feat: connect prepared run and refine presenter notes

2026-07-14 08:59:32 +07:00
parent 5ba9e55f92
commit 184e41858c
26 changed files with 296 additions and 114 deletions
@@ -80,15 +80,16 @@ export const GuidedProductMoment = ({
  const facts = projectDemoRunFacts(demo);
  const runResume = demo.state.events.find((event) => event.stage === "run_resume");
  const revisionRequested = approvalActions?.state === "revision_requested";
+  const isReplay = demo.state.mode === "replay";
  // Replay revision evidence has a separate recording identity; do not let
  // the normal submitted-branch lens imply continuity for that branch.
  const eyebrow = revisionRequested
-    ? demo.state.mode === "replay"
+    ? isReplay
      ? "Prepared branch"
      : "Live branch"
    : lens.eyebrow;
  const headline = revisionRequested
-    ? demo.state.mode === "replay"
+    ? isReplay
      ? "Separate prepared revision recording"
      : "Live revision branch resumes the run"
    : lens.headline;
@@ -106,7 +107,7 @@ export const GuidedProductMoment = ({
      <header className="guided-product-moment__header">
        <span>{eyebrow}</span>
        <strong>{headline}</strong>
-        <p>{statusCopy(moment, approvalActions, demo.state.mode === "replay")}</p>
+        <p>{statusCopy(moment, approvalActions, isReplay)}</p>
      </header>

      <div className="guided-product-moment__primary">
@@ -268,7 +268,8 @@ describe("PresentationRoute", () => {
      .toHaveAttribute("data-phase", "deployment");
  });

-  it("records a Deployment run request locally without an RPC", async () => {
+  it("starts the shared prepared workflow action without advancing the Deployment beat", async () => {
+    window.sessionStorage.setItem("lda.workflowConsole.target", "http://127.0.0.1:8765/rpc");
    window.location.hash = "#scene/prepared-lifecycle/deployment";
    const { PresentationRoute } = await import("./PresentationRoute.js");
    render(<PresentationRoute />);
@@ -277,11 +278,16 @@ describe("PresentationRoute", () => {

    await userEvent.click(screen.getByRole("button", { name: /send message/i }));

-    expect(await screen.findByRole("status")).toHaveTextContent(
-      "Run request prepared for the next execution slice.",
-    );
-    expect(mockedCallOperation.mock.calls.some(([operation]) => operation === "workflow.runs.start"))
-      .toBe(false);
+    expect(await screen.findByText("Prepared run receipt added. Continue when you are ready."))
+      .toBeInTheDocument();
+    expect(window.location.hash).toBe("#scene/prepared-lifecycle/deployment");
+    expect(screen.getByText(/three proposed issues are ready for review/i)).toBeInTheDocument();
+    await waitFor(() => {
+      expect(mockedCallOperation.mock.calls.some(([operation]) => operation === "workflow.deployments.inspect"))
+        .toBe(true);
+      expect(mockedCallOperation.mock.calls.some(([operation]) => operation === "workflow.runs.start"))
+        .toBe(true);
+    }, { timeout: 3000 });
  });

  it("keeps Diagnose submission on the current beat", async () => {
@@ -109,6 +109,11 @@ export const PresentationStage = ({
                motionDisabled={state.motionDisabled}
                approvalActions={approvalActions}
                onPreparedLifecycleAdvance={onPreparedLifecycleAdvance}
+                onRunPreparedWorkflow={
+                  demoRail.kind === "action" && timelineAgent
+                    ? () => timelineAgent.runPreparedWorkflow(demoRail.mode)
+                    : undefined
+                }
              />
            )}
          </section>
@@ -30,15 +30,16 @@ describe("RunInputFileBrowser", () => {

    const files = within(browser).getByRole("list", { name: /included in prepared run/i });
    expect(within(files).getAllByRole("listitem")).toHaveLength(2);
-    expect(within(files).getAllByText("selected")).toHaveLength(2);
+    expect(within(files).getAllByText("selected")).toHaveLength(1);
    expect(within(files).getAllByRole("button")).toHaveLength(2);
    expect(within(files).queryAllByRole("link")).toHaveLength(0);
    for (const path of ["docs/project-brief.md", "docs/architecture-notes.md"]) {
      const row = within(files).getByText(path).closest("li");
      expect(row).not.toBeNull();
      expect(row).toHaveAttribute("data-file-path", path);
-      expect(row).toHaveTextContent(/selected/i);
    }
+    expect(within(files).getByText("docs/project-brief.md").closest("li")).toHaveTextContent(/selected/i);
+    expect(within(files).getByText("docs/architecture-notes.md").closest("li")).not.toHaveTextContent(/selected/i);

    const destination = within(browser).getByRole("group", { name: /workflow output/i });
    expect(destination).toHaveTextContent("artifacts/issue-board.json");
@@ -57,6 +58,11 @@ describe("RunInputFileBrowser", () => {
    expect(preview).toHaveTextContent(/architecture-notes\.md/i);
    expect(preview).toHaveTextContent(/fixture preview/i);
    expect(preview).toHaveTextContent(/not execution evidence/i);
+    expect(screen.getByRole("button", { name: /architecture-notes\.md/i })).toHaveAttribute(
+      "aria-pressed",
+      "true",
+    );
+    expect(screen.getByRole("button", { name: /project-brief\.md/i })).not.toHaveTextContent(/selected/i);
  });

  it("shows an honest empty preview for files absent from the fixture catalog", () => {
@@ -34,7 +34,9 @@ export const RunInputFileBrowser = ({
              >
                <span className="run-input-file-browser__icon" aria-hidden="true">md</span>
                <code>{path}</code>
-                <span className="run-input-file-browser__marker">selected</span>
+                {path === visibleSelectedPath ? (
+                  <span className="run-input-file-browser__marker">selected</span>
+                ) : null}
              </button>
            </li>
          ))}
@@ -33,6 +33,7 @@ type SceneBodyProps = {
  readonly motionDisabled: boolean;
  readonly approvalActions?: DemoApprovalActions | undefined;
  readonly onPreparedLifecycleAdvance?: (() => void) | undefined;
+  readonly onRunPreparedWorkflow?: (() => Promise<void>) | undefined;
 };

 const workflowDemoSceneIds = new Set([
@@ -290,7 +291,7 @@ const assertNever = (value: never): never => {
  throw new Error(`Unexpected view: ${value}`);
 };

