@signalridge/pi-workflows 1.7.5 → 1.9.0

package/src/inspection.ts

@@ -0,0 +1,544 @@
+/** Read-only, bounded presentation facts. No prompts, args, histories, or full results. */
+import { createHash, randomUUID } from "node:crypto";
+import type { ScriptRun } from "./journal.js";
+import {
+  type AgentRunOptions,
+  MAX_RUNTIME_IDENTITY_LENGTH,
+  type WorkflowRunOptions,
+  type WorkflowRuntimeEvent,
+} from "./runtime.js";
+
+export const INSPECTION_LIMITS = {
+  calls: 2_048,
+  phases: 256,
+  tasks: 256,
+  events: 128,
+  text: 512,
+  preview: 300,
+} as const;
+export type InspectionStatus =
+  | "unknown"
+  | "pending"
+  | "queued"
+  | "running"
+  | "completed"
+  | "failed"
+  | "stopped"
+  | "interrupted";
+export interface WorkflowInspectionPhase {
+  readonly id: string;
+  readonly frameId: string;
+  /** Opaque frame-scoped hash of the full, untruncated title. */
+  readonly phaseKey?: string;
+  readonly title: string;
+  readonly detail?: string;
+  readonly declared: boolean;
+  /** Only an explicit phase event proves entry; attribution and metadata do not. */
+  readonly entered: boolean;
+}
+export interface WorkflowInspectionCall {
+  readonly id: string;
+  readonly frameId: string;
+  readonly index: number;
+  readonly kind: "agent" | "workflow" | "checkpoint" | "unknown";
+  readonly label?: string;
+  readonly phase?: string;
+  /** Optional for legacy adapters; never infer this from a display label. */
+  readonly phaseKey?: string;
+  readonly agentType?: string;
+  readonly tier?: string;
+  readonly graphId?: string;
+  readonly taskKey?: string;
+  readonly taskId?: string;
+  readonly status: InspectionStatus;
+  /** Physical terminal status can differ from the script's logical outcome. */
+  readonly physicalStatus?: InspectionStatus;
+  readonly physicalDispatches?: number;
+  readonly replayed?: boolean;
+  readonly replayedAgentCount?: number;
+  readonly childFrameId?: string;
+  readonly observed: boolean;
+  readonly journalOnly: boolean;
+  readonly agentId?: string;
+  readonly attemptId?: string;
+  readonly outputFile?: string;
+  readonly tokenCount?: number;
+  readonly compactionCount?: number;
+  readonly resultPreview?: string;
+  readonly errorPreview?: string;
+}
+export interface WorkflowInspectionTask {
+  readonly id: string;
+  readonly frameId: string;
+  readonly graphId: string;
+  readonly taskId: string;
+  readonly description?: string;
+  readonly phase?: string;
+  readonly attempt: number;
+  readonly errorPreview?: string;
+  readonly status: InspectionStatus | "skipped";
+}
+export interface WorkflowInspectionEvent {
+  readonly id: string;
+  readonly type: WorkflowRuntimeEvent["type"] | "log";
+  readonly frameId: string;
+  readonly stage?: string;
+  readonly childFrameId?: string;
+  readonly title?: string;
+  readonly graphId?: string;
+  readonly taskKey?: string;
+  readonly helperId?: string;
+  readonly helper?: string;
+  readonly outcome?: string;
+  readonly accepted?: boolean;
+  readonly attempt?: number;
+  readonly message?: string;
+}
+export interface WorkflowInspection {
+  readonly runId: string;
+  /** Opaque engine-state/execution identity; never use runId alone to own a UI mount. */
+  readonly executionId: string;
+  readonly executionGeneration: number;
+  readonly revision: number;
+  readonly name: string;
+  readonly status: ScriptRun["status"];
+  readonly observationSource: "execution" | "journal-only";
+  readonly restored: boolean;
+  readonly currentPhase?: string;
+  readonly currentPhaseKey?: string;
+  readonly phases: readonly WorkflowInspectionPhase[];
+  readonly calls: readonly WorkflowInspectionCall[];
+  readonly tasks: readonly WorkflowInspectionTask[];
+  readonly events: readonly WorkflowInspectionEvent[];
