Session replay glossary

A checkout, in rrweb's sense of the word, is a scheduled re-baseline of a recording: the recorder takes a fresh full snapshot mid-session so the event stream contains periodic self-sufficient starting points instead of one ever-growing dependency chain.

A custom event is an application-defined entry in the rrweb event stream (type 5), added via `record.addCustomEvent(tag, payload)`, that timestamps domain context, such as "checkout-started" or "error-shown", alongside the recorded behavior.

Deserialization in session replay is the reverse of serialization: rebuilding real DOM nodes from the stored data structure so the page can be rendered again inside the player.

A DOM snapshot is a serialized copy of a web page's document object model at one moment: every element, attribute, and text node converted into a data structure that can be stored and later rebuilt into an identical page.

The event stream is the ordered sequence of timestamped events that makes up a session recording, from the opening full snapshot through every incremental change.

A full snapshot is the session replay event that captures the complete serialized DOM at a point in time, giving the replayer a self-sufficient starting state.

An incremental snapshot is a session replay event that records only what changed since the previous event, such as DOM mutations, mouse movement, scrolling, or input values.

The incremental source is the field inside an rrweb incremental snapshot that identifies what kind of change the event records, distinguishing DOM mutations from mouse movement, clicks, scrolls, viewport resizes, input changes, and other interaction types.

Live mode is replaying a session while it is still being recorded: events stream from the user's browser to a viewer with a short buffer delay, so support or engineering can watch a session in progress.

A meta event is the rrweb event (type 4) that records context about the page rather than its content: the URL, and the viewport width and height at recording start.

The mirror is rrweb's bidirectional mapping between the numeric node ids used in the event stream and the actual DOM nodes, maintained on both the recording and replay sides.

In session replay, the mutation observer is the mechanism that watches the page's DOM for changes such as added nodes, removed nodes, and attribute or text edits, so each change can be recorded as an event.

The replayer is the component that turns a stored event stream back into a watchable session: it rebuilds the initial full snapshot, then applies incremental events on a timeline with play, pause, seek, and speed controls.

rrdom is rrweb's virtual DOM implementation: a lightweight, browser-independent model of the document that can apply rrweb events without rendering anything.

An rrweb event is one entry in a recording's event stream: a JSON object with a numeric `type`, a type-specific `data` payload, and a millisecond `timestamp`.

Sandboxing in session replay is the isolation of the rebuilt page inside a restricted iframe so that nothing from the recorded session can execute during playback.

Serialization in session replay is the conversion of live browser state, primarily the DOM tree, into a plain data structure that can be stored, transmitted, and later rebuilt.

Session recording is the capture side of session replay: collecting a user's page state and interactions as structured events while they browse.

Session replay is the reconstruction of a user's visit to a website or application as a watchable playback, built from recorded browser events such as DOM changes, clicks, scrolls, and inputs.

Skip inactive is the playback feature that fast-forwards through stretches of a recording where nothing happens, such as a user reading, idling, or leaving the tab.

A dead click is a click that produces no observable response: no DOM change, no navigation, no network activity within a short window after the interaction.

Drop-off is the point at which users abandon a flow, page, or session, measured as the share who leave at each step rather than continuing.

An error click is a click that coincides with a JavaScript error or a failed request, connecting a user's action directly to a technical failure.

A frustration signal is any behavioral pattern that indicates a user is struggling, including rage clicks, dead clicks, error clicks, thrashing cursors, rapid back-and-forth navigation, and form field churn.

Funnel analysis measures how users progress through an ordered sequence of steps toward a goal, such as signup or checkout, and where they abandon it.

A heatmap is an aggregate visualization that overlays interaction density, such as clicks, moves, or scroll reach, onto a page screenshot or rendering, with warmer colors marking higher activity.

A rage click is a burst of repeated rapid clicks on the same element or spot, and it is the most widely used behavioral signal of user frustration.

Scroll depth is how far down a page users actually scroll, usually expressed as a percentage of page height or as reach past defined milestones.

Session segmentation is filtering and grouping recorded sessions by properties such as page, device, user cohort, presence of errors or frustration signals, funnel stage reached, or custom events.

A thrashing cursor is fast, erratic, high-amplitude mouse movement without productive interaction, such as circling, zigzagging, or shaking the pointer.

Consent mode is the practice of gating session recording on the user's consent state: recording starts, stops, or degrades to a minimal mode depending on what the visitor agreed to, usually via the site's consent management platform.

Data masking in session replay is the replacement of sensitive content with placeholders at record time, in the user's browser, so the real values never enter the recording.

Data residency is the requirement that data be stored and processed within a specific geography, such as keeping EU users' session recordings on EU infrastructure.