-export const SceneBody = ({ location, demo, selectedNodeId, selectNode, openEvidence, openDiscussion, onFocusPathChange, motionDisabled, approvalActions, onPreparedLifecycleAdvance }: SceneBodyProps) => {
+export const SceneBody = ({ location, demo, selectedNodeId, selectNode, openEvidence, openDiscussion, onFocusPathChange, motionDisabled, approvalActions, onPreparedLifecycleAdvance, onRunPreparedWorkflow }: SceneBodyProps) => {
  const sceneId = location.kind === "main" ? location.sceneId : "positioning";
  const beatId = location.kind === "main" ? location.beatId : "landscape";
  const scene = findScene(sceneId) ?? findScene("thesis")!;
@@ -331,6 +332,7 @@ export const SceneBody = ({ location, demo, selectedNodeId, selectNode, openEvid
          scene={scene}
          beat={beat}
          onAdvance={onPreparedLifecycleAdvance}
+          onRunPreparedWorkflow={onRunPreparedWorkflow}
          discussionRail={scene.id === "prepared-lifecycle" ? discussionLinks : undefined}
        />
      );
@@ -2,6 +2,7 @@ import { AssistantOperatorThread } from "../chat/AssistantOperatorThread.js";
 import {
  authoringToolGroupId,
  projectPreparedAuthoringThread,
+  projectPreparedRunRequestExchange,
  type AuthoringPhaseId,
 } from "./authoring-recording.js";
 import {
@@ -17,6 +18,7 @@ type AuthoringConversationProps = {
  readonly requestOverride?: string | undefined;
  readonly requestOverrides?: PreparedLifecycleSubmittedOverrides | undefined;
  readonly scrollMode?: "active" | "start" | undefined;
+  readonly runRequested?: string | null | undefined;
 };

 /** Renders the same prepared conversation at full-stage or compact-dock scale. */
@@ -27,17 +29,25 @@ export const AuthoringConversation = ({
  requestOverride,
  requestOverrides,
  scrollMode,
-}: AuthoringConversationProps) => (
-  <AssistantOperatorThread
-    mode={surface === "stage" ? "full" : "dock"}
-    surface={surface}
-    messages={projectPreparedAuthoringThread(
+  runRequested,
+}: AuthoringConversationProps) => {
+  const messages = [
+    ...projectPreparedAuthoringThread(
      recordingPhaseForStep(throughPhase),
      requestOverride,
      requestOverrides,
-    )}
-    activeToolGroupId={authoringToolGroupId(recordingPhaseForStep(activePhase))}
-    scrollMode={scrollMode}
-    ariaLabel="prepared authoring conversation"
-  />
-);
+    ),
+    ...(runRequested ? projectPreparedRunRequestExchange(runRequested) : []),
+  ];
+
+  return (
+    <AssistantOperatorThread
+      mode={surface === "stage" ? "full" : "dock"}
+      surface={surface}
+      messages={messages}
+      activeToolGroupId={authoringToolGroupId(recordingPhaseForStep(activePhase))}
+      scrollMode={runRequested ? "end" : scrollMode}
+      ariaLabel="prepared authoring conversation"
+    />
+  );
+};
@@ -7,7 +7,12 @@ import { PreparedAuthoringLifecycleScene } from "./PreparedAuthoringLifecycleSce

 afterEach(() => cleanup());

-const renderBeat = (beatId: string, onAdvance?: () => void, discussionRail?: ReactNode) => {
+const renderBeat = (
+  beatId: string,
+  onAdvance?: () => void,
+  discussionRail?: ReactNode,
+  onRunPreparedWorkflow?: () => Promise<void>,
+) => {
  const scene = findScene("prepared-lifecycle");
  const beat = findBeat("prepared-lifecycle", beatId);
  if (!scene || !beat) throw new Error(`missing prepared-lifecycle/${beatId}`);
@@ -16,6 +21,7 @@ const renderBeat = (beatId: string, onAdvance?: () => void, discussionRail?: Rea
      scene={scene}
      beat={beat}
      onAdvance={onAdvance}
+      onRunPreparedWorkflow={onRunPreparedWorkflow}
      discussionRail={discussionRail}
    />,
  );
@@ -173,6 +179,21 @@ describe("PreparedAuthoringLifecycleScene", () => {
    expect(onAdvance).toHaveBeenCalledTimes(advances ? 1 : 0);
  });

+  it("answers the terminal run request without advancing the presentation", async () => {
+    const user = userEvent.setup();
+    const onAdvance = vi.fn();
+    const onRunPreparedWorkflow = vi.fn().mockResolvedValue(undefined);
+    renderBeat("deployment", onAdvance, undefined, onRunPreparedWorkflow);
+
+    await user.click(screen.getByRole("button", { name: /send message/i }));
+
+    expect(screen.getByRole("textbox", { name: /message to authoring assistant/i })).toHaveValue("");
+    expect(screen.getByRole("button", { name: /run.*1 tool call/i })).toBeInTheDocument();
+    expect(screen.getByText(/three proposed issues are ready for review/i)).toBeInTheDocument();
+    expect(onRunPreparedWorkflow).toHaveBeenCalledOnce();
+    expect(onAdvance).not.toHaveBeenCalled();
+  });
+
  it.each([
    ["discover", "Discover"],
    ["draft", "Draft"],
@@ -18,6 +18,7 @@ type PreparedAuthoringLifecycleSceneProps = {
  readonly scene: SceneDefinition;
  readonly beat: SceneBeatDefinition;
  readonly onAdvance?: (() => void) | undefined;
+  readonly onRunPreparedWorkflow?: (() => Promise<void>) | undefined;
  readonly discussionRail?: ReactNode;
 };