+  /** Counters cover observed callbacks, not anonymous historical journal records. */
+  readonly logicalCalls: number;
+  readonly replayedCalls: number;
+  readonly physicalDispatches: number;
+  readonly truncated: boolean;
+  readonly resultPreview?: string;
+  readonly errorPreview?: string;
+}
+
+export type WorkflowInspectionHeader = Pick<
+  WorkflowInspection,
+  "currentPhase" | "currentPhaseKey" | "restored" | "revision" | "executionId" | "executionGeneration"
+>;
+
+type Mutable<T> = { -readonly [K in keyof T]: T[K] };
+type StartEvent = Parameters<NonNullable<WorkflowRunOptions["onAgentStart"]>>[0];
+type EndEvent = Parameters<NonNullable<WorkflowRunOptions["onAgentEnd"]>>[0];
+const text = (value: unknown, limit: number = INSPECTION_LIMITS.text): string | undefined =>
+  typeof value === "string" ? value.slice(0, limit) : undefined;
+function identity(value: unknown): string | undefined {
+  if (value === undefined) return undefined;
+  if (typeof value !== "string" || value.length === 0 || value.length > MAX_RUNTIME_IDENTITY_LENGTH)
+    throw new Error(
+      `Inspection identity must be a non-empty string of at most ${MAX_RUNTIME_IDENTITY_LENGTH} characters`,
+    );
+  return value;
+}
+function preview(value: unknown): string | undefined {
+  if (value === undefined) return undefined;
+  if (typeof value === "string") return value.slice(0, INSPECTION_LIMITS.preview);
+  try {
+    return JSON.stringify(value)?.slice(0, INSPECTION_LIMITS.preview);
+  } catch {
+    return "[unavailable]";
+  }
+}
+const phaseKey = (frameId: string, title: string): string =>
+  createHash("sha256")
+    .update(JSON.stringify([frameId, title]))
+    .digest("hex");
+function status(value: unknown): InspectionStatus {
+  return value === "pending" ||
+    value === "queued" ||
+    value === "running" ||
+    value === "completed" ||
+    value === "failed" ||
+    value === "stopped" ||
+    value === "interrupted"
+    ? value
+    : "unknown";
+}
+
+/** Engine-owned projection; public consumers receive detached snapshots only. */
+export class WorkflowPresentation {
+  private executionId = randomUUID();
+  private revision = 0;
+  private observed = false;
+  private truncated = false;
+  private currentPhase?: string;
+  private currentPhaseKey?: string;
+  private readonly calls = new Map<string, Mutable<WorkflowInspectionCall>>();
+  private readonly phases = new Map<string, Mutable<WorkflowInspectionPhase>>();
+  private readonly tasks = new Map<string, Mutable<WorkflowInspectionTask>>();
+  private readonly events: WorkflowInspectionEvent[] = [];
+  private logicalCalls = 0;
+  private replayedCalls = 0;
+  private physicalDispatches = 0;
+  constructor(
+    private readonly rootFrameId: string,
+    private readonly restored = false,
+  ) {
+    identity(rootFrameId);
+  }
+
+  reset(observed = true): void {
+    this.executionId = randomUUID();
+    this.observed = observed;
+    this.currentPhase = undefined;
+    this.currentPhaseKey = undefined;
+    this.calls.clear();
+    this.phases.clear();
+    this.tasks.clear();
+    this.events.length = 0;
+    this.logicalCalls = this.replayedCalls = this.physicalDispatches = 0;
+    this.truncated = false;
+    this.revision++;
+  }
+
+  changed(): void {
+    this.revision++;
+  }
+
+  private call(id: string, frameId: string, index: number): Mutable<WorkflowInspectionCall> {
+    identity(id);
+    identity(frameId);
+    const existing = this.calls.get(id);
+    if (existing) return existing;
+    this.logicalCalls++;
+    if (this.calls.size >= INSPECTION_LIMITS.calls) {
+      this.calls.delete(this.calls.keys().next().value ?? "");
+      this.truncated = true;
+    }
+    const call: Mutable<WorkflowInspectionCall> = {
+      id,
+      frameId,
+      index,
+      kind: "unknown",
+      status: "unknown",
