fix: resolve all four audit hardening residuals (#567-#570)

.github/workflows/ci.yml

@@ -102,6 +102,7 @@ jobs:
           scripts/__tests__/releaseWorkflowContracts.test.ts
           convex/domains/redesign/chatRuns.responseShape.test.ts
           src/features/redesign/components/UniversalComposer.test.tsx
+          src/features/redesign/surfaces/ScratchnodeEventsSurface.test.tsx
 
   scratchnode-launch-gates:
     name: ScratchNode launch gates

CHANGELOG/pages/redesign-chat.md

@@ -3,6 +3,48 @@
 Append-only lane for the public redesign chat, reproducible answer receipts, and their
 transition into an authenticated live conversation. Newest entries first.
 
+## 2026-07-17 - Resolve the four hardening residuals from the production audit
+
+The 2026-07-16 audit cycle left four filed residuals (#567-#570); all four land here.
+
+1. **Red tests CI could not see (#567)** - three ScratchnodeEventsSurface tests failed on
+   clean main (`TypeError: reading 'product'`) because the test's api mock predated
+   ImportRecapButton's `domains.product.scratchnodeImport` query, and the runtime-smoke
+   allowlist never ran the file. The api mock is now a Proxy that degrades unknown
+   function refs to unresolved queries, the missing `useMutation` mock is added, two
+   scenarios cover the import affordance's published/unresolved gates, and the file joins
+   the CI allowlist.
+2. **Timer-only cancel guard (#568)** - Stop and submit are both always rendered: Stop in
+   a reserved `visibility:hidden` slot, submit staying at identical coordinates while
+   streaming (disabled). A double-click's second click lands on an inert control at ANY
+   user double-click interval; the 400ms arming stays as defense in depth.
+3. **Shape detector gap (#569)** - "in one sentence", "a single paragraph", and
+   "under N words" are now detected, instructed via an exhaustive
+   `responseShapeSystemInstructions` switch, and deterministically enforced (first-sentence
+   extraction, prose collapse, sentence-accumulating word budget). Honesty survives
+   compaction: unsupported runs carry `Source needed:` within the requested shape.
+   JSON/table stay model-side deliberately - the render path is markdown prose.
+4. **No computed-geometry regression capture (#570)** - one-flow-regression (Tier B,
+   per-PR) gains a 390x844 test asserting `.rd-shell__main` resolves to ONE grid track and
+   `main#main-content` stays >300px wide - the audit's 70px collapse was invisible to both
+   the CSS-source string guard and the document-overflow boolean.
+
+**PR / canonical main commit**: `PENDING #NNN MAIN SHA / FINAL QA`.
+
+**Evidence state**:
+- Source: `pending`
+- Checks: `npx tsc --noEmit` -> 0 errors. Full runtime-smoke allowlist locally ->
+  312 passed / 25 skipped (integration skips without env keys, matching CI). Adjacent
+  guards (ChatResponseShape, ChatRunLifecycle, ChatContinuation, sourceVerificationPolicy)
+  -> 12 passed.
+- Visual proof: `not recorded` - behavior changes are runtime/composer mechanics.
+- Preview: Tier B one-flow-regression includes the new 390px geometry test on this PR.
+- Production live: `not recorded` at entry time.
+
+**Author**: Homen Shum + Claude Fable 5.
+**Touches**: [components/fast-agent-panel.md](../components/fast-agent-panel.md) is NOT
+touched - the composer change is redesign-chat's UniversalComposer, not FastAgentPanel.
+
 ## 2026-07-17 - Ground "best source" superlatives on their own citation
 
 `applyDeterministicResponsePolicy` only rewrote unsupported "best/strongest source|claim"
@@ -22,10 +64,10 @@ constant used by both the sanitizer and the gate so they cannot disagree.
 This closes the residual left open by the 2026-07-16 audit follow-up (#565 fixed four of
 the five P1s; this is the fifth). Tracked as #566.
 
-**PR / canonical main commit**: `PENDING #NNN MAIN SHA / FINAL QA`.
+**PR / canonical main commit**: `#571` / `6198dc39`.
 
 **Evidence state**:
-- Source: `pending`
+- Source: `merged`
 - Checks: `npx tsc -p convex --noEmit --pretty false` -> 0 errors. `npx vitest run
   convex/domains/redesign/chatRuns.responseShape.test.ts` -> 16 passed (3 new: grounded
   citation kept, cached-label citation rewritten, uncited superlative rewritten). Gate
@@ -63,10 +105,10 @@ names in UI / provider names appear only in the trace" - the opposite of the dis
 contract shipped in #550 - and pinned `DEFAULT_TIERS` to the runtime's `modelForTier` with
 a parity test, since the two were hand-maintained mirrors with no test binding them.
 
-**PR / canonical main commit**: `PENDING #NNN MAIN SHA / FINAL QA`.
+**PR / canonical main commit**: `#565` / `b2d12ae0`.
 
