugs, reports and acceptance
test card: A four-line plan for an experiment: the hypothesis, the test, what you will measure, and the success criteria.In Testing ideas cheaply
test case: A test with its inputs, preconditions and expected result, written so anyone can run it and say pass or fail.In Designing tests that find bugs
test data builder: A small helper object with sensible defaults and chainable methods, such as anOrder().asStaff().build(), that creates fresh test data for each test.In Unit and API tests
test double: Any object that stands in for a real dependency during a test. Stubs, spies, mocks and fakes are kinds of test doubles.In Unit and API tests
test levels: A group of test activities aimed at one stage of the software, from single components up to the whole system and its acceptance.In What testing is for, A test strategy, not a pile of scripts
test oracle: The way a test decides whether a result is right, such as an expected number, a status code or a rule.In A test strategy, not a pile of scripts
test period: A final period of data that no model, setting or decision has been tuned on, scored once at the end to estimate real-world error.In Training and comparing models
test pyramid: A picture of a healthy test suite: many fast unit tests at the bottom, fewer service tests in the middle, and few slow UI tests at the top.In A test strategy, not a pile of scripts
test scenarios: A short story of a real task from start to finish, such as a Monday reorder that goes over the credit limit, used to test whether the solution supports the work.In From change control to acceptance
Testcontainers: A library that starts real services such as PostgreSQL in throwaway Docker containers for the length of a test run.In Unit and API tests
text files: A file that stores plain characters, such as code, JSON or Markdown. You can open it in any editor and read it.In How code becomes a running program
tf-idf: Term frequency times inverse document frequency: a score that rises when a word appears often in this page but rarely in all pages.In Google Search: crawling, indexing and answering in a fraction of a second
the bridge: The old asynchronous channel between JavaScript and native code in React Native, which passed batches of messages encoded as JSON.In How React Native works
The EU AI Act: Regulation (EU) 2024/1689, the European Union's law on artificial intelligence. It sorts AI uses by risk and sets duties for the companies that build and use them.In Governance and accountability
theirs: In a merge, the other side: the branch you named in git merge. VS Code calls it the incoming change.In Merge conflicts without panic
thin vertical slice: A first version that does the whole job end to end in a basic way, through every part of the app: screen, API and database.In From idea to one-page spec
third story: A neutral description of the difference between two people's views, the way a fair observer would tell it.In Managing up and across
thousands separator: The character that groups thousands in a number: a comma in 1,204.50, a full stop in 1.204,50, a space in some countries.In Cleaning data in a spreadsheet
threat: Anything that could cause harm to an asset: a person or group (a threat actor) or an event such as a flood or a power cut.In Thinking like an attacker
threat actor: A person or group who might attack an organisation, such as a criminal gang, an opportunist, an insider or a fraudster.In Thinking like an attacker
threat model: A written list of the ways a system can be attacked or misused, how much harm each could do, how likely it is, and which control is meant to stop it.In Guardrails and red-teaming
Threat modelling: A structured way to find what could go wrong with a system, decide what to do about it, and check the work, done while you design it.In Threat modelling a feature
Three amigos: A short talk about one piece of work between the business side (BA or product owner), a developer and a tester, before building starts.In What business analysts do, User stories and acceptance criteria
Three Lines Model: A model from the Institute of Internal Auditors (2020) that describes how a governing body, management and internal audit share the work of governance and risk management.In Risk, security and compliance
Threshold: The score above which you act; choose it from the cost of each error. Churn
throttled: When a process has used its CPU allowance for the current period, the kernel pauses it until the next period starts. Requests get slower but nothing crashes.In Containers from the inside
Throughput: How many requests the system finishes per unit of time, for example 500 requests per second.In Speed numbers and back-of-the-envelope math, Working with engineers, Slicing and planning work
throws: Stop the normal flow and hand an error object up the call stack until some code catches it, or the program crashes.In Errors, debugging and testing
tidy: A table where each variable is a column, each observation is a row, and each kind of thing is its own table. The idea comes from Hadley Wickham's 2014 paper Tidy Data.In Cleaning data in a spreadsheet
Tidy data: One variable per column, one observation per row, one value per cell. Python and pandas
tie-breaker: A second sort column, usually the unique id, used when the first one has equal values, so the order of rows is always the same.In Designing an API people can use
Tier 0: The highest tier in the admin model: the systems that control identity itself, such as domain controllers and the accounts that administer them.In Privilege escalation and lateral movement
Time and materials: Paying for the hours or days worked at agreed rates, plus any materials. The buyer carries the risk that the work takes longer.In Build, buy or outsource
TIME model: A portfolio model credited to Gartner that sorts applications by business (functional) fit and technical fit into Tolerate, Invest, Migrate and Eliminate.In Application portfolio
