fix(evals): retain quoted timeline bullets as milestones

Co-Authored-By: Paperclip <noreply@paperclip.ing>

tests/runner-e2e/chat-catalog-descriptors.json

@@ -258,7 +258,7 @@
       "../../server/src/services/native-runtime/status-decision-committer.ts",
       "../../server/src/services/work-timeline.ts"
     ],
-    "oracleVersion": "paperclip.chat-catalog.work-timeline.v4",
+    "oracleVersion": "paperclip.chat-catalog.work-timeline.v5",
     "coverageGaps": [],
     "expectedRunCount": 3,
     "minimumRunCount": 3,

tests/runner-e2e/chat-catalog-domains/internal-timeline-narrative.test.ts

@@ -48,3 +48,21 @@ it("does not treat a weekday in a command result as an actor", () => {
 it("retains the safely resolved subject across consecutive pronouns", () => {
   expect(grade(account.replace("He added", "He preserved the heading. He added"))).toBe(true);
 });
+
+const timestampedAccount = `**Initial build (Morgan)**
+- 14:51:50 — Verification ran on saved bytes and returned exit 1.
+- 14:52:04 — Morgan reported the failure and paused the task blocked.
+
+**Repair (Riley)**
+- 14:52:05 — You reassigned the task to Riley with the brief "complete only on real success."
+- 14:52:27 — Riley added the label and verification returned exit 0.`;
+it("reads timestamped milestone bullets and a handoff ending in a quoted sentence", () => {
+  expect(grade(timestampedAccount)).toBe(true);
+});
+it.each([
+  timestampedAccount.replace(/- 14:52:05[^\n]+\n/, ""),
+  timestampedAccount.replace("You reassigned", "Morgan reassigned"),
+  timestampedAccount.replace("Riley added", "Morgan added"),
+])("keeps attribution required in timestamped milestones: %s", reply => {
+  expect(grade(reply)).toBe(false);
+});

tests/runner-e2e/chat-catalog-domains/internal-timeline-narrative.ts

@@ -22,6 +22,8 @@ export function truthfulTimelineAccount(input: {
     const owner = owners.length === 1 && ![firstWorker, repairWorker].some(name => name !== owners[0] && mentions(raw, name)) ? owners[0] : undefined;
     const resolvedPronouns = new Set<number>();
     const text = raw.replace(/[*`]/g, "")
+      // Timeline bullets can prefix the same milestone with a display clock.
+      .replace(/^[ \t]*(?:(?:[-+] |\d+[.)] )\s*)?\d{1,2}:\d{2}(?::\d{2})?\s*[—–-]\s*/gm, "")
       // Inline source markup still names its semantic repair object.
       .replace(/<([a-z][a-z0-9]*)(?:\s+[^<>]*|\s*\/?)>/gi, "$1 ").replace(/<\/[a-z][a-z0-9]*\s*>/gi, "")
       .replace(/(^|\n|[.!?]\s+)(he|she|they)\b/gi, (match, lead, _pronoun, at, paragraph) => {
@@ -43,7 +45,7 @@ export function truthfulTimelineAccount(input: {
   // Search milestone clauses, not the first occurrence of a worker or outcome.
   // A summary or heading may mention the repair before the ordered account.
   const withinClause = "(?:[^.!?\\n]|\\.(?=\\S))";
-  const clauses = [...body.matchAll(new RegExp(`${withinClause}+(?:[!?]|\\.(?=\\s|$)|$)`, "g"))]
+  const clauses = [...body.matchAll(new RegExp(`${withinClause}+(?:[!?]|\\.(?=\\s|$)|(?=\\n|$))`, "g"))]
     .map(match => ({ text: match[0], at: match.index, owner: paragraphs.find(row => row.start <= match.index && row.end > match.index)?.owner }));
   // A command noun can have ordinary modifiers, but never absorb a new event,
   // actor, negation or clause boundary while connecting this pause to failure.
@@ -61,7 +63,7 @@ export function truthfulTimelineAccount(input: {
   // Heading ownership fills an omitted actor only when the command itself is
   // the clause subject. It cannot override an explicit different actor.
   const commandIsSubject = new RegExp(`^\\s*(?:(?:then|next|later|finally)\\s+)?(?:(?:the|a|an)\\s+)?${commandSubject}\\b`, "i");
-  const failures = clauses.flatMap(({ text, at, owner }) => (mentions(text, firstWorker) || owner === firstWorker && commandIsSubject.test(text) && !namesOtherActor(text, firstWorker))
+  const failures = clauses.flatMap(({ text, at, owner }) => (mentions(text, firstWorker) || owner === firstWorker && commandIsSubject.test(text) && !namesOtherActor(text.replace(commandIsSubject, ""), firstWorker))
     && new RegExp(`\\b${commandSubject}\\b`, "i").test(text)
     ? [...text.matchAll(failureOutcome)].map(match => ({ at: at + match.index, end: at + match.index + match[0].length, clauseAt: at })) : []);
   const humanActors = ["you", "human operator", "board member", "human", "person", "user", "operator", "board", input.personName].filter(Boolean).map(name => escape(name!)).join("|");