 **Evidence state**:
-- Source: `pending`
+- Source: `merged`
 - Checks: `npx tsc --noEmit --pretty false` -> 0 errors. `npx vitest run
   convex/domains/redesign/chatRuns.responseShape.test.ts
   src/features/redesign/components/UniversalComposer.test.tsx` -> 16 passed. Fix reverted

convex/domains/redesign/chatRuns.responseShape.test.ts

@@ -63,6 +63,91 @@ describe("redesign chat runtime response policy", () => {
     }
   });
 
+  // Issue #569 — every explicit shape beyond title/bullets used to fall
+  // through to the five-section memo. These scenarios model an analyst
+  // pasting exact-output constraints into the composer.
+  it("detects single-sentence, single-paragraph, and word-limit requests", () => {
+    expect(detectRequestedResponseShape("Summarize the brief in one sentence.")).toEqual({ kind: "sentence" });
+    expect(detectRequestedResponseShape("Give me just one sentence on the risk.")).toEqual({ kind: "sentence" });
+    expect(detectRequestedResponseShape("Write a one-sentence summary.")).toEqual({ kind: "sentence" });
+    expect(detectRequestedResponseShape("Explain the funding round in a single paragraph.")).toEqual({ kind: "paragraph" });
+    expect(detectRequestedResponseShape("One-paragraph answer please.")).toEqual({ kind: "paragraph" });
+    expect(detectRequestedResponseShape("Summarize in under 50 words.")).toEqual({ kind: "word_limit", limit: 50 });
+    expect(detectRequestedResponseShape("Answer in 30 words or fewer.")).toEqual({ kind: "word_limit", limit: 30 });
+    expect(detectRequestedResponseShape("No more than 80 words.")).toEqual({ kind: "word_limit", limit: 80 });
+  });
+
+  it("keeps memo for incidental or degenerate shape phrasing", () => {
+    // The memo template itself says "one sentence" per heading — user prose
+    // that merely mentions sentences/paragraphs must not flip the shape.
+    expect(detectRequestedResponseShape("The first sentence of their pitch is misleading.")).toEqual({ kind: "memo" });
+    expect(detectRequestedResponseShape("Compare the opening paragraph across both reports.")).toEqual({ kind: "memo" });
+    // Degenerate or absurd limits never bind.
+    expect(detectRequestedResponseShape("Summarize in under 2 words.")).toEqual({ kind: "memo" });
+    expect(detectRequestedResponseShape("Summarize in under 90000 words.")).toEqual({ kind: "memo" });
+    // Bullets outrank a word limit when both appear.
+    expect(
+      detectRequestedResponseShape("Give me exactly 3 bullets, under 50 words."),
+    ).toEqual({ kind: "bullets", count: 3 });
+  });
+
+  it("returns exactly one sentence while omitting memo-only fields", () => {
+    // idx binds the [1] marker to the URL-backed row so the #571 superlative
+    // gate recognizes the claim as grounded and leaves the sentence intact.
+    const shaped = applyDeterministicResponsePolicy(
+      "Summarize the brief in one sentence.",
+      memo,
+      [{ idx: 1, source: "https://example.com/acme", quote: "Acme launched a new product." }],
+    );
+    expect(shaped.shortAnswer).toBe("Acme is the strongest supported claim in the current packet. [1]");
+    expect(shaped.whyItMatters).toBe("");
+    expect(shaped.risks).toEqual([]);
+    expect(shaped.nextAction).toBe("");
+  });
+
+  it("collapses the memo into one paragraph and keeps the source-needed limitation inline when unsupported", () => {
+    const supported = applyDeterministicResponsePolicy(
+      "Explain it in a single paragraph.",
+      memo,
+      [{ source: "https://example.com/acme", quote: "Acme launched a new product." }],
+    );
+    expect(supported.shortAnswer).not.toContain("\n");
+    expect(supported.whyItMatters).toBe("");
+
+    const unsupported = applyDeterministicResponsePolicy(
+      "Explain it in a single paragraph.",
+      memo,
+      [{ source: "Setup" }],
+    );
+    // Honesty survives the compact shape: the limitation rides inside the
+    // paragraph because compact renders hide the risks section entirely.
+    expect(unsupported.shortAnswer).toContain("Source needed");
+    expect(unsupported.risks).toEqual([]);
+  });
+
+  it("enforces an explicit word limit deterministically, including the honesty prefix", () => {
+    const longMemo = {
+      ...memo,
+      shortAnswer:
+        "Acme closed the round. The filing lists twelve investors across three continents and two follow-on commitments.",
+    };
+    const shaped = applyDeterministicResponsePolicy(
+      "Summarize in under 8 words.",
+      longMemo,
+      [{ source: "https://example.com/acme", quote: "Acme closed the round." }],
+    );
+    expect(shaped.shortAnswer.split(/\s+/).length).toBeLessThanOrEqual(8);
+    expect(shaped.whyItMatters).toBe("");
+
+    const unsupported = applyDeterministicResponsePolicy(
+      "Summarize in under 8 words.",
+      longMemo,
+      [{ source: "Setup" }],
+    );
+    expect(unsupported.shortAnswer.startsWith("Source needed:")).toBe(true);
+    expect(unsupported.shortAnswer.split(/\s+/).length).toBeLessThanOrEqual(8);
+  });
+
   it("does not mistake incidental prose about titles for a shape request", () => {
     expect(
       detectRequestedResponseShape("Summarize the report and explain why its title is misleading."),