TempuraML's picture
test(e2e): Playwright deep-dive suite (local-deep + staging-smoke)
df692ec
|
Raw
History Blame Contribute Delete
1.46 kB

scripts/ β€” dev + ops tooling (not shipped runtime)

Standalone scripts run with ./.venv/Scripts/python.exe scripts/<name>.py. Nothing here is imported by the app.

What's here

  • dev_seed.py β€” run the app locally with a dev-token login (no Clerk) + in-memory SQLite, seeded with clickable batch/booking leads (overlay image + outline geometry + a prior edit). Serves http://127.0.0.1:8000; open /dashboard?token=devtoken. Seeds the in-process viz cache too, so it must run the server in-process (it does).
  • browser_smoke.py β€” Playwright-for-Python click-test of the dashboard popup / drag-editor / history. No Node or MCP β€” pip install -e ".[dev]" gets playwright, then python -m playwright install chromium once. Run against a live dev_seed.py. Screenshots β†’ scripts/_browser_shots/ (gitignored).
  • debug_hardscape.py β€” per-class segmentation predictions for one address.
  • run_eval.py / run_ablation.py / audit_pipeline.py β€” engine accuracy analysis (labeled before/after evals β†’ data/evals/<label>/<ts>/).
  • dogis_export_probe.py β€” DOGIS tile export probing (imagery research).

The full browser suite is e2e/ (pytest e2e -m local β€” auto-starts dev_seed.py and covers dashboard/widget/admin); browser_smoke.py is the fast standalone smoke. Run one of them after any dashboard/widget JS change β€” the Python unit suite does not exercise the browser.