Start recording before the difference.

The tested command sequence

The native fixtures were recorded with @playwright/cli@0.1.22, published September 28, 2026. That package used Playwright engine 1.64.0-alpha-1790635538000 in this test. The observed library trace format was 10. A current package version alone does not establish compatibility with every trace source.

# Run these in a directory you control.
# Replace the example URL and selectors with your own staging app.
playwright-cli open about:blank
playwright-cli tracing-start
playwright-cli goto https://your-staging-app.example/
playwright-cli fill '#email' 'your-test-address@example.com'
playwright-cli click '#verify'
playwright-cli tracing-stop
playwright-cli close

Inspect the output paths reported by tracing-stop. Keep the matching .trace and .network files together. Capture the first run, change only the condition you intend to test, and record a second run. If a command fails, still stop recording and close your owned session. The exact selectors, address and wait condition must come from your application.

What this import accepts

A plain .trace and a Playwright Test trace.zip are different inputs. Do not rename a ZIP to bypass the file picker. Use the official Trace Viewer for full archives, screenshots, DOM snapshots and richer single-capture inspection. It is already a free, local-processing option; locality alone is not a TracePair advantage.

Keep the reproduction bounded

The browser checks 8 MiB and 40,000 physical lines per file, with at most 5,000 actions and 20,000 network records. A single record cannot exceed the parser’s one-megacharacter bound. Files parse in a Web Worker; Cancel stops an active read/parse operation from replacing the current capture. An invalid replacement leaves the previous capture intact.

Record a focused flow instead of a whole browsing session. The tool does not stream arbitrarily large recordings or decompress archives. Browser memory still grows with the supplied files and derived comparison, even though input is local. UI lists paginate large sequences; exports retain every selected interval request and every unassigned request.

Why a request may be missing

The supplied files define the scope. A request that happened before recording, after recording stopped, in another page, or through an API path not captured in these streams may be absent. Our complex native fixture sent a request through page.request.get(); the owned server received it, but these supplied trace/network files did not include it. A zero count is therefore a statement about these files, never proof that the application made no request.

Consult the pinned CLI tracing reference for the upstream recording workflow. If your format differs, use the official tooling and report the format through the correction contact; do not send secret-bearing traces by email.