time series: A sequence of values over time for one thing, such as daily units of one product in one store.In Forecasting demand, Observability with OpenTelemetry
Time to detect: Time from the start of customer impact to the moment the team knew about it, by an alert or a report.In Leading an incident, Postmortems and learning
time to first token: Time to first token: how long until the first piece of the answer appears.In ChatGPT: serving a large language model to hundreds of millions, How AI features work, just enough for PMs, How LLM APIs work, Cost, latency and safety, Designing AI features: LLMs, RAG and vector search, Observability for LLM systems
Time to mitigate: Time from the start of customer impact until customers were no longer affected.In Leading an incident, Postmortems and learning
Time to resolve: Time from the start of customer impact until the incident was declared resolved and the response ended.In Leading an incident
Time to value: How long it takes a new user to get the product's core value for the first time, usually measured as a median from sign-up.In Funnels, activation and adoption, Onboarding and the first-time experience
time travel: A warehouse feature that lets you query a table as it was at an earlier time, for a limited retention period.In Experiments, versions and the model registry
time value of money: The idea that money now is worth more than the same amount later, because it can earn a return in the meantime and the future is uncertain.In Business cases
timeline: A numbered branch of WAL history. Each time recovery ends and the server opens for writes, it starts a new timeline so new WAL never mixes with the old history.In Backups you have actually restored, Replication and failover
timeout: A limit on how long you wait for an operation. When it passes, you stop waiting and treat the call as failed.In Failure is normal: timeouts, retries, backoff and circuit breakers, Screens and flows
tip: The newest commit on a branch, the one the branch name currently points to.In Branches and merging
TLS: Transport Layer Security: the encryption behind HTTPS. It proves the server is who it claims to be and encrypts everything sent.In What happens when you open an app or website, Cryptography without the maths
to-be model: A model of how the work should run after the change, used to agree the design and derive requirements.In Modelling processes
TOAST: A side table where Postgres keeps very large values, such as long text, outside the main table.In Speed numbers and back-of-the-envelope math
todo list: The text file that interactive rebase opens in your editor: one line per commit, oldest first, each with a command such as pick, reword, squash, fixup or drop.In Rewriting history safely: rebase, squash, cherry-pick, stash
TOGAF: An open standard for enterprise architecture from The Open Group, with a method, content guidance and governance advice. The current version is the TOGAF Standard, 10th Edition, from 2022.In Frameworks and governance
toil: Work tied to running a production service that is manual, repetitive, automatable, tactical, has no enduring value, and grows linearly as the service grows.In Toil and automation
token: The unit a model reads and writes: a word, part of a word or a symbol. One English word is roughly 1.3 tokens.In ChatGPT: serving a large language model to hundreds of millions, Your first NestJS service, Working with AI coding assistants, How AI features work, just enough for PMs, Data and security basics, How LLM APIs work
Token bucket: A rate-limiting method: tokens refill at a steady rate, each request spends one, and bursts are allowed up to the bucket size. Rate limiting
tokenizer: The fixed table and rules a model uses to split text into tokens. Different model families use different tokenizers, so the same text can give different counts.In How LLM APIs work
tombstone: A marker that a record was deleted, kept so that sync can tell other devices to remove their copies. Usually a deleted_at column instead of a real DELETE.In Offline-first apps and data sync
Tool calling: A way for a model to ask your code to run a named function, such as reading a GRN; your code checks permissions, runs it and returns the result.In How AI features work, just enough for PMs, Tools and the Model Context Protocol, Designing AI features: LLMs, RAG and vector search
tool registry: A list your code keeps of the functions the agent may use, each with a name, a description for the model, an argument schema and the code that runs it.In What makes an agent
Top-p: Nucleus sampling: keep only the most likely tokens whose probabilities add up to p, such as 0.9, and sample from those.In How LLM APIs work
topological order: An ordering of items in which every item comes after all the items it depends on.In Target state and roadmap
tornado chart: A chart of horizontal bars, one per input, showing how far the result moves when that input changes; sorted with the biggest bar on top, so it looks like a tornado.In Business cases a managing director will sign, Business cases
Total cost of ownership: Everything an option costs over its useful life, including one-time costs, recurring costs, your own staff time and the cost of leaving, not only the price on the quote.In Build, buy or partner, Application portfolio, Build, buy or outsource, Money: budgets, headcount and vendors
Total cost of ownership (TCO): All costs of a system over its life: licences, cloud, people, support and exit. Build or buy
TOTP: Time-based one-time password: a 6-digit code computed from a shared secret and the current time, changing every 30 seconds.In Accounts, SSO and MFA, Identity: passwords, MFA and passkeys
Touch time: Time when someone is actively working on the item.In Product management for internal tools, Modelling processes
