Verify

scripts/verify.py runs the same commands as CI, plus the brain check and the interface budget from docs/UI_PLAN.md §7. It prints one line per step with the counts the tools reported. Those lines are the only test numbers you may quote.

Choose the run

• Add PG_TEST_URL=… REQUIRE_PG=1 (the URL is in CLAUDE.md) to run the PostgreSQL tests as well. Without it they skip, and the summary shows how many.
• The end-to-end step uses the pre-installed Chromium at /opt/pw-browsers/chromium when it exists.
• The e2e run rewrites docs/screenshots/. Restore them with git checkout -- docs/screenshots unless the change is meant to update them.

Report

• Quote the PASS and FAIL lines as printed, with their counts, in the form "Python tests: N tests, K skipped".
• If a step failed, show the failing test from the output tail, and fix it or say plainly why not.
• Never round a count, carry one over from an earlier session, or call a skipped step passed.
• When a count changes, update README.md, the one place test counts are kept. tests/test_docs.py checks other numeric claims.

Performance

Lighthouse is not part of the script, because it needs a running server. To measure:

1. Run python scripts/serve.py --local-only --port 8011 in the background.
2. In frontend/, run CHROME_PATH=/opt/pw-browsers/chromium npx lighthouse http://127.0.0.1:8011/ --quiet --output=json --output-path=<scratch>/lh.json --chrome-flags="--headless=new --no-sandbox". Run it three times and quote the median.
3. Quote the scores and LCP, CLS and TBT as measured, with the date. The targets are in docs/UI_PLAN.md §7.