@@ -40,7 +41,7 @@ const steps = [
 * Each beat shows a persistent prepared assistant beside one dominant phase
 * projection sourced from the prepared authoring recording.
 */
-export const PreparedAuthoringLifecycleScene = ({ scene, beat, onAdvance, discussionRail }: PreparedAuthoringLifecycleSceneProps) => {
+export const PreparedAuthoringLifecycleScene = ({ scene, beat, onAdvance, onRunPreparedWorkflow, discussionRail }: PreparedAuthoringLifecycleSceneProps) => {
  const [messageState, dispatch] = useReducer(
    preparedLifecycleMessageReducer,
    initialPreparedLifecycleMessageState,
@@ -81,7 +82,12 @@ export const PreparedAuthoringLifecycleScene = ({ scene, beat, onAdvance, discus
            dispatch({ type: "artifact_submitted" });
            onAdvance?.();
          }
-          if (step === "deployment") dispatch({ type: "run_requested" });
+          if (step === "deployment") {
+            dispatch({ type: "run_requested" });
+            // This is the same mode-aware action as the footer control; the
+            // presentation remains on this beat so the presenter advances it.
+            void onRunPreparedWorkflow?.();
+          }
        }}
      />
      <div className="prepared-lifecycle-scene__presentation">
@@ -41,7 +41,8 @@ describe("PresentationAssistantPane", () => {
    );
    expect(screen.getByRole("heading", { name: /authoring assistant/i })).toBeInTheDocument();
    expect(screen.getByText(/current phase: validate/i)).toBeInTheDocument();
-    expect(screen.getByText(/prepared replay only/i)).toBeInTheDocument();
+    expect(screen.getByText(/prepared assistant transcript/i)).toBeInTheDocument();
+    expect(screen.getByText(/final run request uses the configured demo target/i)).toBeInTheDocument();
  });

  it("exposes an explicit support role when composed in the lifecycle scene", () => {
@@ -129,6 +130,15 @@ describe("PresentationAssistantPane", () => {
    cleanup();
    renderPane("deployment", { runRequested: "Run this deployment" });
    expect(screen.getByRole("button", { name: /send message/i })).toBeDisabled();
+    expect(screen.getByText("Run this deployment")).toBeInTheDocument();
+    expect(screen.getByText(/starting lda_report_case_study\.default with the prepared input/i))
+      .toBeInTheDocument();
+    const runTools = screen.getByRole("button", { name: /run.*1 tool call/i });
+    expect(runTools).toHaveAttribute("aria-expanded", "false");
+    await user.click(runTools);
+    await user.click(screen.getByRole("button", { name: /startRun/i }));
+    expect(screen.getByText(/run_recorded_lda_report/i)).toBeInTheDocument();
+    expect(screen.getByText(/paused at the typed issue-review boundary/i)).toBeInTheDocument();
  });

  it("renders submitted overrides while keeping the replay tools canonical", () => {
@@ -58,7 +58,10 @@ export const PresentationAssistantPane = ({
  };
  const { submit, handleKeyDown } = usePreparedComposerSubmit(
    canSubmit,
-    () => onSubmit(draft),
+    () => {
+      onSubmit(draft);
+      setDraft("");
+    },
  );

  return (
@@ -74,7 +77,7 @@ export const PresentationAssistantPane = ({
        <h2>Authoring assistant</h2>
        <p>Current phase: {phaseLabels[phase]}</p>
        <p className="presentation-assistant-pane__disclosure">
-          Prepared replay only. No live actions or RPC calls.
+          Prepared assistant transcript. The final run request uses the configured demo target.
        </p>
      </header>
      <div className="presentation-assistant-pane__conversation">
@@ -83,6 +86,7 @@ export const PresentationAssistantPane = ({
          activePhase={phase}
          surface="stage"
          requestOverrides={submittedOverrides}
+          runRequested={runRequested}
        />
      </div>
      <form className="presentation-assistant-pane__composer" onSubmit={submit}>
@@ -106,7 +110,7 @@ export const PresentationAssistantPane = ({
        </p>
        {runRequested !== null ? (
          <p role="status" className="presentation-assistant-pane__run-status">
-            Run request prepared for the next execution slice.
+            Prepared run receipt added. Continue when you are ready.
          </p>
        ) : null}
      </form>
@@ -4,6 +4,7 @@ import {
  authoringToolGroupId,
  projectPreparedAuthoring,
  projectPreparedAuthoringThread,
+  projectPreparedRunRequestExchange,
  type AuthoringPhaseId,
 } from "./authoring-recording.js";

@@ -225,3 +226,27 @@ describe("projectPreparedAuthoringThread", () => {
    ]);
  });
 });
+
+describe("projectPreparedRunRequestExchange", () => {
+  it("projects the prepared start receipt and typed interrupt without navigation actions", () => {
+    const messages = projectPreparedRunRequestExchange("Run the saved deployment.");
+    const parts = messages.flatMap((message) => message.parts);
+
+    expect(messages[0]).toMatchObject({ role: "user", parts: [{ text: "Run the saved deployment." }] });
+    expect(parts).toContainEqual(expect.objectContaining({
+      type: "tool-call",
+      call: expect.objectContaining({ name: "startRun" }),
+    }));
+    expect(parts).toContainEqual(expect.objectContaining({
+      type: "tool-result",
+      result: expect.objectContaining({
+        output: expect.objectContaining({
+          run_id: "run_recorded_lda_report",
+          status: "interrupted",
+          resume_readiness: "ready",
+        }),
+      }),
+    }));
+    expect(parts.some((part) => part.type === "presentation-action")).toBe(false);
+  });
+});
@@ -354,5 +354,62 @@ export const projectPreparedAuthoringThread = (
  });
 };