+      observed: true,
+      journalOnly: false,
+    };
+    this.calls.set(id, call);
+    return call;
+  }
+
+  agentStart(event: StartEvent): void {
+    for (const key of [event.id, event.frameId, event.graphId, event.taskKey, event.taskId, event.phaseKey])
+      identity(key);
+    const frameId = event.frameId ?? event.id.slice(0, event.id.lastIndexOf(":"));
+    const index = event.callIndex ?? Number(event.id.slice(event.id.lastIndexOf(":") + 1));
+    const call = this.call(event.id, frameId, index);
+    Object.assign(call, {
+      kind: "agent",
+      label: text(event.label),
+      phase: text(event.phase),
+      phaseKey: identity(event.phaseKey),
+      agentType: text(event.agentType),
+      tier: text(event.tier),
+      graphId: identity(event.graphId),
+      taskKey: identity(event.taskKey),
+      taskId: identity(event.taskId),
+      status: "running",
+      replayed: event.replayed === true,
+      physicalDispatches: call.physicalDispatches ?? 0,
+    });
+    if (event.replayed) this.replayedCalls++;
+    this.revision++;
+  }
+
+  agentEnd(event: EndEvent): void {
+    identity(event.id);
+    const call = this.calls.get(event.id);
+    if (!call) return;
+    call.status = event.error ? "failed" : "completed";
+    call.tier = text(event.tier) ?? call.tier;
+    call.resultPreview = preview(event.result);
+    call.errorPreview = text(event.error, INSPECTION_LIMITS.preview);
+    call.tokenCount = event.tokens;
+    this.revision++;
+  }
+
+  dispatch(options: AgentRunOptions, attemptId: string): void {
+    identity(attemptId);
+    const frameId = options.runId ?? this.rootFrameId;
+    const index = options.callIndex ?? 0;
+    const call = this.call(`${frameId}:${index}`, frameId, index);
+    call.physicalDispatches = (call.physicalDispatches ?? 0) + 1;
+    call.physicalStatus = "pending";
+    call.attemptId = identity(attemptId);
+    // An earlier retry's terminal identity/preview is not this new attempt.
+    call.agentId = call.resultPreview = call.errorPreview = call.outputFile = undefined;
+    call.tokenCount = call.compactionCount = undefined;
+    this.physicalDispatches++;
+    this.revision++;
+  }
+
+  managed(
+    nodeId: string,
+    fact: {
+      status?: string;
+      agentId?: string;
+      tier?: string;
+      outputFile?: string;
+      tokenCount?: number;
+      compactionCount?: number;
+    },
+  ): void {
+    identity(nodeId);
+    identity(fact.agentId);
+    const id = nodeId.includes(":") ? nodeId : `${this.rootFrameId}:${nodeId}`;
+    const call = this.calls.get(id);
+    if (!call) return;
+    if (fact.status !== undefined) call.physicalStatus = status(fact.status);
+    if (fact.agentId !== undefined) call.agentId = identity(fact.agentId);
+    if (fact.tier !== undefined) call.tier = text(fact.tier);
+    if (fact.outputFile !== undefined) call.outputFile = text(fact.outputFile);
+    if (fact.tokenCount !== undefined) call.tokenCount = fact.tokenCount;
+    if (fact.compactionCount !== undefined) call.compactionCount = fact.compactionCount;
+    this.revision++;
+  }
+
+  journalResult(frameId: string, index: number, result: unknown): void {
+    identity(frameId);
+    const call = this.calls.get(`${frameId}:${index}`);
+    if (!call) return;
+    call.resultPreview = preview(result);
+    this.revision++;
+  }
+
+  log(message: string, frameId = this.rootFrameId): void {
+    identity(frameId);
+    this.record({ type: "log", frameId, message: text(message) });
+  }
+
+  runtime(event: WorkflowRuntimeEvent): void {
+    const rawFrameId = typeof event.frameId === "string" ? event.frameId : this.rootFrameId;
+    const frameId = identity(rawFrameId) ?? this.rootFrameId;
+    // Validate before retaining state: display limits must never alias keys.
