Skip to content

Test playwright tests with Jev - #8

Open
ostenbom wants to merge 12 commits into
mainfrom
oliver/jev-test-loop
Open

ostenbom wants to merge 12 commits into
mainfrom
oliver/jev-test-loop

Conversation

@ostenbom

@ostenbom ostenbom commented Sep 22, 2026 •

Copy link
Copy Markdown
Contributor

Summary

  • Add a Bun command for running the Jev-driven live name-change test loop
  • Add an isolated Playwright harness, frozen test specification, and run documentation
  • Capture observations, action decisions, screenshots, usage estimates, and independent completion checks

Testing

  • Added and ran Bun unit tests for candidate generation and action-code safety
  • Live Playwright validation is available through bun run jev:test

@vercel

vercel Bot commented Sep 22, 2026 •

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated
endform-playwright-tutorial Ready Ready Preview Oct 20, 2026 4:44am UTC

Request Review

Benchmark variants now run interleaved per scenario. Adding the page
screenshot to the no-gate controller lowered verified passes from
109/150 to 79/150 on Jev 1.14. The stepwise controller (settling,
compact history, per-field values, finish vs stop, no-op skipping,
ordered aim progress, Escape) reached 150/150 with screenshots and
149/150 without. Three verifiers now check outcomes rather than the
final page, and the fault matrix shows no pass with a broken outcome.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@endform-dev

endform-dev Bot commented Oct 20, 2026 •

Copy link
Copy Markdown

Endform

Passed on 151005f

Suite Result Tests Duration
(repository root) 🟢 Passed 🟢 15 passed 33s

Clairic, Cloudflare's extraction model, answers text fields as spans of
the input, so the loop needs no candidate actions: it asks for the next
element ref, the value to type, a direct URL, and a quote proving
completion, and code acts by the element's role. In one interleaved
batch the naive Jev option menu verified 110/150, naive Clairic 42/150,
Clairic with a screenshot 30/150, and Clairic with Clef judging
completion 53/150. Clairic mostly quoted the aim itself as completion
evidence, and without that it did not know when to stop.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Clairic with Clef also passed activity-after-account-update 10/10, so the
2/10 ceiling for multi-step scenarios applies to the naive loop only.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

This branch was successfully deployed

1 active deployment
Preview — 151005f6 Deployed Oct 20, 2026 by vercel[bot]
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant