fix(redesign): continue reproducible answers in live chat

AGENT_COORDINATION.md

@@ -57,6 +57,12 @@ Server Error) for an unknown slug.
 
 ## Active claims (who is editing what RIGHT NOW)
 
+- **2026-07-16 · Codex `/root` →** `convex/schema.ts#redesignChatRuns`,
+  `convex/domains/redesign/chatRuns.ts#continuation-context`,
+  `src/features/redesign/{pages/ReproducibleChatPage,surfaces/ChatSurface,RedesignShell}` ·
+  add durable, privacy-redacted receipt-to-live-chat continuation ·
+  branch `fix/receipt-continue-chat`, PR #557. CI deploy only; no out-of-band Convex deploy.
+
 - **2026-06-03 · Claude →** `convex/*#handoff-token`, `src/.../ScratchnodePrivateBridge`,
   `src/App.tsx#events-private-route`, `public/proto/home-v5.html#private-handoff` ·
   shipping the cross-domain private-notes token bridge (opaque stateful token, PR #496) ·
@@ -68,6 +74,13 @@ Server Error) for an unknown slug.
 
 ## Hand-offs (built + ready for the other agent to call)
 
+- **2026-07-16 - Codex `/root` ->** Receipt continuation contract is additive and ready
+  on `fix/receipt-continue-chat` (PR #557): `startChat` accepts optional
+  `conversationContext: Array<{role: "user" | "assistant"; text: string; sourceUrls?: string[]}>`
+  plus `parentRunHash?: string`. Context is bounded and sanitized before persistence and
+  grounding. Public `getByHash` receipts deliberately redact both fields so a shared hash
+  never exposes a private follow-up transcript.
+
 - **2026-07-16 - Codex `/root` -> frontend/runtime agents** - additive optional
   `redesignChatRuns.clientRequestId/provider/runtimeReceiptId/cancelRequestedAt/cancelledAt` fields;
   `startChat({ ..., clientRequestId? })` dedupes by owner + request key;

CHANGELOG/integrations/pipeline-runtime.md

@@ -3,6 +3,13 @@
 Append-only lane for pipeline launch, activity, streaming, evaluation, schedule,
 and secret-gated MCP bridge ownership contracts. Newest entries first.
 
+## 2026-07-16 - Ground chat follow-ups in bounded receipt context
+
+The pending candidate extends redesign chat runs with optional, sanitized conversation
+turns and a parent receipt hash. The live action includes this transcript as context while
+requiring factual claims to be re-grounded, and public hash receipts redact the private
+continuation fields. See [`../pages/redesign-chat.md`](../pages/redesign-chat.md).
+
 ## 2026-07-15 — Derive pipeline ownership on the server
 
 The pending candidate removes browser-selected `ownerKey` authority from public pipeline launch, history, detail, bundle, stream, scorecard, and schedule APIs. Cost-bearing single and composed launches and all schedule controls now require an authenticated server identity. Guest history, detail, bundle, stream, and evaluation reads require an anonymous-session possession credential; schedule changes verify row ownership, public responses omit owner keys, and cron or secret-gated MCP work uses explicit internal service contracts. Legacy anonymous schedules do not execute.

CHANGELOG/pages/redesign-chat.md

@@ -0,0 +1,31 @@
+# Redesign Chat
+
+Append-only lane for the public redesign chat, reproducible answer receipts, and their
+transition into an authenticated live conversation. Newest entries first.
+
+## 2026-07-16 - Continue an immutable answer in live chat
+
+Reproducible answer receipts now separate two intents that were previously collapsed:
+`Re-run prompt` starts a fresh reproducible run, while `Continue in chat` opens the real
+chat surface with the receipt's prompt, answer, and source lineage already visible. The
+composer remains available beneath that context, so the receipt is no longer a dead end.
+
+Follow-up runs persist a bounded, role-aware conversation context and parent receipt hash.
+The backend sanitizes this context before grounding the next answer and strips both fields
+from public hash lookups, preventing a shared receipt URL from leaking private follow-up
+conversation history. Reload recovery reconstructs the same transcript from stored run
+context instead of inventing UI-only state.
+
+The release also excludes the route-lazy markdown editor chunk from Workbox precaching;
+it remains network loaded and runtime cached, avoiding an unrelated 2 MiB precache ceiling
+from blocking production builds as dependency versions move.
+
+**PR / canonical main commit**: pending CI-gated candidate.
+
+**Evidence state**:
+- Root TypeScript and 14 focused continuation/chat contract tests passed locally.
+- Production build reached the full client bundle; final release proof is recorded on the PR.
+- Responsive receipt-to-chat browser evidence remains outside git under `.qa/evidence/`.
+
+**Author**: Homen Shum + Codex.
+**Touches**: [`../integrations/pipeline-runtime.md`](../integrations/pipeline-runtime.md).