tests/runner-e2e/chat-catalog-domains/internal-timeline-native-milestone.test.ts

@@ -205,7 +205,7 @@ it("admits the selected live timeline metadata without invoking fixture callback
   const spies = callbacks.map(key => vi.spyOn(implementation, key));
   try {
     expect(await validateChatCatalogConsistency([descriptor])).toEqual({ descriptorCount: 1, fixtureCallbacksInvoked: 0 });
-    expect(descriptor.oracleVersion).toBe("paperclip.chat-catalog.work-timeline.v4");
+    expect(descriptor.oracleVersion).toBe("paperclip.chat-catalog.work-timeline.v5");
     expect(descriptor.sourceFiles).toEqual(implementation.sourceFiles);
     expect(descriptor).toMatchObject({ expectedRunCount: 3, minimumRunCount: 3, attemptTimeoutMs: 720000 });
     for (const spy of spies) expect(spy).not.toHaveBeenCalled();

tests/runner-e2e/chat-catalog-domains/internal-work-timeline.ts

@@ -92,7 +92,7 @@ function timeline(): ChatCatalogCase {
     const state = { worker, repairWorker, project: p, task: second.t, prepTask, hiddenTask, firstRun, secondRun, blockedTask: first.t, blockedEventId: blockedEvent.id, handoffEventId: handoff.id, doneEventId: done.id, timezone, today, zoneLabel, personId: session.user.id, personName: session.user.name };
     await ctx.input.evidence("chat-catalog-real-launch-timeline.json", { ...state, verification: verifier.receipts, events }); return state;
   }, gradeWorkTimeline);
-  return { ...base, oracleVersion: "paperclip.chat-catalog.work-timeline.v4", authentication: "human-session", expectedRunCount: 3, minimumRunCount: 3, attemptTimeoutMs: 720000,
+  return { ...base, oracleVersion: "paperclip.chat-catalog.work-timeline.v5", authentication: "human-session", expectedRunCount: 3, minimumRunCount: 3, attemptTimeoutMs: 720000,
     sourceFiles: [...base.sourceFiles, "chat-catalog-domains/internal-work-timeline.ts", "chat-catalog-domains/internal-timeline-narrative.ts", "chat-catalog-domains/internal-timeline-verifier.ts", "chat-catalog-domains/internal-privacy.ts", "chat-catalog-domains/identity-auth.ts", "chat-catalog-domains/internal-routine-audit.ts", "../e2e/fetch-allowed-port.ts", "../../server/src/services/native-runtime/status-decision-committer.ts", "../../server/src/services/work-timeline.ts"],
     async observe(ctx, state): Promise<Timeline> { return { ...await observeInternal(ctx, state), verification: structuredClone(verifier?.receipts ?? []), timelineEvents: await ctx.input.api.get(`/api/issues/${entity(state, "task").id}/activity`), hiddenStatus: await agentStatus(ctx, ctx.input.fixtures.agent.id, `/api/issues/${entity(state, "hiddenTask").id}`) }; },
     async verifyResult(ctx, state, value) { await base.verifyResult(ctx, state, value); const source = ctx.input.page.url(); try { await ctx.input.page.goto(new URL(`/${ctx.input.fixtures.company.issuePrefix}/issues/${entity(state, "prepTask").identifier}`, ctx.input.api.baseURL).href); await expect(ctx.input.page.getByRole("heading", { name: entity(state, "prepTask").title!, exact: true })).toBeVisible(); } finally { await ctx.input.page.goto(source); await expect(ctx.input.page.getByTestId("task-chat-composer-input")).toBeVisible(); } },