+/**
+ * Projects the terminal prepared request into the same chat vocabulary as a
+ * live run, while retaining the stable replay run identity used by later scenes.
+ */
+export const projectPreparedRunRequestExchange = (
+  request: string,
+): readonly AgentMessage[] => {
+  const callId = "authoring-run-request-call";
+  return [
+    agentTextMessage("authoring-run-request-user", "user", request),
+    agentTextMessage(
+      "authoring-run-request-starting",
+      "assistant",
+      "Starting lda_report_case_study.default with the prepared input.",
+    ),
+    {
+      // The authoring prefix reuses the compact phase tool-group treatment;
+      // presenters can expand the real-shaped receipt only when useful.
+      id: "authoring-run-tools",
+      role: "assistant",
+      parts: [
+        agentToolCallPart(callId, "startRun", {
+          operation: "workflow.runs.start",
+          deployment_id: "lda_report_case_study.default",
+          input_file: "examples/lda_report_workflow/run-input.json",
+          equivalent_cli:
+            "uv run wf --url http://127.0.0.1:8765/rpc run start lda_report_case_study.default --input-file examples/lda_report_workflow/run-input.json",
+        }),
+        agentToolResultPart(callId, "startRun", "success", {
+          artifact_id: "lda_report_case_study",
+          artifact_version: 1,
+          deployment_id: "lda_report_case_study.default",
+          diagnostics: [],
+          interrupt: {
+            id: "interrupt:review_issues",
+            kind: "issue_review",
+            node_id: "review_issues",
+            outcomes: ["submitted", "cancelled"],
+            proposed_issue_count: 3,
+            resumable: true,
+            typed: true,
+          },
+          resume_readiness: "ready",
+          run_id: "run_recorded_lda_report",
+          status: "interrupted",
+          trace_count: 6,
+        }),
+      ],
+    },
+    agentTextMessage(
+      "authoring-run-request-paused",
+      "assistant",
+      "The run started and paused at the typed issue-review boundary. Three proposed issues are ready for review.",
+    ),
+  ];
+};
+
 /** Returns the full prepared authoring recording. */
 export const projectPreparedAuthoring = (): readonly PreparedAuthoringPhase[] => recording;
@@ -43,14 +43,14 @@ describe("prepared lifecycle staged message state", () => {
    });
  });

-  it("preserves edited draft text exactly when submitting draft", () => {
+  it("preserves submitted text exactly and clears the composer draft", () => {
    const edited = preparedLifecycleMessageReducer(initialPreparedLifecycleMessageState, {
      type: "draft_edited",
      draft: "  Check only the report binding.  ",
    });

    expect(preparedLifecycleMessageReducer(edited, { type: "draft_submitted" })).toEqual({
-      draft: edited.draft,
+      draft: "",
      submittedOverrides: { validate: edited.draft },
      runRequested: null,
    });
@@ -63,7 +63,7 @@ describe("prepared lifecycle staged message state", () => {
    });

    expect(preparedLifecycleMessageReducer(state, { type: "artifact_submitted" })).toEqual({
-      draft: state.draft,
+      draft: "",
      submittedOverrides: { deployment: state.draft },
      runRequested: null,
    });
@@ -76,7 +76,7 @@ describe("prepared lifecycle staged message state", () => {
    });

    expect(preparedLifecycleMessageReducer(state, { type: "discover_submitted" })).toEqual({
-      draft: state.draft,
+      draft: "",
      submittedOverrides: { discover: state.draft },
      runRequested: null,
    });
@@ -106,6 +106,7 @@ describe("prepared lifecycle staged message state", () => {
    const requested = preparedLifecycleMessageReducer(state, { type: "run_requested" });

    expect(requested.runRequested).toBe("Run this deployment");
+    expect(requested.draft).toBe("");
    expect(requested.submittedOverrides).toEqual({});
    expect(preparedLifecycleMessageReducer(requested, { type: "run_requested" })).toBe(requested);
  });
@@ -57,6 +57,9 @@ const submitOverride = (

  return {
    ...state,
+    // The submitted text is preserved in the transcript override; leaving it
+    // in the composer would leak the previous phase's request into the next.
+    draft: "",
    submittedOverrides: {
      ...state.submittedOverrides,
      [destination]: state.draft,
@@ -79,7 +82,7 @@ export const preparedLifecycleMessageReducer = (
      return submitOverride(state, "deployment");
    case "run_requested":
      return state.runRequested === null && hasText(state.draft)
-        ? { ...state, runRequested: state.draft }
+        ? { ...state, draft: "", runRequested: state.draft }
        : state;
  }
 };
@@ -6,7 +6,7 @@ type PreparedComposerSubmitHandlers = {
 };

 export const PREPARED_COMPOSER_HELP =
-  "Shift+Enter adds a new line. This is a deterministic prepared replay, not a live model request.";
+  "Shift+Enter adds a new line. Responses are prepared; the final run request may use the configured workflow target.";

 /** Shares the prepared-replay composer keyboard and form submission contract. */
 export const usePreparedComposerSubmit = (
@@ -232,7 +232,7 @@ describe("AssistantOperatorThread", () => {
    }
  });

-  it("can show the latest response in a static comparison transcript", async () => {
+  it("can show the latest response even when an authoring group is synchronized", async () => {
    const setScrollTop = vi.fn();
    const descriptors = {
      scrollTop: Object.getOwnPropertyDescriptor(HTMLDivElement.prototype, "scrollTop"),
@@ -246,7 +246,14 @@ describe("AssistantOperatorThread", () => {
        scrollHeight: { configurable: true, get: () => 240 },
        clientHeight: { configurable: true, get: () => 80 },
      });
-      render(<AssistantOperatorThread mode="dock" messages={[]} scrollMode="end" />);
+      render(
+        <AssistantOperatorThread
+          mode="dock"
+          messages={[]}
+          activeToolGroupId="authoring-deployment"
+          scrollMode="end"
+        />,
+      );

      await waitFor(() => expect(setScrollTop).toHaveBeenCalledWith(160));
    } finally {
@@ -276,9 +276,9 @@ export const AssistantOperatorThread = ({
    const viewport = viewportRef.current;
    if (!viewport) return;