convex/domains/redesign/chatRuns.ts

@@ -69,6 +69,12 @@ type TraceRow = {
   durationMs: number;
 };
 
+export type ConversationContextTurn = {
+  role: "user" | "assistant";
+  text: string;
+  sourceUrls?: string[];
+};
+
 interface AnswerPacket {
   shortAnswer: string;
   whyItMatters: string;
@@ -143,6 +149,9 @@ type RuntimeContextPacket = {
 // ───────── Constants ─────────
 
 const MAX_PROMPT_CHARS = 4_000;
+const MAX_CONVERSATION_TURNS = 10;
+const MAX_CONVERSATION_TURN_CHARS = 2_000;
+const MAX_CONVERSATION_SOURCE_URLS = 5;
 const TIMEOUT_MS = 45_000;
 const FALLBACK_SOURCE_TIMEOUT_MS = 12_000;
 const FALLBACK_SOURCE_LIMIT = 5;
@@ -176,11 +185,34 @@ async function assertRunReadable(ctx: any, runId: string): Promise<any> {
   return row;
 }
 
-function redactSharedRun(row: any): any {
+export function sanitizeConversationContext(
+  rows: ConversationContextTurn[] | undefined,
+): ConversationContextTurn[] | undefined {
+  if (!rows?.length) return undefined;
+  const normalized = rows
+    .slice(-MAX_CONVERSATION_TURNS)
+    .map((row) => ({
+      role: row.role,
+      text: row.text.trim().slice(0, MAX_CONVERSATION_TURN_CHARS),
+      ...(row.sourceUrls?.length
+        ? {
+            sourceUrls: row.sourceUrls
+              .filter((url) => /^https?:\/\//i.test(url))
+              .slice(0, MAX_CONVERSATION_SOURCE_URLS),
+          }
+        : {}),
+    }))
+    .filter((row) => row.text.length > 0);
+  return normalized.length ? normalized : undefined;
+}
+
+export function redactSharedRun(row: any): any {
   if (!row) return null;
   const {
     userId: _userId,
     clientRequestId: _clientRequestId,
+    conversationContext: _conversationContext,
+    parentRunHash: _parentRunHash,
     cancelRequestedAt: _cancelRequestedAt,
     ...safeRow
   } = row;
@@ -1030,6 +1062,13 @@ export const startChat = mutation({
       text: v.string(),
       source: v.optional(v.string()),
     }))),
+    /** Bounded prior transcript. Persisted privately for reload recovery. */
+    conversationContext: v.optional(v.array(v.object({
+      role: v.union(v.literal("user"), v.literal("assistant")),
+      text: v.string(),
+      sourceUrls: v.optional(v.array(v.string())),
+    }))),
+    parentRunHash: v.optional(v.string()),
     /** Phase 5 — counterfactual probe. When set, the run is a probe
      *  re-evaluation of an earlier run with the cited source masked. */
     probeOriginRunId: v.optional(v.string()),
@@ -1053,6 +1092,8 @@ export const startChat = mutation({
       if (existing) return existing.runId;
     }
     const normalizedTier = normalizeChatTier(args.tier);
+    const conversationContext = sanitizeConversationContext(args.conversationContext);
+    const parentRunHash = args.parentRunHash?.trim().slice(0, 80);
     const model = modelForTier(normalizedTier);
     const runId = generateRunId();
     await ctx.db.insert("redesignChatRuns", {
@@ -1064,6 +1105,8 @@ export const startChat = mutation({
       model,
       provider: "google-gemini",
       runtimeReceiptId: `redesign-chat:${runId}`,
+      ...(conversationContext ? { conversationContext } : {}),
+      ...(parentRunHash ? { parentRunHash } : {}),
       status: "pending",
       createdAt: Date.now(),
     });
@@ -1074,6 +1117,8 @@ export const startChat = mutation({
       contextRef: args.contextRef,
       model,
       pinnedClaims: args.pinnedClaims,
+      conversationContext,
+      parentRunHash,
       probeOriginRunId: args.probeOriginRunId,
       probeMaskedSourceUrl: args.probeMaskedSourceUrl,
       probeMaskedSourceIdx: args.probeMaskedSourceIdx,
@@ -1337,6 +1382,12 @@ export const runStreamingChat = internalAction({
       text: v.string(),
       source: v.optional(v.string()),
     }))),
+    conversationContext: v.optional(v.array(v.object({
+      role: v.union(v.literal("user"), v.literal("assistant")),
+      text: v.string(),
+      sourceUrls: v.optional(v.array(v.string())),
+    }))),
+    parentRunHash: v.optional(v.string()),
     /** Phase 5 — counterfactual probe origin */
     probeOriginRunId: v.optional(v.string()),
     probeMaskedSourceUrl: v.optional(v.string()),
@@ -1469,6 +1520,12 @@ export const runStreamingChat = internalAction({
       const pinnedSection = args.pinnedClaims && args.pinnedClaims.length > 0
         ? `\n\nPinned claims (carry forward as established context — do not contradict without explicit re-grounding):\n${args.pinnedClaims.map((p, i) => `  ${i + 1}. ${p.text}${p.source ? ` (source: ${p.source})` : ""}`).join("\n")}`
         : "";
+      const conversationSection = args.conversationContext && args.conversationContext.length > 0
+        ? `\n\nPrior conversation (context only; preserve the user's intent, but re-ground factual claims when needed):\n${args.conversationContext.map((turn, i) => {
+            const sources = turn.sourceUrls?.length ? `\n     sources: ${turn.sourceUrls.join(", ")}` : "";
+            return `  ${i + 1}. ${turn.role.toUpperCase()}: ${turn.text}${sources}`;
+          }).join("\n")}`
+        : "";
       // Phase 5 — counterfactual probe instruction
       const probeSection = args.probeMaskedSourceUrl
         ? `\n\nIMPORTANT — counterfactual probe: The source previously at <${args.probeMaskedSourceUrl}> (originally cited as [${args.probeMaskedSourceIdx ?? "?"}] in run ${args.probeOriginRunId ?? "?"}) is being treated as UNRELIABLE for this answer. DO NOT cite it. DO NOT use it as the basis for any claim. Re-answer the same prompt and explicitly note in "Risks / unknowns" how the conclusion changes (or holds) if that source is excluded. Prefer alternative grounded sources.`
@@ -1509,14 +1566,21 @@ ${citationInstruction} ${liveGrounding.useLiveGrounding ? "Keep claims grounded
 
