Stand the backend up, and drive the primary journey on it

[HomenShum]

Nine of the twelve promotion conditions are judged on what a browser shows, and all nine were UNVERIFIED for one reason: every product route rendered "Convex backend not configured" (defect D1). This provisions an isolated Convex dev deployment, drives J1, J2 and J4 end to end against it, and commits both the captures and the scripts that regenerate them.

Two things were actually broken behind that door

Neither names itself, and both stop a cold clone dead.

1. @erquhart/convex-oss-stats imports @convex-dev/crons and declares it nowhere — not in dependencies, not in peerDependencies. With package-lock.json gitignored, a fresh install can land a tree without it and the very first convex dev push dies on Could not resolve "@convex-dev/crons/convex.config". Now a direct dependency.

2. A missing OPTIONAL model key failed the ENTIRE deploy. Convex analyses every backend module on every push; coordinatorAgent.ts builds DEFAULT_MODEL (kimi-k2.6, OpenRouter) at module scope because a Convex function must be a module-level export; and buildLanguageModel threw at construction for an unconfigured provider. So convex dev failed with InvalidModules: Failed to analyze domains/agents/digestAgent.js unless you had an OpenRouter account — for a provider /redesign/chat never calls. The error now lives on doGenerate/doStream, so an unconfigured provider fails the call that needs the key rather than the deploy that does not. Gated by a new modelResolver.test.ts case that deletes the key, constructs the model, and asserts both halves.

What now runs, and what proves it

node scripts/capture-live-journey.mjs --port 4902 → exit 0, 10/10 checks. It asserts, it does not narrate:

• the surface mounts with data-empty="true" on a live backend;
• the run streams with a named-stage checklist and a durable data-chat-run-id;
• the sealed packet's source count and evidence agree, in both directions;
• the run row reaches complete with 30 ordered redesignChatStreamEvents rows read back out of Convex — so "it streamed" and "it persisted" are separate claims;
• /redesign/chat/r/<hash> opened cold replays rather than re-runs, proven by getLatestOwnedRun().runId being unchanged;
• Stop yields an honest terminal state with no fabricated packet, and the session keeps working;
• a 2-character prompt is rejected by startChat at the trust boundary;
• zero console errors and zero failed requests, at 1280 and at 375.

Plus scripts/audit-web-quality.mjs (Lighthouse 13.4.1 + axe-core 4.13.0 on the production build) and scripts/review-web-interface-guidelines.mjs with the written review at promotion/evidence/wig-review/REVIEW.md. That review is a review, not a Lighthouse score — the two measure different things and are committed separately.

Scorecard: 1/12 → 7/12, and UNVERIFIED is gone from the table

Conditions 1, 3, 4, 5, 6 and 9 move on committed evidence. Conditions 2, 7, 8, 10 and 11 FAIL on measured numbers, which is the point — running the thing found three new defects:

• D5 — a rejected prompt shows "The live chat run could not be started." and sends the actual reason only to the console.
• D6 — LCP 10 827 ms on the production build, plus four Web Interface Guidelines deviations (no h1, no skip link, sub-44px touch targets including the submit button, 14.5px composer font).
• D7 — the same question returned 3, 2, 1, 0, 1 and 0 grounded sources across six runs.

Setup is written down in docs/START_HERE.md "Before Step 1" and docs/codebase/CONCERNS.md C7b, including both traps, so the next cold reader does not rediscover them.

Not touched: the known api→never typecheck cascade (D2).

🤖 Generated with Claude Code

[gitguardian]

⚠️ GitGuardian has uncovered 1 secret following the scan of your pull request.

Please consider investigating the findings and remediating the incidents. Failure to do so may lead to compromising the associated services or software components.

🔎 Detected hardcoded secret in your pull request

GitGuardian id	GitGuardian status	Secret	Commit	Filename
36087749	Triggered	Generic Password	a4e9cd1	scripts/capture-live-journey.mjs	View secret

🛠 Guidelines to remediate hardcoded secrets

1. Understand the implications of revoking this secret by investigating where it is used in your code.
2. Replace and store your secret safely. Learn here the best practices.
3. Revoke and rotate this secret.
4. If possible, rewrite git history. Rewriting git history is not a trivial act. You might completely break other contributing developers' workflow and you risk accidentally deleting legitimate data.

To avoid such incidents in the future consider

• following these best practices for managing and storing secrets including API keys and other credentials
• install secret detection on pre-commit to catch secret before it leaves your machine and ease remediation.

---

^{🦉 GitGuardian detects secrets in your source code to help developers and security teams secure the modern development process. You are seeing this because you or someone else with access to this repository has authorized GitGuardian to scan your pull request.}

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
nodebench-ai	Ready	Preview	Aug 14, 2026 3:43am

[github-actions]

PR size advisory

This PR adds 19309 lines of substantive change. CONTRIBUTING.md defines a soft limit of ~400 LOC.

If the PR is genuinely cohesive (e.g. an architecture map, a generated migration, a deletion of a dead module), no action is needed. Otherwise consider:

• splitting into 2-3 PRs along independent concerns
• pre-discussing the architecture change in a GitHub Discussion before merge

This is advisory — it does not block the merge.

[HomenShum]

Superseded by #616, which carries the same work on a conventionally-named branch with a conventional title, rebased onto main so the expired MCP publish token is no longer in the tree.