-    // Static comparison transcripts should show their final answer; live
-    // presentation docks instead keep the beat-owned tool group in view.
-    if (!activeToolGroupId && scrollMode === "end") {
+    // A terminal exchange must override phase anchoring so its result and
+    // final assistant response remain visible after the composer submits.
+    if (scrollMode === "end") {
      viewport.scrollTop = Math.max(0, viewport.scrollHeight - viewport.clientHeight);
      return;
    }
@@ -25,7 +25,9 @@ describe("PresenterNote", () => {
      />,
    );

-    expect(screen.getByRole("region", { name: "Beat goal" })).toHaveTextContent(note.goal);
+    const goal = screen.getByRole("region", { name: "Beat goal" });
+    expect(goal).toHaveTextContent(note.goal.replaceAll("**", ""));
+    expect(within(goal).getByText("one public operation").tagName).toBe("STRONG");

    const anchors = screen.getByRole("region", { name: "Anchor terms" });
    const anchorList = within(anchors).getByRole("list");
@@ -28,7 +28,7 @@ export const PresenterNote = ({ note, cumulativeSeconds, next, covered, onCovere

      <section className="presenter-note__goal" aria-labelledby="presenter-goal">
        <span id="presenter-goal">Beat goal</span>
-        <p>{note.goal}</p>
+        <div className="presenter-note__goal-copy"><ReactMarkdown>{note.goal}</ReactMarkdown></div>
      </section>

