/vitals-review

Verifying a fix with local Unlighthouse

Use this only when the issue says the metric reproduces in the lab. That covers LCP, FCP, and TBT, and CLS only when lab CLS is at or above 0.1. It never covers INP, which needs a real interaction.

Build and serve both sides

Measure the default branch and the fix on the same machine, in the same session. A local number never predicts production. Only the difference between the two runs means anything.

pnpm install --frozen-lockfile
pnpm build
PORT=4173 node .output/server/index.mjs &

Use the repository's own preview command instead when it has one, for example pnpm preview. Stop the server after each run.

Scan the route

pnpm dlx unlighthouse-ci --site http://localhost:4173 \
  --urls <route> --<mobile|desktop> --throttle --samples 3 \
  --reporter jsonExpanded --output-path "$HOME/scratch/vitals-review/<run>/<before|after>"

• Pass the strategy the issue names. A mobile issue needs --mobile --throttle, or the number is a desktop number.
• --urls takes the one route. Never crawl the Site.
• --samples 3 takes the median of three runs.

Read ci-result.json in the output path. Record the metric for the route from both runs.

Report it

Put both values, the command, and the commit each side was built from in the pull request body. If the metric did not move past the Threshold in lab-series.md, say so and recommend closing. Never raise a Threshold, skip a route, or change the strategy to make the number move.