TPUs: Tensor Processing Unit: a chip designed by Google specifically for the matrix maths in machine learning.In Claude: how Anthropic serves Claude at scale
trace: The record of one request's whole journey through your system, made of timed steps called spans that share one trace id.In Observability: logs, metrics and traces, Evals: measuring AI quality, What makes an agent, Shipping an agent safely, Observability for LLM systems, End-to-end tests with Playwright
traceability: Being able to follow a link from any piece of work to the goal it serves, and from any goal down to the work that delivers it.In Why enterprise architecture exists, Requirements that work, User stories and acceptance criteria, From change control to acceptance
traceability matrix: A table with one row per requirement and columns for its source, its stories, its tests and its status, used to check coverage and the impact of changes.In From change control to acceptance
traceparent: The standard HTTP header that carries trace context between services: version, trace id, parent span id and flags.In Observability with OpenTelemetry
tracked: A file Git knows about because it is in the last commit or the staging area. Git watches tracked files for changes.In What Git really is: snapshots and the three areas
tracking plan: The agreed list of every event a product records: its name, exactly when it fires, its properties and their allowed values, where it is sent from, who owns it and which metric it serves.In Instrumentation: tracking plans and data you can trust
trade-off: A choice where getting more of one quality means accepting less of another, for example faster answers in exchange for slightly older data.In What system design really is
trade-off point: In ATAM, a property of the design that affects two or more quality attributes in opposite directions, so improving one makes another worse.In Choosing an architecture style
training-serving skew: A difference between how features were computed for training and how they are computed when the model runs for real.In Features and leakage
Training/serving skew: A difference between how a feature was computed for training and how it is computed when the model makes real predictions.In Data and feature pipelines
trajectory: The ordered list of steps an agent took in one run: every tool call with its arguments and result. Also called the trace or transcript.In Evaluating agents
Trajectory checks: A rule checked against the trace of a run, such as a tool that must be called, a tool that must never be called, or an approval that must come before a risky call.In Evaluating agents
transaction: A group of statements that the database treats as one unit: all of them take effect, or none of them do.In Transactions, ACID, isolation and locks, Data with PostgreSQL and Prisma, Data model and API, Offline first
transaction fact table: A fact table with one row per event, written when the event happens and never changed afterwards.In From questions to a reporting model
transactional outbox: Saving outgoing events in a database table in the same transaction as the business change, and publishing them from that table afterwards.In Events, the outbox pattern and sagas, Background jobs and idempotency
transition architectures: A description of the architecture at a planned point between the baseline and the target, stable enough to run the business on for a while.In Target state and roadmap
transitional architecture: Code and infrastructure you build only so that the old and new systems can run side by side during a migration, and that you plan to remove later.In Evolving a legacy system
transitive dependencies: A package you never asked for directly: it is installed because one of your dependencies, or one of theirs, needs it.In Supply chain and secrets
Tree-shaking: A bundler removing exports that no code imports, so they are not shipped in the final bundle.In Prototypes, classes and modules
Trend: A slow, steady rise or fall in the level over weeks or months.In Forecasting demand
Triage: Sorting new bug reports to decide what happens to each one, how urgent it is and who owns it.In Bugs, reports and acceptance, Triage and investigation
triangulation: Checking a finding against at least two other kinds of source, such as what people say, what you saw them do and what the records show.In Stakeholders and elicitation
Trunk-based development: A branching strategy where everyone merges small changes into the main branch (the trunk) at least daily, hiding unfinished work behind feature flags.In Branching strategies, releases and versions
trust boundaries: A line on the diagram where data moves between parts with different levels of trust or different owners, such as the internet and your API.In Threat modelling a feature
Trust boundary: A line in a data-flow diagram where data passes between parts with different levels of trust, such as the internet and your API. Threat modelling
truthy: Any value that counts as true in a condition: every value that is not one of the eight falsy values.In JavaScript in one sitting
tsvector: A Postgres type holding the normalised words of a text with their positions, ready for fast full-text matching.In Beyond SQL: choosing the right data store, Retrieval-augmented generation
TTID: Time to initial display: how long Android takes from launching the app to drawing its first frame. Android prints it in logcat as a Displayed line.In Fast on a cheap phone
TTL: Time to live: how many seconds a cached DNS answer may be reused before it must be looked up again.In What happens when you open an app or website, Caching
TTL (time to live): How long a cached value or DNS record is kept before it must be fetched again. Caching
tuple routing: The step where Postgres sends a new row from the parent table into the partition whose range matches the row's key value.In Partitioning big tables
