Learning center

Reproduce the browser behavior someone else saw

A maintainer can only fix what they can make happen again. When a report says "it breaks for me", the missing piece is usually one of five things the reporter had and the maintainer does not: the build, the browser, the stored state, the exact input or the timing. This guide shows how to capture each one so the behavior replays on another machine.

The five things a replay needs#

Work down the list. Each layer is cheap to record at the moment of the bug and expensive to reconstruct a day later.

LayerWhat differs between two machinesWhat to record
BuildThe page's HTML, JavaScript bundle and the API version behind itThe page URL, the time in UTC, and any build or request ID the page or its responses expose
BrowserEngine, exact version, extensions, device emulation, preferenceschrome://version, the OS, and the extensions that were on
StateCookies, local storage, IndexedDB, the service worker, feature flagsA storage snapshot from a test account, the service worker's version and the flags the client evaluated
InputThe exact text, file, click target and navigation pathCode points for odd text, the file's size and type, the URL sequence including redirects
TimingResponse order, network speed, the clock, how long the tab was openA trace with timestamps, the throttling used, the reporter's local time and time zone

A report that pins all five usually replays on the first try. A report that pins only the input ("click Export") leaves the maintainer to guess the other four.

Choose the format the maintainer can run#

The best evidence is something the maintainer executes, not something they watch.

The further down that list, the more the maintainer has to reconstruct by hand.

Worked example: an export that only fails for one tester#

This example is fictional. A QA engineer reports: Export CSV downloads an empty file. Steps: open Orders, click Export. The maintainer follows the steps and gets a full file.

The maintainer asks for a trace instead of more steps. The tester runs the steps in a Playwright session with tracing on and sends trace.zip. In the trace, the request at the Export click is:

GET /api/orders/export?from=2026-11-01&to=2026-10-31&status=open
200  text/csv  0 rows

The date range is inverted. The trace's earlier actions show no date picker interaction, so the range came from stored state. The maintainer asks for the page's local storage, and the tester copies it from DevTools' Application panel: orders.filter = {"from":"2026-11-01","to":"2026-10-31"}, written by an earlier version of the date picker. The maintainer seeds that value before the page loads and the empty export replays on their machine every time.

The report the maintainer would have needed from the start fits in four lines:

Build: orders page, 2026-11-21 14:05 UTC, x-request-id 7f3c…
Browser: Chrome 153.0.8010.12, macOS 15.1, no extensions
State: localStorage orders.filter = {"from":"2026-11-01","to":"2026-10-31"}
Input: Orders > Export CSV; request has from > to; response 0 rows

Guides in this collection#

Pick the guide for the layer you are missing.

Formats a maintainer can open or run

The browser the reporter used

The reporter's settings and surroundings

State the reporter had

Input the reporter used

Navigation and network

Time and speed

The maintainer's side

When the question is which cause explains a symptom, rather than how to make it happen again, start from the browser bug report method and its symptom guides.

Limits of a replay#

A replay proves that the captured conditions produce the behavior. It does not prove they were the only conditions on the reporter's machine, and it cannot replay what was never captured, such as a server that has since been redeployed. Say which layers you pinned and which you assumed. When a replay fails to reproduce, that is evidence too: it tells the maintainer the cause is in a layer you did not capture.

Keep real customer data out of every artifact. A trace, a HAR, a storage snapshot and a heap snapshot can each hold session tokens and personal data. Reproduce with a test account or synthetic records whenever you can, and see Share a HAR file for what to remove before sending.

Record the timeline with Session Recorder#

Session Recorder, a free Chrome extension from Proofline, records clicks, console errors, network activity and screenshots on one timeline while you reproduce a failure, then exports a Markdown report. Recording and export need no account. The setup guide covers installation and export. Review the export before you share it, because automatic redaction can miss values.

Questions#

Which layer should I capture first?

Capture the build and the time first, because they change soonest: a deploy can land within the hour. Then capture state, because the reporter's next action may overwrite it. Browser details and input can usually be collected later.

The maintainer still cannot reproduce it. What now?

Ask which layer differs. Run the maintainer's attempt and the reporter's attempt side by side and compare the first request or state value that differs. If none differs, the cause is in something neither of you captured, such as an extension, a proxy or a server-side record.

Is a reproduction rate useful evidence?

Yes. "Fails 6 of 20 runs" tells the maintainer the bug depends on timing or order, and gives them a baseline to show that a fix worked. Run the same script repeatedly and count, rather than estimating.

One month of free code review.

New organizations get one calendar month of hosted code review for private and public GitHub repositories. No card or cloud connection required. Eligibility and fair-use terms.

Start free with GitHub

Related reading