Answer mockup: take the symbol off the turn, drop the safety rail, and make the two panels one

[claude]

Summary

Design scratch only. Nothing under src/components/clinical-dashboard/ or src/app/globals.css changes, and no production route renders any of it. classifyPullRequestFiles on this diff returns clinicalRisk: false, operationalRisk: false, ragRanking: false.

• The assistant avatar comes off the chat turn, and the answer prose runs the full column width. On a 390px phone that column cost about 2.75rem of every line of a clinical answer — the exact measurement --answer-message-gutter exists to hold — and bought nothing: there are two speakers, the person's turn is already a right-aligned bubble, and the answer is the one element on this surface that wants the full measure. What identifies the turn instead is the provenance line, which is information rather than decoration, and which now sets as two deliberate lines (what wrote the answer, then what the reader owes it) instead of one sentence breaking mid-phrase. Its wording is unchanged.
• A hairline and a quiet CITED DOCUMENTS / SOURCES READ eyebrow above the card row. With no gutter organising the column, the answer needed a clean end before its evidence.
• The card row fades at its right edge, so it reads as scrollable without a scrollbar — the mechanism .answer-suggestion-chips-scroll already ships in globals.css. Written as an inline style because the arbitrary Tailwind property did not generate a rule here; checked in the browser, not assumed.
• Copy became Copy with sources. On this product an answer copied without its citations is the actual hazard, so the control names what it puts on the clipboard.
• The safety card is drawn without its coloured rail.AnswerSupportSummaryCard runs border-t-2 border-t-[color:var(--warning)] across the top of a card that also holds Evidence gaps and Report a problem, so the rule colours two controls carrying no state. Colour moves into the icon tile plus a short RED FLAG chip — the direction also-matches-accent-mockups.tsx already records for the sibling surface. The signal does not weaken: the design-system contract holds colour-only status indicators at zero either way. The evidence-gaps count is drawn neutral rather than amber, because a status-coloured numeral is what that same contract holds at zero.
• "Also in your library" and "Also matches" become one list. They stack directly on top of each other under an answer and read as one panel repeated; they answer one question, so each row names what it is (Medication, Factsheet, Mode) instead of leaving that to be inferred from which panel it landed in. "4 related modes" goes — "mode" is internal vocabulary. Two directions are drawn so the owner picks rather than approving the only option on the page.
• Found while drawing it:answer-result-surface.tsx wraps AnswerFollowUpSuggestions in hidden sm:block, so the most likely next tap on the answer surface has been desktop-only. Both directions restore it, above the matches. That fix belongs to the production PR, not this one.

Verification

• npx tsc --noEmit — no diagnostics.
• npx eslint on both changed component files — exit 0, no findings.
• npm run check:design-system-contract — passed (1012 production files); no ratchet moved. Status-colour boundary: colour-only status indicators 0; status-coloured numerals 0; image inversions 0.Scale ratchets: raw padding literals 42; raw radius literals 20; raw gap literals 20; raw margin literals 52; raw line-height literals 3.
• npm run format then repo-wide npx prettier --check . — "All matched files use Prettier code style!"
• Browser proof on /mockups/answer-chat-perfected-v2 at 390px and 1280px, in light, dark, and Chromium forcedColors: "active". Alignment measured with text ranges rather than element boxes, because the prose box bleeds 8px for its highlight wash: notice, prose, eyebrow and card row all start at x = 59.
• npm run verify:ui — not run, by risk. No production route renders these components; the Chromium journeys do not reach /mockups/**, which 404s in production. It runs before the production PR, where shared answer-surface foundations actually change.
• npm run test — not run. No test references any symbol in this diff (searched tests/ for the changed component and token names).

Risk and rollout

• Risk: low and contained. These routes are design scratch, blocked in production by src/proxy.ts, and exempt from the wiring and reachability gates — but not from typecheck, lint, the design-system contract, or check:bundle-budget, all of which ran or apply. The one thing a reviewer should weigh is whether the safety card reads as a warning without its rail; the before-and-after is drawn side by side in section Seven for exactly that judgement.
• Rollback: git revert either commit independently. They share no hunks.
• Provider or production effects: none. No provider-backed gate was run and no live service was called.
• RAG impact: none — no file under src/lib/rag/**, no retrieval RPC, no ranking surface, no eval harness, no golden fixture. classifyPullRequestFiles returns ragRanking: false.

Notes

• The production change is deliberately a separate pull request, after the owner picks a direction: removing the shield and --answer-message-gutter from answer-content.tsx / answer-card.tsx / globals.css, restyling the real safety card, capping in-text marks at one number plus +N, and restoring the follow-up suggestions on phones. That one is clinical-risk and carries the full governance preflight.
• Both decisions are written into docs/answer-page-redesign-handover.md (§12.6, §12.7), which is what the production change will be built from.

---

Generated by Claude Code

[supabase]

This pull request has been ignored for the connected project sjrfecxgysukkwxsowpy because there are no changes detected in supabase directory. You can change this behaviour in Project Integrations Settings ↗︎.

---

Preview Branches by Supabase.
Learn more about Supabase Branching ↗︎.

[cursor]

Stale comment

Approved. Cursor Bugbot was not present after the first check poll, so that signal was skipped; no approval policy required human review and there is no current approval from this automation. No reviewers were assigned.

_{Sent by Cursor Approval Agent: Pull Request Router and Approver}

[github-actions]

CI triage

CI failed on this PR. Automated classification of the 2 failed job(s):

• Build — needs investigation: inspect the failing step and uploaded diagnostics; rerun only after classifying the cause.
• PR required — needs investigation: inspect the failing step and uploaded diagnostics; rerun only after classifying the cause.

Compared with main CI run #14008 (failure).

Classification is evidence routing, not permission to ignore a failure. Exact quarantined Playwright identities remain governed by the flake ledger.

[cursor]

Stale comment

Approved. Cursor Bugbot was not present after the first check poll, so that signal was skipped; no approval policy required human review and this automation has no current approval. No reviewers were assigned.

_{Sent by Cursor Approval Agent: Pull Request Router and Approver}

[BigSimmo]

Build was red on mockups scratch weight: 610.6 KiB gzip vs 487.6 KiB baseline (+25.2%, ceiling 25%). Production stayed within tolerance. There were no unresolved review threads.

501dfe7d6 refreshes only bundle-budget.jsonmockups.gzipBytes to the CI-measured scratch size. Production and route budgets are unchanged. This is the documented hygiene-ceiling path for design-scratch growth, not a production-budget raise.

[cursor]

Approved. Cursor Bugbot was not present after the first check poll, so that signal was skipped; no approval policy required human review and this automation has no current approval. No reviewers were assigned.

_{Sent by Cursor Approval Agent: Pull Request Router and Approver}