 ${calculationInstruction}
 
-Context: ${JSON.stringify(contextBundle)}${pinnedSection}${probeSection}${verifiedCalculationSection}`;
+Context: ${JSON.stringify(contextBundle)}${conversationSection}${pinnedSection}${probeSection}${verifiedCalculationSection}`;
       // Emit a stage event so the UI can show "Probing without [N]" / "Carrying forward N pins"
       if (probeSection) {
         await append("stage", { stage: "probe", maskedUrl: args.probeMaskedSourceUrl, maskedIdx: args.probeMaskedSourceIdx, originRunId: args.probeOriginRunId });
       }
       if (pinnedSection) {
         await append("stage", { stage: "pinned", count: args.pinnedClaims!.length });
       }
+      if (conversationSection) {
+        await append("stage", {
+          stage: "conversation_context",
+          count: args.conversationContext!.length,
+          parentRunHash: args.parentRunHash,
+        });
+      }
 
       const controller = new AbortController();
       const timeoutId = setTimeout(() => controller.abort(), TIMEOUT_MS);

convex/schema.ts

@@ -16026,6 +16026,14 @@ export default defineSchema({
     /** Honest runtime receipt metadata; no provider response id is inferred. */
     provider: v.optional(v.string()),
     runtimeReceiptId: v.optional(v.string()),
+    /** Private, bounded transcript supplied to a continuation run. Never returned by getByHash. */
+    conversationContext: v.optional(v.array(v.object({
+      role: v.union(v.literal("user"), v.literal("assistant")),
+      text: v.string(),
+      sourceUrls: v.optional(v.array(v.string())),
+    }))),
+    /** Reproducible receipt this run continued from. Optional for existing rows. */
+    parentRunHash: v.optional(v.string()),
     cancelRequestedAt: v.optional(v.number()),
     cancelledAt: v.optional(v.number()),
     createdAt: v.number(),

src/features/redesign/RedesignShell.tsx

@@ -47,6 +47,7 @@ import { WorkspaceSurface } from "./surfaces/WorkspaceSurface";
 import { ReproducibleChatPage } from "./pages/ReproducibleChatPage";
 import { useLiveArtifacts, type LiveArtifactDetail } from "./hooks/useLiveArtifacts";
 import type { ReportCardData, SurfaceId } from "./fixtures";
+import { parseChatLaunchParams } from "./lib/chatContinuation";
 
 const PATH_TO_SURFACE: Record<string, SurfaceId | "workspace"> = {
   "": "home",
@@ -75,11 +76,6 @@ function pathToChatHash(pathname: string): string | null {
   return match?.[1] ?? null;
 }
 
