Performance

The repository measures local compute, database queries, payloads, and painted browser data. Budgets are decision targets, not claims that current code meets them.

Choose a measurement

Browser readiness

The probe requires a destination-matching data element that is visible and not hidden/inert/busy, then waits for animation frames. A heading, skeleton, or API response alone is insufficient. CDP measures transferred bytes separately from decoded API JSON. Tests also check duplicate GETs and font hints in both HTML and Link headers.

Cold entry uses fresh contexts; navigation measures real in-app actions without document reload. Simulated internet settings model a repeatable lab link, not deployment latency. Remote mode needs a real host and authenticated storage state; it does not use the local test bypass. See the performance runbook for environment flags and artifact locations.

Using results

Source budgets live in src/lib/perf/budgets.ts. Read the HTML/JSON summary with the exit status: failed and slow samples remain in the report. Compare like-for-like dataset, build, network profile and browser settings. Do not infer throughput or prove an alternative architecture slower from a target number.

Change checks

A new list field can increase payload on every page; a new provider subscription can affect every keystroke. Measure the affected path, preserve source/decoded/transfer distinctions, and keep readiness selectors aligned with UI changes. Fixture review keys are UUIDs from fixture helpers or current data, not hand-built FFY URLs.

Read the code

• src/lib/perf/budgets.ts
• src/lib/perf/measure.ts
• src/lib/server-timing.ts
• e2e/perf/measure.ts
• e2e/perf/probe.ts
• e2e/perf/reporter.ts
• playwright.perf.config.ts

Related file notes