Use these boundaries when deciding what a shared bundle can establish and what needs another tool or a separate execution summary.

Timeline and session limits

Recorder does not stop after a fixed number of recorded actions. Capture begins only after Start recording, and at Stop Recorder retains up to the last 120 seconds of timeline events. If the recording is shorter than this limit, the retained interval begins at Start; otherwise, older events are discarded as the recording continues.

The elapsed duration in session.md measures the whole recording, not how much evidence remains. If the attempt exceeded the retained timeline, treat early setup as unrecorded unless another supplied artifact establishes it. Ask the tester for separate checkpoint bundles when the missing interval matters.

This timeline limit is the same for local and connected recordings. Signing in or navigating does not extend it. There is no pre-Start action replay, and screenshots or written observations cannot recover discarded events. Metadata, notes, attachments, and downloaded or uploaded copies have separate lifetimes; this window is not a deletion schedule for all session data.

Separate per-session budgets protect storage and processing. The current defaults allow a report of up to 16 MiB, up to 5,000 HTTP requests including 3,000 non-static requests, and up to 10 attachments. Each attachment can be up to 2 MiB, with 20 MiB available across attachments. The Data Plane enforces these limits for uploads and processing; local review and export can remain available when an upload is rejected. These are payload and request limits, not a recorded-action limit.

Browser scope

Recorder capture is supported in Google Chrome on a computer. Browser-internal and extension pages are restricted; account-free mode also excludes local file URLs. Broad host permission does not bypass the local hostname scope or connected project rules. Use these facts to interpret a gap in the bundle, not as steps for opening the bundle in Recorder.

Actions, frames, and components

Supported action evidence includes clicks, input/control changes, selected keyboard actions, scrolling, and navigation. session.md contains a numbered action summary, while steps.json and generated scripts contain their own converted action forms. These files do not reproduce the complete event log or action detail panels.

Iframe provenance and open-shadow-root targets can be recorded. Untrusted iframe values, closed shadow roots, unavailable target information, and complex widgets can limit capture or generated locators. Keyboard evidence is not a complete focus history.

Bodies and size limits

Payload limits depend on configuration; the account-free default is 512 KiB. Bodies can be truncated, unavailable, filtered, or represented by binary placeholders. Browser protocol availability and streaming behavior affect completeness.

Screenshot attachments also face destination-specific size limits. A tracker rejecting an image does not necessarily mean issue creation failed.

Performance evidence

Available observations include FCP, LCP, long tasks, and selected resource timing. Observer support varies. Buffered paint/resource measurements can predate Start; long tasks are observed during capture with a 50 ms threshold. INP and CLS are not calculated.

Bundle versus runtime evidence

The runtime Report, HAR, DevTools JSON, and generated scripts are distinct formats. A standard bundle is not a serialized runtime Report. WebSocket frames, environment panels, performance panels, filters, source navigation, and privacy decisions are not reconstructed from the download. Screenshots are added explicitly and are not automatically masked.

Use Inspect a shared artifact bundle for the portable file set and its omission rules.

See artifact formats before building tooling or asserting complete coverage. If a necessary observation exists only in Recorder, ask the tester to write it into Description or a source-linked note and provide a new reviewed bundle.