Skip to content

docs: visual judge proof and finished Build Week summary#4

Merged
AlexPaiva merged 2 commits into
mainfrom
feature/final-visual-story
Jul 21, 2026
Merged

docs: visual judge proof and finished Build Week summary#4
AlexPaiva merged 2 commits into
mainfrom
feature/final-visual-story

Conversation

@AlexPaiva

Copy link
Copy Markdown
Owner

Final presentation and provenance pass. Documentation, images, and one Action README wording fix only. No product, verifier, evaluator, evidence, schema, seeded-defect, or model-boundary behavior changed.

What changed

  • README front matter reordered for a non-technical first read: a 30-second version, a two-shot verifier proof (PASS then tampered BROKEN_PROMISE), "Why this is more than a privacy test", the authority split, a verified-facts table, and a single "Try it yourself" path (watch, challenge, run in CI).
  • docs/screenshots/: the two hosted-verifier screenshots (1920x1080), committed so the proof renders in the README.
  • BUILD_WEEK.md: a finished-product summary and a tag-by-tag timeline at the top; the original milestone log is preserved.
  • JUDGE_START_HERE.md: a "The model cannot award itself PASS" section that names the existing model-authority tests (no new test added).
  • .github/actions/verify/README.md: removed the stale sentence that told consumers to swap an already-pinned tag.
  • Limitations: clarified that the reusable Action runs on Windows, Ubuntu, and macOS while the full repair lifecycle is Windows-verified.

Verification

  • Every facts-table value checked against the repo and CI (2 defects, 2 replays, 1 approved repair, 5 clauses, tri-OS Action, 0 model-owned verdicts, 0 model calls in the verdict path).
  • npm run typecheck passes. No src/, landing/, or public/ source changed, so no site redeploy is required.
  • git diff --check clean. Screenshots confirmed 1920x1080 with no private data, paths, or keys.

Owner follow-up

  • Capture 01-broken-promise.png and 02-models-without-authority.png during the video, drop them into docs/screenshots/, and uncomment the ready image block in the README "See the proof" section to complete the four-image sequence.

… week

Restructure the README front matter for a non-technical first read: a
30-second version, a two-shot verifier proof, why it is more than a privacy
test, the authority split, a verified-facts table, and a single Try it
yourself path covering watch, challenge, and CI.

Commit the two hosted-verifier screenshots under docs/screenshots. Add a
finished-product summary and tag timeline to the top of BUILD_WEEK.md. Surface
the existing model-authority tests by name in JUDGE_START_HERE.md.

Correct the Action README so it no longer tells consumers to swap an
already-pinned tag, and clarify in Limitations that the reusable Action runs on
Windows, Ubuntu, and macOS while the repair lifecycle is Windows-verified.

No product, verifier, evaluator, evidence, or model-boundary behavior changed.
Add a fourth Try it yourself path that reproduces a committed sealed report
byte for byte and fails the gate on a committed broken bundle, so a judge can
see both PASS and BROKEN_PROMISE on files that ship in the repo, with no init,
scaffolding, or key. Document the committed artifacts/verify set and honest
programmatic use of the evaluatePromise function in ADOPTION. Add a CI status
badge to the README.

No product, verifier, evaluator, evidence, or model-boundary behavior changed.
@AlexPaiva
AlexPaiva merged commit a44c4b3 into main Jul 21, 2026
7 checks passed
@AlexPaiva
AlexPaiva deleted the feature/final-visual-story branch July 21, 2026 17:52
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