TurboModules: The New Architecture's native module system: modules are loaded only when first used and called through JSI with a typed contract generated by Codegen.In How React Native works
twelve-factor app: A widely used set of 12 rules for building apps that run well in the cloud, such as keeping config in environment variables and treating processes as stateless.In Scaling up vs scaling out
two-pizza teams: Amazon's rule that a team should be small enough to feed with two pizzas, usually under 10 people, and own one service or product fully.In Amazon: the store that never closes
two-sided: A test that counts surprises in both directions, a rise or a fall, because a change could help or hurt.In Hypothesis tests without the fog
two-way door: A decision that is easy and cheap to reverse, so it can be made quickly by one person or a small group.In Writing for decisions
type coercion: JavaScript automatically converting a value from one type to another, for example the string '5' to the number 5, during an operation.In JavaScript in one sitting
Type I and Type II error: A false alarm, and a missed real effect. Hypothesis tests
type I error: Rejecting the null when it is true: declaring a change that is not there. Its rate is alpha.In Hypothesis tests without the fog
type II error: Failing to reject the null when a real effect exists: missing a change that is there. Its rate is called beta.In Hypothesis tests without the fog
Type narrowing: TypeScript working out a more specific type after a check, such as if (r.ok) on a union. TypeScript for a JavaScript team
type stripping: A Node.js feature (on by default from Node 22.18 and 23.6) that removes TypeScript type annotations as it loads a .ts file, so it can run without a separate compile step. It does not check the types.In Project: ship a small app end to end
typosquatted: Publishing a package whose name is one typo or one small change away from a popular one, hoping someone installs it by mistake.In Supply chain and secrets
Typosquatting: Publishing a malicious package with a name very close to a popular one, hoping someone mistypes it or confuses the two.In Security and quality in JavaScript projects

U

UAT: User acceptance testing: real users check that the solution lets them do their work before it goes live.In What business analysts do
UDP: User Datagram Protocol: sends single messages with no handshake, no confirmation and no ordering.In Networking you actually need
UEM: Unified endpoint management: an MDM that also manages laptops and desktop PCs, so one console covers every kind of device.In Looking after 1,400 devices
UI thread: The main native thread of the app. It lays out and draws native views, handles touches and runs native scrolling and native animations.In Fast on a cheap phone
ULID: A 128-bit id like a UUID whose first part is a timestamp, so ids sort roughly by creation time. UUID version 7 has the same property.In Offline-first apps and data sync, Case study: syncing POS sales from hundreds of stores
Unbundling: Pulling one job out of a general tool that does many, and building a specialist product that does that one job much better.In Idea lab: finding new software and feature ideas
Underfitting: When a model is too simple to capture the pattern, so its error is high on training and validation data alike.In Training and comparing models
Union types: A type that allows one of several options, written with a vertical bar: string | undefined, or 'cash' | 'card'.In TypeScript for a JavaScript team
unit: The single repeatable thing a business sells or serves, such as one order, one ride, one subscriber-month or one delivery, used to check whether the business makes money at the smallest scale.In Unit economics: does each order make money?
unit cost: Cloud cost divided by a business unit, such as cost per 1,000 orders, so cost can be compared with growth.In Cloud architecture and cost, Cloud and SaaS costs with FinOps
Unit economics: Revenue and costs for one unit, such as an order or a customer. Unit economics
unit file: A small text file that tells systemd how to run one thing: which program, as which user, when, and what to do if it fails.In Linux from the command line
unit of randomisation: The thing that gets randomly assigned to control or variant: a customer, a session, a store or a time slot.In Designing and reading an A/B test
unit test: Code that runs a small piece of your code with known inputs and checks that the result is what you expect.In Errors, debugging and testing, Testing what matters
universal links: Apple's verified links: https URLs on a domain you own that open your app directly once iOS has checked a file on that domain.In Navigation and app structure
Unpivot (melt): Turning month columns into rows of month and value, so tools can group them. Python and pandas
unpivoting: Turning a set of columns, such as one column per week, into rows with two columns: one for the old header (the week) and one for the value.In Cleaning data in a spreadsheet
upcasters: A function that converts data written in an old format into the current format, so the rest of the code only deals with one version.In Case study: syncing POS sales from hundreds of stores
update window: A fixed time range, usually at night, when devices are allowed to download and install updates and restart.In Looking after 1,400 devices
updater function: A function you give to a set function, such as n => n + 1. React calls it with the latest queued value, so several updates in a row add up.In State, effects and where data should live
uplift: The change in an outcome caused by an action, such as how much a coupon lowers a customer's chance of churning. Measured by comparing a treated group with a random untreated group.In Classification and churn
upsert: Insert a row, or update it if a row with the same unique key already exists, in one statement.In SQL you'll actually use
upstream: The remote-tracking branch your local branch is linked to, for example origin/main for main. git status compares the two, and plain git pull and git push use it.In GitHub: remotes, clone, push, pull and authentication
uptime check: A service that opens your app's address every few minutes and alerts you when it does not answer.In Ship it and know your limits
Urgency: How fast the damage grows if nobody acts: work stopped now, slowed with a workaround, or able to wait.In Incidents, problems and changes
usability heuristic: A rule of thumb used to judge a design, such as keeping users informed of what the system is doing.In Design for PMs
usability test: A session where real users try to do real tasks with a design while you watch, to find where they struggle.In Testing ideas cheaply
usage tiers: Levels of API access with higher limits; an organisation moves up as its usage history grows.In Claude: how Anthropic serves Claude at scale
USE: Utilization, Saturation, Errors: the three numbers to track for every resource such as CPU, disk, memory or a connection pool.In Observability: logs, metrics and traces
Use case: A description of how an actor reaches a goal with the system, with main and alternate flows. Requirements
USE method: A checklist by Brendan Gregg: for every resource, check its utilisation (how busy), saturation (how much work is waiting) and errors.In Capacity, resilience and game days
useful life: The number of years a company expects to use an asset; its cost is spread over that period.In Budgets, capex and opex
User acceptance testing: User acceptance testing: the buyer's real users test the delivered software against the agreed acceptance criteria before it goes live or is paid for.In Working with vendors and agencies, From change control to acceptance
user acceptance testing (UAT): User acceptance testing: the intended users try the system in conditions close to real use, to decide whether it meets their needs.In Bugs, reports and acceptance
user flow: The path one user takes through an app to finish one task, written as steps: the screen they see, the action they take and what the app does in reply.In Screens and flows
user journey: A short sequence of steps a user takes to get one thing done, such as placing an order or paying for it.In How reliable is reliable enough?
user story: A short description of something a specific user needs and why, used as the unit of planning and delivery.In User stories and acceptance criteria, User stories and acceptance criteria
UTC: Coordinated Universal Time, the reference clock that local time zones are defined from, for example UTC+6 is six hours ahead.In Analysing with an AI assistant
Utilisation: The share of what you committed to that you actually used. Unused commitment is paid for anyway.In Cloud and SaaS costs with FinOps
utility tree: A tree with utility at the root, quality attributes and their refinements as branches, and concrete scenarios as leaves, each rated for business importance and difficulty.In Quality attributes and constraints, Decisions, diagrams and reviews
UUID: Universally unique identifier: a 128-bit random value like 7f3c9a1e-... that is practically never generated twice.In APIs done right, Modeling data in tables, Offline-first apps and data sync

V

V8: The JavaScript engine made by Google. It runs JavaScript in Chrome, in Node.js and in Edge.In Performance and memory in JavaScript
VACUUM: The Postgres maintenance command that finds dead row versions, marks their space reusable, updates the visibility map and freezes old transaction IDs.In Running Postgres in production, Inside a PostgreSQL server
Validation: Checking that a product meets the needs of its users and other stakeholders in its real environment: did we build the right thing?In What testing is for
validation gate: A pipeline step that runs checks on the data and decides whether the run continues, continues without some data, or stops.In Data and feature pipelines
value item: The piece of value a single stage produces, such as a confirmed order or an accurate invoice.In Value streams and processes
value metric: The unit your price scales with, such as per user, per store, per truck, per order or per active member. A good one grows as the customer gets more value.In Pricing and packaging
value proposition: The benefit the stakeholder gets when the stream finishes, stated from their point of view.In Value streams and processes
Value streams: An end-to-end series of stages that creates an outcome for a stakeholder, such as a partner shop getting its stock and Mangrove getting paid.In Why enterprise architecture exists, Value streams and processes
Value-based pricing: Setting a price from the value the customer gets and the price of their next-best alternative, rather than from your own costs.In Pricing and packaging
vanity metric: A number that looks good and usually only goes up, such as total sign-ups or downloads, but does not show whether anyone is getting value.In North star, input and guardrail metrics
variable: A name that holds a value, such as a number, a piece of text or a list, so the code can use it later.In Reading and fixing what the AI wrote
Variable costs: A cost that rises and falls with each unit sold or served, such as the goods in a basket, the rider for that trip or the payment fee.In Unit economics
variance: The difference between actual and budget. Spending above budget is adverse; below budget is favourable.In Budgets, capex and opex
VAT: Value added tax, or sales tax: money the shop collects from customers on behalf of the government. It is not the shop's income.In Metrics everyone agrees on
vector clock: A small list of (node, counter) pairs attached to each version. Comparing two clocks tells you if one version came after the other or if they are in conflict.In Amazon: the store that never closes