The GDPR governs session replay whenever recordings capture personal data of people in the EU, which in practice means nearly always.

PII, personally identifiable information, is any data that identifies a person directly or in combination, such as names, email addresses, phone numbers, government IDs, or precise location.

Redaction is the removal or obscuring of sensitive content from session data, either at capture time or from recordings that already exist.

A retention policy defines how long session recordings are kept before deletion, and under what rules, such as 30 days for routine sessions and 90 for error sessions.

Sampling in session replay means recording a subset rather than everything, and the word covers two distinct levels: session sampling, capturing only a percentage of sessions, and event sampling, throttling high-frequency event types like mouse moves within a recorded session.

The accessibility tree is the browser's structured representation of a page built for assistive technology: a hierarchy of elements with roles, names, and states, stripped of purely visual detail.

An action space is the set of operations an agent can perform on its environment; for web agents, typically click, type, select, scroll, navigate, and wait.

Agent observability is the practice of capturing what an AI agent did and why, in enough detail to debug, evaluate, and audit its behavior.

An agent trajectory is the complete recorded path of one agent run: the sequence of observations, reasoning steps, actions, and resulting states from task start to termination.

Behavioral anomaly detection flags sessions that deviate from learned patterns of normal use: impossible speed, inhuman input cadence, paths no legitimate user takes, or sudden population-level shifts.

Behavioral cloning is the supervised form of imitation learning: train a model to predict the expert's action from the state the expert saw, over a dataset of state-action pairs.

A behavioral substrate is the foundational data layer that captures how interfaces are used, by humans or by agents, in a form other systems build on.

A browser agent is an AI system that operates a web browser autonomously to complete tasks: navigating, reading pages, filling forms, and clicking through flows on behalf of a user or another system.

Computer use is the class of AI capability where a model operates a computer the way a person does: viewing the screen or interface state, moving a pointer, clicking, typing, and navigating applications to complete tasks.

Deterministic action replay is re-executing a recorded sequence of actions against a live or restored environment, as opposed to visual replay, which renders what was recorded without executing anything.

DOM grounding is the step where an agent's intended action, expressed in language or coordinates, gets mapped to an actual element in the page: "click the submit button" becomes a specific node.

An episode is reinforcement learning's unit of experience: one complete run through an environment from initial state to termination, containing the sequence of states, actions, and rewards.

Evals are structured evaluations of AI system behavior: defined tasks, graded outcomes, and comparisons across model versions, prompts, or configurations.

Imitation learning trains an agent from demonstrations of a task rather than from trial-and-error reward.

Intent inference is deriving what a user was trying to accomplish from what they observably did: classifying sessions by goal (compare plans, find an invoice, cancel) from navigation, interaction, and timing patterns.

An LLM session is the grouped record of a multi-step interaction with a language model, linking the individual requests, responses, and tool calls of one task or conversation into a single traceable unit.

A pixel-based agent perceives interfaces through screenshots, deciding actions from rendered pixels rather than from the DOM or accessibility tree.

Prompt injection is an attack where content an AI system processes contains instructions that hijack its behavior: a web page, email, or document tells the model to ignore its task and do something else.

Semantic session search is querying a session archive by meaning rather than metadata: "sessions where users struggled to apply a coupon" instead of filters over URLs and dates.

Session summarization is generating a natural-language account of what happened in a recorded session: the user's apparent goal, the path taken, the friction met, and the outcome.

A synthetic user is an AI agent standing in for a human user: exercising flows, testing changes, or generating load by operating the product's real interface.

Training data, in the interface-behavior context, is recorded interaction data used to teach models how software is operated: demonstrations of tasks, successful agent trajectories, and human sessions that show real usage patterns.

A world model is a learned model of how an environment responds to actions: given a state and an action, it predicts the next state.

Canvas recording is the capture of `<canvas>` element content in session recordings.

A cross-origin iframe is a frame whose content comes from a different origin than the parent page, such as an embedded payment form, video player, or third-party widget.

Event compression is the reduction of recording payload size before transmission or storage, in rrweb typically via the packer, which deflate-compresses events client-side.

The MutationObserver API is the browser interface that notifies JavaScript of DOM changes, delivering batched records of added and removed nodes, attribute changes, and text edits.

Shadow DOM is the browser feature that gives an element an encapsulated subtree with scoped styles, hidden from normal document traversal; it is the foundation of web components.

Web worker recording is the pattern of moving the expensive parts of session recording, such as serialization work, compression, and network transmission, off the page's main thread into a web worker.

WebSocket ingest is the delivery of recording events to the backend over a persistent socket connection instead of batched HTTP requests.