fix(ui): preserve canonical typography and home restoration

[sgardoll]

Summary

• Restore the canonical mobile title letter-spacing of -0.025em.
• Replace Mac-specific intrinsic-geometry expectations with same-browser comparisons against a SHA256-pinned canonical reference, using a tolerance of <=1px.
• Restore Home presentation: abandoned runs expose the composer; completed Results restore fullscreen presentation and the persisted tab across two navigation cycles, including reduced motion.
• Apply the shared cancellation invariant: hiding the stage clears its presentation marker unless navigation explicitly preserves meaningful results. Regression coverage includes editing, failure, and cancellation paths under normal and reduced motion.

Coder-reported local verification — round 3

These results apply to the frozen snapshot published in this PR; they are not independent reviewer approval.

• Unit tests: 398/398 passed, zero failures or skips.
• Browser tests: 275/275 passed, 2 workers, 0 retries, zero failures.
• Build: passed.
• Accessibility: 7 modal shells passed.
• Deployment guard: 9/9 passed.
• Proxy allowlist: 30/30 passed.
• Production bundle leak scan: passed; canonical credential scan: 12 patterns, 0 hits.
• 8 new regressions failed before the fix; the subsequent targeted run passed 14 tests.

Evidence and snapshot identity

Local evidence directory: /Users/home/Projects/ccc-diagnostics/stu445-r2/round3/

• Authoritative manifest: /Users/home/Projects/ccc-diagnostics/stu445-r2/round3/final-manifest.json
• Manifest SHA256: d91313db4539c5c01023bd97a747c42c482fe06eca4a471c28188572337381dd
• Full-run logs: /Users/home/Projects/ccc-diagnostics/stu445-r2/round3/full-verification/ (02-unit.log, 03-build.log, 05-browser.log, 06-accessibility.log, 07-deploy-guard.log, 08-proxies.log, 09-leaks.log).
• Regression evidence: /Users/home/Projects/ccc-diagnostics/stu445-r2/round3/red.log and /Users/home/Projects/ccc-diagnostics/stu445-r2/round3/green.log.
• Release checks confirmed raw file SHA256 values and byte counts for all 5 authored files and 16 dist files, including their committed blobs, against the manifest. Patch serialization hashes are not the snapshot identity.
• Published bundle: dist/assets/index-vWxoLxnc.js, SHA256 83506f0431dbdf0c4ea0d72eafe79659471f62732f1ecbdb224d8e8ffbcb8e85.
• Commit: cee6fc2f9659ddd05a1dbaf124e677af1546e73d; sole parent: dc3c48834fcfd6b44110c567cd9d20e8125255e2.

Draft review status and canonical whitespace exception

• Linux GitHub CI pending. The coder's local verification ran on macOS (PHP 8.5.8 locally versus CI 8.3).
• Independent final refuter review pending on the same bytes; no independent approval is claimed.
• The canonical HTML fixture intentionally retains its five original trailing-whitespace lines to remain byte-identical. Its SHA256 is 548e628fe94f2b5e647cb4849fbc3200b43dcead60ea53c076f0085aca8917dc.
• Draft targets the existing milestone/stu-implement-updated-ui-ux branch.

[sgardoll]

Current-head release review record — declared internal fallback

Exact approved source commit: cee6fc2f9659ddd05a1dbaf124e677af1546e73d
Approved source tree: db8cd435a57340aafc4f6dccfae5827798ce1f7a
Authorized target: existing milestone/stu-implement-updated-ui-ux, pre-merge tip dc3c48834fcfd6b44110c567cd9d20e8125255e2.

CI and snapshot evidence

• Linux CI run 38412341896, attempt 1 completed successfully on this exact source SHA; release administration confirmed all 16 recorded steps succeeded. Recorded suite totals: 398 unit tests, 275 browser tests, zero flaky tests/retries, 7 accessibility shells, 9 deployment-guard cases, 30 proxy cases, production build and bundle leak scan passing.
• GitGuardian Security Checks passed on this head. This is a passing check, not a review approval.
• Authoritative round-three manifest SHA256: d91313db4539c5c01023bd97a747c42c482fe06eca4a471c28188572337381dd. All 5 authored and 16 dist files were verified against that manifest; release publication also verified their committed blobs.

Bounded independent internal review

The orchestrator reports the final independent internal refuter result as VERIFIED WITH CAVEATS, with no blockers on these same source bytes. Reported scope includes all 21 manifest files, 15 targeted checks, and 6 production adversarial checks. The adversarial checks passed after a timeout-only extension of the harness; source bytes were unchanged.

Evidence directory: /Users/home/Projects/ccc-diagnostics/stu445-r3-independent-5sDOGD/.

Caveats: this does not establish exhaustive painted-frame equivalence, and there was no external-model review. The canonical fixture's five original trailing-whitespace lines remain intentionally byte-identical.

Approval provenance and authorized fallback

At the release gate, GitHub reports 0 current-head reviews, no inline review comments, and no unresolved review threads. No external-model or bot review approval is claimed. The earlier AI security review route was unavailable because of unsupported-model infrastructure. The user explicitly accepted this declared internal fallback under the Stage 2 flow and authorized a normal merge commit for PR #129 into the existing milestone only.

GitHub reports no configured required checks; both current-head check runs are successful. Merge remains subject to normal repository policy and an exact-head guard. The feature branch is retained for audit. Product post-integration checks are delegated separately to the coder.

[greptile-apps]

This PR does not match any of the 1 configured review trigger rule.

[devin-ai-integration]

Devin Review found 1 potential issue.

[devin-ai-integration]

🔴 Abandoned runs expose unreviewed results

Leaving during Code Review makes switchView display Results because pipelineState.artifactBundle already exists. The review request is aborted, so Home presents unreviewed code as a completed result.

Learn more

Generator populates pipelineState.artifactBundle before Code Review completes in runThinkingPipeline. On navigation, updateSelectedArtifactPanels renders that bundle as Results even though pipelineState.step3Result has not arrived. The newly preserved panel marker then makes switchView('home') reveal those results. The no-bundle branch clears the marker, but the presence of a generated bundle does not prove a completed run.

Example: Architect and Generator succeed, then the user opens Account while Review is pending. Review is aborted; returning Home displays Results with the generated code instead of returning to the composer or a reviewed previous result.

Recommended fix: In switchView, distinguish a reviewed, completed result from a partially populated bundle before calling updateSelectedArtifactPanels. Clear the marker and return to the composer for an abandoned first run; if replacing an earlier result, restore its completed snapshot instead.

---

Was this helpful? React with 👍 or 👎 to provide feedback.