-function queryPrompt(search: string): string {
-  const params = new URLSearchParams(search);
-  return params.get("q")?.trim() ?? "";
-}
-
 function normalizeEntityKey(value: string): string {
   return value.toLowerCase().replace(/[^a-z0-9]+/g, "");
 }
@@ -117,7 +113,7 @@ export default function RedesignShell() {
   const reportId = useMemo(() => pathToReportId(location.pathname), [location.pathname]);
   const chatHash = useMemo(() => pathToChatHash(location.pathname), [location.pathname]);
   const workspace = useMemo(() => workspaceParams(location.search), [location.search]);
-  const initialChatPrompt = useMemo(() => queryPrompt(location.search), [location.search]);
+  const chatLaunch = useMemo(() => parseChatLaunchParams(location.search), [location.search]);
   const focusHomeComposer = useMemo(
     () => new URLSearchParams(location.search).get("focus") === "home-composer",
     [location.search],
@@ -360,7 +356,8 @@ export default function RedesignShell() {
               )}
               {!isPrototypeKit && surface === "chat" && (
                 <ChatSurface
-                  initialPrompt={initialChatPrompt}
+                  initialPrompt={chatLaunch.prompt}
+                  continuationHash={chatLaunch.continuationHash}
                   onActiveContextChange={setActiveChatDetail}
                   onAgentRailChange={setActiveChatAgentRail}
                 />

src/features/redesign/agent-workspace.css

@@ -37,20 +37,61 @@
 [data-redesign] .rd-chat-workspace {
   display: grid;
   grid-template-rows: minmax(0, 1fr) auto auto auto;
+  width: 100%;
+  min-width: 0;
   height: 100%;
   min-height: 0;
   overflow: hidden;
   background: var(--rd-paper);
 }
 
+[data-redesign] .rd-chat-workspace > * {
+  min-width: 0;
+}
+
+[data-redesign] .rd-chat-workspace[data-continuation="true"] {
+  grid-template-rows: auto minmax(0, 1fr) auto auto auto;
+}
+
+[data-redesign] .rd-continuation-context {
+  width: min(100% - 32px, 840px);
+  box-sizing: border-box;
+  margin: 12px auto 0;
+  padding: 10px 12px;
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  gap: 16px;
+  border: 1px solid var(--rd-accent-ring);
+  border-radius: var(--rd-r-md);
+  background: var(--rd-accent-tint);
+  color: var(--rd-ink);
+}
+
+[data-redesign] .rd-continuation-context p {
+  margin: 3px 0 0;
+  color: var(--rd-ink-mute);
+  font-size: 12px;
+  line-height: 1.4;
+}
+
+[data-redesign] .rd-continuation-context[data-state="unavailable"] {
+  border-color: var(--rd-amber-ring, var(--rd-line));
+  background: var(--rd-amber-tint, var(--rd-panel));
+}
+
 [data-redesign] .rd-chat-transcript {
+  width: 100%;
+  min-width: 0;
   min-height: 0;
   height: 100%;
   overflow: hidden;
 }
 
 [data-redesign] .rd-chat-transcript__content {
   width: min(100%, 840px);
+  min-width: 0;
+  box-sizing: border-box;
   min-height: 100%;
   margin: 0 auto;
   padding: clamp(20px, 4vw, 44px) clamp(18px, 4vw, 40px) 48px;
@@ -71,6 +112,9 @@
 
 [data-redesign] .rd-composer-dock {
   position: relative;
+  width: 100%;
+  min-width: 0;
+  box-sizing: border-box;
   bottom: auto;
   z-index: 5;
   padding: 10px clamp(14px, 4vw, 40px) max(14px, env(safe-area-inset-bottom));
@@ -81,6 +125,7 @@
 
 [data-redesign] .rd-composer-shell {
   width: min(100%, 840px);
+  min-width: 0;
   margin: 0 auto;
 }
 
@@ -183,6 +228,14 @@
 }
 
 @media (max-width: 760px) {
+  [data-redesign] .rd-continuation-context {
+    width: calc(100% - 24px);
+    margin-top: 8px;
+    align-items: flex-start;
+    flex-direction: column;
+    gap: 8px;
+  }
+
   [data-redesign] .rd-run-scope > summary span:nth-of-type(2),
   [data-redesign] .rd-run-scope > summary span:nth-of-type(3) {
     display: none;