      <section className="presenter-note__anchors" aria-labelledby="presenter-anchors">
@@ -59,7 +59,7 @@ export const presenterNotes = [
    "thesis",
    "title",
    15,
-    "Separate the AI-agent ambition from the implemented contribution.",
+    "Separate the **AI-agent ambition** from the implemented contribution.",
    ["AI-agent goal", "platform underneath"],
    "The title describes the original goal: an AI agent for workspace automation. My contribution is the platform underneath that agent.",
    ["Thesis Abstract and Introduction"],
@@ -72,7 +72,7 @@ export const presenterNotes = [
    "thesis",
    "substrate",
    15,
-    "State what the platform lets its users do.",
+    "State **what the platform lets its users do**.",
    ["agents and humans", "build, run, inspect"],
    "It lets agents and humans build workflows, run them, and inspect what happened.",
    ["Thesis Abstract and Introduction", "Thesis Contributions"],
@@ -82,7 +82,7 @@ export const presenterNotes = [
    "problem",
    "direct-actions",
    15,
-    "Show why one successful chat is not yet automation.",
+    "Show why **one successful chat is not yet automation**.",
    ["tool calls", "not reusable"],
    "Like the chat example, an agent can call tools and finish one task. But that conversation is not yet a reusable workflow.",
    ["Thesis Problem Statement and Requirements"],
@@ -92,7 +92,7 @@ export const presenterNotes = [
    "problem",
    "missing-contracts",
    15,
-    "Name the minimum durable properties reusable automation needs.",
+    "Name the **minimum durable properties** reusable automation needs.",
    ["saved definition", "validation", "execution records"],
    "Reusable automation needs a saved definition, validation, execution records, and a clear way to pause and continue.",
    ["Thesis Problem Statement and Requirements"],
@@ -102,7 +102,7 @@ export const presenterNotes = [
    "positioning",
    "landscape",
    18,
-    "Place the work beside familiar adjacent systems.",
+    "Place the work beside **familiar adjacent systems**.",
    ["Python / n8n / Zapier", "LangGraph", "MCP"],
    "Existing systems solve different parts of this problem: Python scripts, n8n, Zapier, LangGraph, and MCP.",
    ["Thesis Positioning and Related Systems"],
@@ -112,7 +112,7 @@ export const presenterNotes = [
    "positioning",
    "lda-position",
    17,
-    "State the platform's narrow position without a superiority claim.",
+    "State the platform's **narrow position**, without a superiority claim.",
    ["provider-neutral", "workflow layer", "not a replacement"],
    "My platform does not replace them. It provides a provider-neutral workflow layer that agents and humans can operate.",
    ["Thesis Positioning and Related Systems", "Thesis Source Model"],
@@ -122,7 +122,7 @@ export const presenterNotes = [
    "planner-runtime",
    "planner",
    12,
-    "Assign workflow decisions to an external planner.",
+    "Assign workflow decisions to an **external planner**.",
    ["human or AI planner"],
    "A human or AI planner decides what workflow to build.",
    ["Thesis Architecture Overview"],
@@ -132,7 +132,7 @@ export const presenterNotes = [
    "planner-runtime",
    "runtime",
    16,
-    "Assign execution and recording to the runtime.",
+    "Assign **execution and recording** to the runtime.",
    ["validation", "step-by-step execution", "state and traces"],
    "The runtime validates the graph, executes it step by step, records state and traces, and pauses at declared boundaries.",
    ["Thesis Workflow Core", "Thesis Architecture Overview"],
@@ -146,7 +146,7 @@ export const presenterNotes = [
    "planner-runtime",
    "boundary",
    12,
-    "Introduce the public seam between clients and runtime.",
+    "Introduce the **public seam** between clients and runtime.",
    ["Workflow API", "CLI", "JSON-RPC"],
    "Both sides communicate through the Workflow API. Today, clients reach it through the CLI or JSON-RPC without accessing runtime internals directly.",
    ["Thesis Architecture Overview", "docs/source_architecture.md"],
@@ -156,7 +156,7 @@ export const presenterNotes = [
    "lifecycle",
    "draft",
    9,
-    "Introduce the editable lifecycle state.",
+    "Introduce the **editable lifecycle state**.",
    ["Draft", "being built"],
    "A workflow moves through four lifecycle stages. Draft means the workflow is still being built.",
    ["Thesis Workflow Lifecycle"],
@@ -166,7 +166,7 @@ export const presenterNotes = [
    "lifecycle",
    "artifact",
    9,
-    "Introduce the immutable saved definition.",
+    "Introduce the **immutable saved definition**.",
    ["Artifact", "immutable version"],
    "Artifact is a saved, immutable version.",
    ["Thesis Workflow Lifecycle"],
@@ -175,7 +175,7 @@ export const presenterNotes = [
    "lifecycle",
    "deployment",
    9,
-    "Connect a saved definition to a runnable environment.",
+    "Connect a saved definition to a **runnable environment**.",
    ["Deployment", "sources", "ready"],
    "Deployment connects that version to the sources it needs and checks whether it is ready.",
    ["Thesis Workflow Lifecycle"],
@@ -184,7 +184,7 @@ export const presenterNotes = [
    "lifecycle",
    "run",
    9,
-    "Introduce one persisted execution record.",
+    "Introduce one **persisted execution record**.",
    ["Run", "status", "output and trace"],
    "Run is one recorded execution, including its status, output, and trace.",
    ["Thesis Workflow Lifecycle"],
@@ -194,27 +194,27 @@ export const presenterNotes = [
    "architecture",
    "overview",
    6,
-    "Show how the implementation realizes the earlier concepts.",
+    "Show how the **implementation realizes the earlier concepts**.",
    ["architecture spine"],
-    "This is how those concepts are organized in the implementation.",
+    "Now I will show how those concepts map to the implementation.",
    ["Thesis System Architecture", "docs/project_map.md"],
  ),
  beatNote(
    "architecture",
    "client",
    8,
-    "Show that humans and agents share one public surface.",
+    "Show that humans and agents share **one public surface**.",
    ["shared operations"],
-    "Humans and agents use the same public workflow operations.",
+    "At the top, humans and agents use the same public workflow operations.",
    ["Thesis System Architecture", "docs/project_map.md"],
  ),
  beatNote(
    "architecture",
    "api",
    9,
-    "Identify the system's public front door.",
+    "Identify the system's **public front door**.",
    ["Workflow API", "public boundary"],
-    "The Workflow API is the front door. It exposes lifecycle operations without exposing runtime internals.",
+    "Those operations enter through the Workflow API, the public boundary that keeps clients out of runtime internals.",
    ["Thesis System Architecture", "docs/source_architecture.md"],
    { qnaBranchIds: ["not-just-cli"] },
  ),
@@ -222,9 +222,9 @@ export const presenterNotes = [
    "architecture",
    "runtime",
    9,
-    "Explain what the server composes behind the API.",
+    "Explain what the server composes **behind the API**.",
    ["WorkflowServer", "records and capabilities", "execution core"],
-    "Behind it, the workflow server brings together stored records, available capabilities, and the execution core.",
+    "Behind that boundary, the workflow server brings together stored records, available capabilities, and the execution core.",
    ["Thesis System Architecture", "docs/source_architecture.md"],
    { qnaBranchIds: ["provider-security"] },
  ),
@@ -232,7 +232,7 @@ export const presenterNotes = [
    "agent-handoff",
    "request",
    12,
-    "Disclose the prepared demonstration before it begins.",
+    "Disclose the **prepared demonstration** before it begins.",
    ["prepared example", "not an autonomous planner"],
    "This is a prepared example, not a live autonomous AI agent. It shows how an agent could use the platform to build and run a workflow.",
    ["Constrained demo agent and prepared replay recipe"],
@@ -245,7 +245,7 @@ export const presenterNotes = [
    "prepared-lifecycle",
    "discover",
    7,
-    "Show that authoring starts with interface discovery.",
+    "Show that authoring starts with **interface discovery**.",
    ["sources", "capabilities"],
    "First, the agent checks which sources and operations are available.",
    ["examples/lda_report_workflow", "deployment inspect replay evidence"],
@@ -255,7 +255,7 @@ export const presenterNotes = [
    "prepared-lifecycle",
    "draft",
    7,
-    "Show mutable workflow authoring.",
+    "Show **mutable workflow authoring**.",
    ["Draft", "editable workflow"],
    "Then it builds an editable workflow draft.",
    ["examples/lda_report_workflow", "deployment inspect replay evidence", "CLI documentation", "Draft authoring API"],
@@ -265,7 +265,7 @@ export const presenterNotes = [
    "prepared-lifecycle",
    "diagnose",
    7,
-    "Show a concrete structured validation failure.",
+    "Show a **concrete structured validation failure**.",
    ["validation", "missing_outcome_edge"],
    "Validation finds that the analyze step has no route for its ok outcome.",
    ["examples/lda_report_workflow", "deployment inspect replay evidence"],
@@ -275,7 +275,7 @@ export const presenterNotes = [
    "prepared-lifecycle",
    "repair",
    7,
-    "Show the exact focused correction and revalidation.",
+    "Show the **focused correction and revalidation**.",
    ["set-route", "validation passes"],
    "The agent adds that route, and validation passes.",
    ["Validation and diagnostics", "Challenge UX findings"],
@@ -285,7 +285,7 @@ export const presenterNotes = [
    "prepared-lifecycle",
    "artifact",
    7,
-    "Show the transition to an immutable saved version.",
+    "Show the transition to an **immutable saved version**.",
    ["Artifact", "immutable"],
    "The valid workflow is saved as an immutable artifact.",
    ["examples/lda_report_workflow", "deployment inspect replay evidence"],
@@ -294,7 +294,7 @@ export const presenterNotes = [