Vector database: A store that finds the embeddings nearest to a query, such as pgvector in Postgres. RAG
vector search: Finding the stored vectors closest to a query vector, so you find text with similar meaning instead of the same words.In Designing AI features: LLMs, RAG and vector search
vendor scorecard: A regular, shared rating of a supplier's performance on agreed measures such as delivery, quality, support and communication, backed by evidence.In Working with vendors and agencies
Verification: Checking that a product meets its specified requirements: did we build it right?In What testing is for
Version control: A system that records every saved version of a project's files, with who changed what, when and why, so you can compare and return to any version.In What Git really is: snapshots and the three areas
version manager: A small tool that installs several Node.js versions side by side and switches between them per project, such as fnm, nvm or Volta.In How code becomes a running program, Node.js and npm
versioned filenames: Putting a version or a hash of the file's content in its name, such as app.3f9a2c.js, so a changed file gets a new URL.In Files, object storage and CDNs
Vertical scaling: Making one machine bigger: more CPU cores, more RAM, a faster disk. Also called scaling up.In Scaling up vs scaling out
vertical slice: A small story that goes through every layer (screen, API, back office) and gives a user something they can try, as opposed to a story for one technical layer.In User stories and acceptance criteria, Slicing and planning work
vertical split: Moving whole tables or features to their own database server, for example loyalty data on one server and sales on another.In Sharding and consistent hashing
vibe coding: Building software by prompting an AI, trying the result and asking for changes, without reading the code it writes.In Prototyping with AI, What AI coding tools can and cannot do
view: One actual picture or model of the architecture, made with a viewpoint, that addresses specific concerns of specific stakeholders.In Why enterprise architecture exists
viewpoint: The rules for making one kind of view: which concerns it addresses, for whom, and which notation and models it uses.In Why enterprise architecture exists
viral coefficient: The number of new users each user brings in per round of a viral loop: invites sent per user times the share of invites that convert. Above 1, growth sustains itself.In Growth loops and local go-to-market
virtual IP: An IP address that can move between two machines. If the active machine dies, the standby takes over the address within seconds.In Load balancers
virtual nodes: Many points on the hash ring for each real server, for example 100 points named A#0 to A#99, so each server owns many small arcs and the load evens out.In Sharding and consistent hashing
Virtualization: Rendering only the rows inside, and just around, the visible part of a scrolling list, with empty spacers standing in for the rest.In Accessible, fast pages, Fast on a cheap phone
Virtualized list: A list that renders only the rows on screen, such as React Native's FlatList. Mobile performance
visibility map: A map with bits per table page that say whether all rows on the page are visible to everyone. Index-only scans use it to skip reading the table.In Finding and fixing slow queries
visibility timeout: In Amazon SQS, the time a received message stays hidden from other consumers. If it is not deleted in that time, it becomes visible again and is delivered again.In Queues, background jobs and pub/sub
Vitest: A test runner for JavaScript and TypeScript that runs test files, reports failures and supports TypeScript without extra setup.In Testing what matters
VLANs: Virtual LAN: a way to run several separate networks over the same switches and cables, each with its own number.In Networks, endpoints and the cloud
VMAF: Video Multimethod Assessment Fusion: an open-source score from 0 to 100, developed by Netflix with university researchers, that predicts how good a video looks to people.In Netflix: streaming to every screen
volume check: A check that counts the rows that arrived, for example per store per day, and compares the count with what is usual for that store and day.In Fresh, tested, trusted
VP of engineering: An executive who runs the engineering organisation day to day: delivery, people, hiring and process. In companies with both roles, the VP of engineering often reports to the CTO or works beside them.In The department head's job
VPC: Virtual Private Cloud: your own private network inside a cloud provider, with private addresses that the internet cannot reach directly.In Networking you actually need
VPN: Virtual private network: an encrypted tunnel that puts a remote device on the company network as if it were in the office.In Networks, endpoints and the cloud
vulnerability: A weakness that a threat could use: a reused password, an unpatched server, a process with no second check.In Thinking like an attacker, Networks, endpoints and the cloud
vulnerability scanner: A tool that compares the packages in an image or SBOM against databases of known vulnerabilities (CVEs and security advisories) and reports matches with a severity.In Secure supply chain and delivery metrics, Networks, endpoints and the cloud, Attacking networks and services
VUs: Virtual user: one k6 worker that runs your test function in a loop or on demand. Each VU runs one iteration at a time.In Performance and load testing

W

WAF: Web application firewall: a filter in front of your app that inspects requests and blocks or flags ones matching known attack patterns.In Injection and unsafe input, Attacking web applications
waiting room: A queue in front of a busy page: visitors get a place in line and are let in at a steady rate, so the system behind it is never overloaded.In Case study: ordering without overselling stock
