The final count.
The important distinction is active source versus stale artifact. The current build and the 0S mount both had 339 business pages and no missing mounted pages. The stale dist-cloudflare-pages artifact still had the old business-profile fallback path and only a partial business set, so it was removed as a deploy target instead of being allowed to confuse the production story.
What changed.
| Area | Final behavior | Proof |
|---|---|---|
| Business pages | Business routes publish from src/handbuilt-pages/<business-id>/index.html. Missing pages fail the build instead of falling back to a generated company page. |
metraiyux_0s_site/_platform-sources/valley-verified/proof/no-generated-business-pages.json |
| Fallback renderer | /business-profile/, /profile-template-preview/, and /data/profile-template-options.json are not published. |
Live Pages HTTP checks returned 404 for all three removed fallback routes. |
| Owner action | Businesses accept a free SkyEmail account with shared 0S/SkyGate sign-in and a 24-hour workspace activation window. | Business-page proof checks require Accept SkyEmail. |
| 0S mount | Valley Verified is mounted in the 0S and still protected by the shared FS27/Free99 gate. | Unauthenticated /valley-verified/ and sample business routes redirect to /admin/login.html. |
Why kAIxU had to own the workflow.
The lesson did not change just because the closure is clean now. External models can write code, but the workflow is the intelligence. If the model does not check which businesses were already hand-built, which pages are still generated, which route is active, and which artifact is stale, it can ship effort in the wrong place.
kAIxU has to own the acceptance criteria because my system is not a generic repo. There is a shared 0S gate, no local app-specific admin password lane, no Codex-run browser proof unless I re-enable it, MCP mining receipts, Cloudflare Pages receipts, 0S Worker receipts, and a hard line between public proof and private authority. A model that is not trained on that workflow needs a smaller lane and a stricter checklist.
The corrected workflow was simple: count first, assign agents to audit the real state, patch the build so generated business pages cannot return, prove the active dist and mounted 0S copy, deploy, then live-check with HTTP. No hand-waving.
Production receipt.
- Cloudflare Pages canonical: https://valley-verified.pages.dev/
- Cloudflare Pages deployment: https://0159b2a0.valley-verified.pages.dev
- Deployment ID:
0159b2a0-2519-4a3a-a10a-979512c7aa75 - 0S Worker version:
0b561f44-5f5c-473d-ad51-0c2541845ab2 - 0S gated mount: https://metraiyux-0s-full-system.graylondonskyes.workers.dev/valley-verified/
- Pages upload receipt:
test-artifacts/cloudflare-pages/valley-verified-no-generated-business-pages-receipt.json - HTTP smoke receipt:
test-artifacts/cloudflare-pages/valley-verified-no-generated-business-pages-http-smoke.json - 0S Worker receipt:
test-artifacts/0s-browser-end-to-end/valley-verified-worker-deploy-receipt.json - Valley smoke suite:
1055 checks passed
Browser proof was not run because this repo's owner policy disables Codex-run browser proof. That is not a failure. That is the boundary. The live production evidence for this pass is build proof, smoke proof, direct Pages upload receipt, 0S Worker deploy, and non-browser HTTP checks.
The rule going forward.
A page is not finished because a model touched it. A page is finished when the count moves, the route is active, the fallback is gone, the owner action is correct, and production proof says the same thing.