    "prepared-lifecycle",
    "deployment",
    7,
-    "Show source binding and readiness before execution.",
+    "Show **source binding and readiness** before execution.",
    ["Deployment", "three local sources"],
    "Finally, a deployment connects it to the three local sources it needs.",
    ["examples/lda_report_workflow", "deployment inspect replay evidence", "Thesis deterministic report case study"],
@@ -304,7 +304,7 @@ export const presenterNotes = [
    "run-from-deployment",
    "input",
    11,
-    "Show the concrete inputs supplied before execution.",
+    "Show the **concrete inputs** supplied before execution.",
    ["run input", "selected documents"],
    "The deployment receives selected local documents and an issue-board path.",
    ["workflow.runs.start replay evidence"],
@@ -313,7 +313,7 @@ export const presenterNotes = [
    "run-from-deployment",
    "operation",
    12,
-    "Show that one public operation creates a persisted execution.",
+    "Show that **one public operation** creates a persisted execution.",
    ["workflow.runs.start", "persisted Run"],
    "The public **workflow.runs.start** operation validates the deployment and input, creates a **persisted Run**, and begins the reusable graph.",
    ["workflow.runs.start replay evidence"],
@@ -323,7 +323,7 @@ export const presenterNotes = [
    "run-from-deployment",
    "graph",
    12,
-    "Show the reusable workflow executing beyond the chat conversation.",
+    "Show the **reusable workflow** executing beyond the chat conversation.",
    ["workflow graph", "declared interrupt"],
    "The graph reads documents, analyzes them, builds a report, drafts proposed issues, and pauses at a declared review interrupt before issue-board changes.",
    ["workflow.runs.start replay evidence", "examples/lda_report_workflow"],
@@ -333,7 +333,7 @@ export const presenterNotes = [
    "typed-human-boundary",
    "interrupt",
    15,
-    "Show what the paused workflow asks from the operator.",
+    "Show what the **paused workflow asks** from the operator.",
    ["issue_review", "interrupt payload", "resume schema"],
    "Execution pauses at a **typed issue_review interrupt** exposing request data, allowed outcomes, request schema, and resume schema.",
    ["Typed interrupt payload and resume contract"],
@@ -343,9 +343,9 @@ export const presenterNotes = [
    "typed-human-boundary",
    "approval",
    15,
-    "Show that the operator chooses a declared continuation.",
+    "Show that the operator chooses a **declared continuation**.",
    ["submitted", "revision-requested", "typed resume"],
-    "The operator chooses submitted or revision-requested; this is a typed interrupt and resume contract, not a production approval gate, role system, or policy engine.",
+    "The operator can Submit or Request revision. Both continue through declared workflow branches; this typed resume contract is not a production approval system.",
    ["Typed interrupt payload and resume contract"],
    { warning: "Both outcomes resume through declared workflow branches; this is not production approval governance.", qnaBranchIds: ["typed-interrupts", "security-production-boundary"] },
  ),
@@ -353,7 +353,7 @@ export const presenterNotes = [
    "resume-output-evidence",
    "resume",
    16,
-    "Show continuation of the same recorded run.",
+    "Show continuation of the **same recorded run**.",
    ["workflow.runs.resume", "same Run"],
    "On the submitted path, **workflow.runs.resume continues the recorded interrupted Run**.",
    ["workflow.runs.resume replay evidence", "Revision replay identity"],
@@ -363,7 +363,7 @@ export const presenterNotes = [
    "resume-output-evidence",
    "output",
    16,
-    "Show the persisted terminal results of the submitted path.",
+    "Show the **persisted terminal results** of the submitted path.",
    ["report output", "issue-board changes"],
    "The workflow creates the report and issue-board changes, then records terminal output.",
    ["workflow.runs.resume replay evidence", "examples/lda_report_workflow"],
@@ -373,19 +373,19 @@ export const presenterNotes = [
    "resume-output-evidence",
    "trace",
    18,
-    "Show that execution evidence remains inspectable after completion.",
+    "Show that **execution evidence remains inspectable** after completion.",
    ["trace frames", "protocol evidence"],
-    "Trace frames and protocol evidence remain inspectable; this is declared-boundary resumability, not arbitrary crash recovery or exactly-once execution. The revision replay is a separate prepared recording.",
-    ["workflow.runs.resume replay evidence", "Revision replay identity"],
-    { warning: "Never claim run-ID continuity for the prepared revision recording.", qnaBranchIds: ["replay-provenance", "demo-reliability"] },
+    "Trace frames and protocol evidence remain inspectable; this is declared-boundary resumability, not arbitrary crash recovery or exactly-once execution.",
+    ["workflow.runs.resume replay evidence"],
+    { qnaBranchIds: ["replay-provenance", "demo-reliability"] },
  ),
  beatNote(
    "evaluation",
    "cohort",
    40,
-    "Describe the external-agent evaluation design.",
+    "Describe the **external-agent evaluation design**.",
    ["36 trials", "two challenges", "three profiles"],
-    "The evaluation combines conformance tests, deterministic case studies, and a **manually audited external-agent campaign**: 36 trials across two challenges, two hosted models, three instruction profiles, and three waves, with three attempts per cell.",
+    "The evaluation combines conformance tests, deterministic case studies, and a **manually audited external-agent campaign**. Each model, under each instruction profile, attempted each challenge three times, for 36 trials in total.",
    ["Thesis Evaluation and Appendix C"],
    { qnaBranchIds: ["evaluation-validity"] },
  ),
@@ -393,9 +393,9 @@ export const presenterNotes = [
    "evaluation",
    "validity",
    40,
-    "Separate audited valid evidence from contaminated samples.",
+    "Separate **audited valid evidence** from contaminated samples.",
    ["27 pass", "8 invalid", "1 fail"],
-    "The author audit classified 27 trials as clean product-path passes, eight as invalid samples, and one as a failure. Invalid samples included contamination such as reading implementation files, prior artifacts, adjacent attempts, or evaluator state.",
+    "The author audit then classified 27 trials as clean product-path passes, eight as invalid samples, and one as a failure. Invalid samples included contamination such as reading implementation files, prior artifacts, adjacent attempts, or evaluator state.",
    ["Thesis Evaluation and Appendix C", "Author audit"],
    { qnaBranchIds: ["evaluation-validity"] },
  ),
@@ -403,9 +403,9 @@ export const presenterNotes = [
    "evaluation",
    "findings",
    40,
-    "State what the evaluation supports and what it cannot prove.",
-    ["longitudinal evidence", "not a benchmark"],
-    "Because prompts, product snapshots, and hosted conditions changed across waves, these results are **longitudinal engineering evidence**. They expose authoring and diagnostic gaps, **not a benchmark** of model success, token reduction, retry reduction, or superiority.",
+    "State **what the evaluation supports** and what it cannot prove.",
+    ["product evolved", "not a benchmark"],
+    "But the prompts, product, and hosted conditions changed during the campaign. These results show what we learned while improving the product. They expose authoring and diagnostic gaps; they are **not a benchmark** of model success, token reduction, retry reduction, or superiority.",
    ["Thesis Evaluation and Appendix C", "Thesis Threats to Validity"],
    { warning: "Use non-benchmark wording; do not report the counts as general model performance.", qnaBranchIds: ["evaluation-validity"] },
  ),
@@ -413,7 +413,7 @@ export const presenterNotes = [
    "conclusion",
    "limits",
    18,
-    "Bound the prototype claims before the final contribution statement.",
+    "Bound the **prototype claims** before the final contribution statement.",
    ["prototype", "not production security"],
    "The prototype uses trusted in-process Python and file-backed stores; it does not provide production authentication, RBAC, sandboxing, scheduling, arbitrary crash recovery, or a bundled autonomous planner.",
    ["Thesis Limitations"],
@@ -423,7 +423,7 @@ export const presenterNotes = [
    "conclusion",
    "future",
    18,
-    "Name the surrounding layers left as future work.",
+    "Name the surrounding layers left as **future work**.",
    ["live agent", "scheduling", "controlled evaluation"],
    "A live agent interface, transactional storage, richer debugging, security hardening, scheduling, and controlled comparative evaluation remain future work.",
    ["Thesis Future Work"],
@@ -433,7 +433,7 @@ export const presenterNotes = [
    "conclusion",
    "conclusion",
    20,
-    "Restate the implemented contribution and planner-runtime boundary.",
+    "Restate the **implemented contribution** and planner-runtime boundary.",
    ["planner proposes", "platform executes"],
    "The contribution is **architectural and implemented**: external planners can propose workflows while a typed platform **validates, binds, executes, persists, interrupts, resumes, and inspects** them through public operations.",
    ["Thesis Contributions", "Thesis Conclusion"],
@@ -443,7 +443,7 @@ export const presenterNotes = [
    "conclusion",
    "questions",
    19,
-    "Open structured examiner discussion without introducing new claims.",
+    "Open **structured examiner discussion** without introducing new claims.",
    ["defense questions", "evidence"],
    "That boundary is the claim I will defend: reusable agent-operated automation is inspectable because planning and execution have explicit contracts. I welcome questions.",
    ["Thesis Conclusion", "Defense Q&A index"],
@@ -114,7 +114,11 @@ describe("defense storyboard catalog", () => {