WAL: Write-ahead log: the journal where Postgres records every change before it changes the data files. Replicas and backups replay it.In Partitioning big tables, Replication and read replicas, Inside a PostgreSQL server
WAL (write-ahead log): Postgres's log of every change, written before the data files; used for crash recovery, replication and PITR. Replication
WAL archiving: Copying every finished 16 MB WAL file from the server to safe storage, so you keep a complete record of every change since the last base backup.In Backups and disaster recovery
WAL segment: One file of the write-ahead log. PostgreSQL writes the WAL in fixed-size files, 16 MB by default, and archives each one when it is full.In Backups you have actually restored
walking skeleton: The smallest version of a system that works end to end, with every main step present but each step kept very simple.In User stories and acceptance criteria, Slicing and planning work
WAPE: Weighted absolute percentage error: the sum of absolute forecast errors divided by the sum of actual values.In Framing a problem as a model, Features and leakage, Training and comparing models, Forecasting demand
war room: One shared space, in person or a call, where the incident is run so decisions and status are in one place.In Incident response
watchdog: A timer the operating system must keep resetting. If the process stops resetting it, for example because it hung, the machine reboots.In Replication and failover
watermark: The highest value of an ordered column, usually a load timestamp, that a pipeline has already processed. The next run reads rows above it.In Data and feature pipelines
waves: Groups of devices that receive an update one after another, smallest and safest group first. Some tools call them rings.In Looking after 1,400 devices
WCAG: Web Content Accessibility Guidelines, the W3C standard for accessible web content. Its success criteria are also widely used for mobile apps.In Accessibility, devices and the real world
WCAG 2.2: Web Content Accessibility Guidelines, the W3C standard for accessible web content. Version 2.2 is current; most laws and contracts ask for level AA.In Accessible, fast pages
Web content discovery: Probing a web application for paths, files, API endpoints and parameters that are present but not linked, by requesting likely names and watching the responses.In Reconnaissance
Web-first assertion: An assertion that retries until the page reaches the expected state or a timeout passes. Playwright E2E
WebAuthn: The W3C standard browser and platform API for signing in with public-key credentials such as passkeys and security keys.In Authentication, sessions and tokens
webhook: An HTTP request that another system sends to your server when something happens, such as a payment provider telling you a payment finished.In Case study: ordering without overselling stock
WebSocket: A long-lived, two-way connection that starts as an HTTP request and is then upgraded, so client and server can both send messages at any time.In Real-time updates: polling, WebSockets, SSE and push
weighted decision matrix: A table with options as rows and weighted criteria as columns; each option gets a score per criterion, and the weighted average ranks the options.In Choosing an architecture style
Welch's t-test: A t-test for two group means that does not assume the groups have the same spread; it is the safe default.In Hypothesis tests without the fog
whole-team approach: A way of working where everyone on the team is responsible for quality and anyone with the skills can do any task, including testing.In What testing is for
Willingness to pay: The highest price a customer would actually pay for a product or feature, which differs between customers and situations.In Pricing and packaging
Wilson interval: A confidence interval for a proportion that stays between 0 and 1 and behaves well with small samples and rare events.In Samples, uncertainty and confidence intervals
window function: A function that computes a value for each row from a set of related rows (its window), without collapsing the rows the way GROUP BY does.In SQL you'll actually use
Windows directory domain (Active Directory): A Microsoft directory service that stores accounts, computers and groups for a Windows network, and decides who may sign in where. Often shortened to AD.In Privilege escalation and lateral movement
wipe: A command that erases a managed device back to factory settings the next time it connects.In Looking after 1,400 devices
wireframe: A rough drawing of a screen made of plain boxes and labels, used to agree content and layout before building.In Designing a dashboard for decisions
Wizard of Oz test: A test where the user sees what looks like a working product while a person secretly does the work behind it.In Testing ideas cheaply
WMS: Warehouse management system: the software that runs a distribution centre, tracking stock, picking and loading.In Business cases
word splitting: Bash cutting the value of an unquoted variable into separate words at spaces, tabs and newlines.In Shell scripting for daily jobs
work packages: A defined piece of change work with a start, an end, an owner and a set of results it delivers.In Target state and roadmap
work sample: A task that reproduces a real piece of the job, done by every candidate in the same way and scored against the same anchors.In Hiring well
work_mem: The memory one sort or hash step may use before it spills to temporary files on disk. Default 4 MB.In Reading query plans and fixing slow queries, Running Postgres in production, Inside a PostgreSQL server
workaround: A way people get the job done without proper support from the product, such as a paper list, a side spreadsheet or a chat group.In Problems before solutions, Watching the work, Incidents, problems and changes