  it("makes the evaluation beat carry the audited counts and validity boundary", () => {
    expect(findBeat("evaluation", "cohort")?.caption).toMatch(/36|two challenges|two hosted models|three waves/i);
-    expect(findBeat("evaluation", "validity")?.caption).toMatch(/27|8|1|audit/i);
+    const validityCaption = findBeat("evaluation", "validity")?.caption ?? "";
+    expect(validityCaption).toMatch(/\b27\b/);
+    expect(validityCaption).toMatch(/\b8\b/);
+    expect(validityCaption).toMatch(/\b1\b/);
+    expect(validityCaption).toMatch(/audit/i);
    expect(findBeat("evaluation", "findings")?.caption).toMatch(/changing|longitudinal|benchmark/i);
  });

@@ -37,12 +37,23 @@ describe("presentationWorkflowPlan", () => {
      expect(node.capability, node.id).toBeTruthy();
      expect(node.inputSummary, node.id).toBeTruthy();
      expect(node.outputSummary, node.id).toBeTruthy();
-      expect(node.outcomes, node.id).not.toHaveLength(0);
      expect(node.evidencePointer, node.id).toBeTruthy();
+
+      if (node.type !== "end") {
+        expect(node.outcomes, node.id).toEqual(
+          presentationWorkflowPlan.edges
+            .filter((edge) => edge.from === node.id)
+            .map((edge) => edge.outcome),
+        );
+      }
    }

    const interrupt = presentationWorkflowPlan.nodes.find((node) => node.id === "review_issues");
    expect(interrupt?.schemaSummary).toMatch(/request: issue proposals/i);
    expect(interrupt?.outcomes).toEqual(["submitted", "cancelled"]);
+    expect(presentationWorkflowPlan.nodes.find((node) => node.id === "finalise")?.outcomes).toEqual([
+      "completed",
+    ]);
+    expect(presentationWorkflowPlan.nodes.find((node) => node.id === "revision_requested")?.outcomes).toEqual([]);
  });
 });
@@ -99,7 +99,7 @@ export const presentationWorkflowPlan = {
      capability: "State output",
      inputSummary: "Markdown report and issue-board state",
      outputSummary: "Completed run output",
-      outcomes: ["ok"],
+      outcomes: ["completed"],
      evidencePointer: "local.lda_report.finalise_report",
    },
    {
@@ -111,7 +111,7 @@ export const presentationWorkflowPlan = {
      capability: "Revision record",
      inputSummary: "Issue-review decision",
      outputSummary: "Recorded revision request",
-      outcomes: ["cancelled"],
+      outcomes: [],
      evidencePointer: "local.lda_report.record_revision_request",
    },
    {