worker: A process that takes jobs from a queue and does the work. Also called a consumer.In Queues, background jobs and pub/sub, End-to-end tests with Playwright
worker thread: A separate JavaScript thread in Node (from the node:worker_threads module) with its own engine instance, memory and event loop. It communicates with the main thread by passing messages.In The event loop
workflow: A YAML file in .github/workflows that says when to run (on) and what to run (jobs). A repository can have many workflows.In GitHub Actions: continuous integration, AI agents in operations, What makes an agent
working agreement: A short written document in which people who share a team agree who decides, who is asked and who is told for the decisions that come up again and again.In Leading with product and management
working capital: The money tied up in day-to-day trading. Here: inventory plus receivables minus payables to suppliers.In The three statements
working folder: The normal project folder on your disk, where you edit files. Also called the working tree or working directory.In What Git really is: snapshots and the three areas
workload: A set of resources and code that together deliver one business capability, such as order checkout or the stock report.In Cloud architecture and cost
workload model: A written description of the traffic a test sends: which requests, at what rate, in what mix, with which data, and the assumptions behind each number.In Performance and load testing
Workspace settings: VS Code settings in a project's .vscode/settings.json; they override each person's user settings for that project. VS Code essentials
Workspace Trust: A VS Code safety mode for folders you have not trusted yet. It turns off tasks, debugging, some workspace settings and extensions that could run code.In VS Code essentials
workspaces: A feature of npm, pnpm and Yarn that manages several packages in one repo, with one install and one lockfile at the root.In Node.js and npm, Infrastructure as code

X

X-as-a-service: An interaction mode where one team provides something and the other uses it through a clear interface, with little need to talk.In Team design and delivery health
xmin horizon: The oldest transaction ID that any snapshot, replication slot or prepared transaction may still need. Dead versions ended after it must be kept.In Inside a PostgreSQL server
XPath: A query language for finding nodes in an HTML or XML document by their path, such as //div[3]/button.In End-to-end tests with Playwright
XSS: Cross-site scripting: an attack where someone gets their own JavaScript to run in other users' browsers, usually by sneaking HTML into data a page displays.In JavaScript in the browser, Injection and unsafe input
XSS (cross-site scripting): Getting attacker-controlled script to run in your page, usually because user text was inserted as HTML. Security and quality in JavaScript projects
XY problem: Asking for help with your attempted solution (X) instead of the real problem (Y), which leads helpers in the wrong direction.In Problems before solutions
XYZ hypothesis: A belief written as a measurable prediction: at least X percent of Y will do Z.In Testing ideas cheaply

Y

Y-statement: A one-sentence summary of a decision: in the context of a situation, facing a concern, we decided for an option and against others, to achieve a quality, accepting a downside.In Decisions, diagrams and reviews
Year over year (YoY): Comparing a period with the same period last year, which removes seasonality. Charts
Yoga: The C++ layout engine React Native uses to turn Flexbox styles into exact positions and sizes.In How React Native works
young generation: A small area of the heap for new objects. It is collected very often and very quickly, because most objects in it are already garbage.In Performance and memory in JavaScript

Z

z-score: How many standard deviations a value is from the mean: (value minus mean) divided by SD.In Describing data honestly
Zero trust: An approach where no device or network location is trusted by default; every request to every application is checked for who the user is, the device health and the context.In Networks, endpoints and the cloud
zero-touch enrolment: An Android Enterprise service where the reseller registers devices to the company, so they enrol themselves the first time they connect to the internet.In Looking after 1,400 devices
Zuul: Netflix's API gateway: the first service every request from a device hits. It checks, routes and filters requests to the right backend services.In Netflix: streaming to every screen

#

0-switch coverage: State transition coverage in which every single valid transition is exercised at least once.In Designing tests that find bugs
1:1: A regular private meeting between a manager and one person who reports to them.In 1:1s, feedback and coaching
3-2-1 rule: Keep at least 3 copies of your data, on 2 different kinds of storage, with 1 copy in another location.In Backups and disaster recovery
3-way handshake: The three opening messages of a TCP connection: SYN, SYN-ACK, ACK. It costs about one round trip.In What happens when you open an app or website, Networking you actually need
301: HTTP 301 Moved Permanently: the resource has moved for good. Browsers may remember it and skip the old address next time.In Case study: a URL shortener
302: HTTP 302 Found: a temporary redirect. The browser asks the original address again next time, unless cache headers say otherwise.In Case study: a URL shortener
5 whys: A method that asks why an event happened, then why that cause happened, and so on, usually about five times, to dig below the first answer.In Postmortems and learning
95% confidence interval: A range of plausible values for the true effect; if a test were repeated many times, about 95% of intervals built this way would contain the true effect.In Experiments: A/B tests, holdouts and store pilots