+    for (const key of ["runId", "childFrameId", "graphId", "taskKey", "taskId", "helperId", "phaseKey"])
+      identity(event[key]);
+    if (event.type === "phase" && typeof event.title === "string") {
+      const title = text(event.title) ?? "";
+      // Hash exact titles so bounded display labels cannot alias identities or retain unbounded input.
+      const key = phaseKey(rawFrameId, event.title);
+      if (rawFrameId === this.rootFrameId) {
+        this.currentPhase = title;
+        this.currentPhaseKey = key;
+      }
+      if (!this.phases.has(key)) {
+        if (this.phases.size >= INSPECTION_LIMITS.phases) {
+          this.phases.delete(this.phases.keys().next().value ?? "");
+          this.truncated = true;
+        }
+        this.phases.set(key, {
+          id: `${frameId}:phase-${this.revision}`,
+          frameId,
+          phaseKey: key,
+          title,
+          declared: false,
+          entered: true,
+        });
+      }
+    }
+    if (event.type === "task" && typeof event.taskKey === "string" && typeof event.taskId === "string") {
+      const id = identity(event.taskKey) ?? "";
+      if (!this.tasks.has(id) && this.tasks.size >= INSPECTION_LIMITS.tasks) {
+        this.tasks.delete(this.tasks.keys().next().value ?? "");
+        this.truncated = true;
+      }
+      this.tasks.set(id, {
+        id,
+        frameId,
+        graphId: identity(event.graphId) ?? "",
+        taskId: identity(event.taskId) ?? "",
+        description: text(event.description),
+        phase: text(event.phase),
+        attempt: typeof event.attempt === "number" ? event.attempt : 0,
+        errorPreview: text(event.error, INSPECTION_LIMITS.preview),
+        status: event.stage === "skip" ? "skipped" : event.stage === "end" ? status(event.status) : "running",
+      });
+    }
+    if (event.type === "control-attempt" && event.helper === "checkpoint" && typeof event.callIndex === "number") {
+      const call = this.call(`${frameId}:${event.callIndex}`, frameId, event.callIndex);
+      call.kind = "checkpoint";
+      call.label = "checkpoint";
+      call.phase = text(event.phase);
+      call.phaseKey = identity(event.phaseKey);
+      call.graphId = identity(event.graphId);
+      call.taskKey = identity(event.taskKey);
+      call.taskId = identity(event.taskId);
+      call.physicalDispatches = 0;
+      call.replayed = event.stage === "replay";
+      if (call.replayed) this.replayedCalls++;
+      call.status =
+        event.stage === "start"
+          ? "running"
+          : event.outcome === "cancelled"
+            ? "interrupted"
+            : event.outcome === "failed"
+              ? "failed"
+              : "completed";
+    }
+    if (event.type === "workflow" && typeof event.callIndex === "number") {
+      const call = this.call(`${frameId}:${event.callIndex}`, frameId, event.callIndex);
+      call.kind = "workflow";
+      call.childFrameId = identity(event.childFrameId);
+      call.phase = text(event.phase);
+      call.phaseKey = identity(event.phaseKey);
+      call.graphId = identity(event.graphId);
+      call.taskKey = identity(event.taskKey);
+      call.taskId = identity(event.taskId);
+      // Only the resolver can prove a saved name. Unknown/legacy sources fail closed.
+      call.label = event.sourceKind === "saved" ? (text(event.name) ?? "nested workflow") : "nested workflow";
+      call.replayed = event.stage === "replay";
+      call.replayedAgentCount = typeof event.replayedAgentCount === "number" ? event.replayedAgentCount : undefined;
+      if (event.stage === "replay") this.replayedCalls++;
+      call.status =
+        event.stage === "start"
+          ? "running"
+          : event.outcome === "cancelled"
+            ? "interrupted"
+            : event.outcome === "failed"
+              ? "failed"
+              : "completed";
+    }
+    this.record({
+      type: event.type,
+      frameId,
+      stage: text(event.stage),
+      childFrameId: identity(event.childFrameId),
+      title: text(event.title),
+      graphId: identity(event.graphId),
+      taskKey: identity(event.taskKey),
+      helperId: identity(event.helperId),
+      helper: text(event.helper),
+      outcome: text(event.outcome),
+      accepted: typeof event.accepted === "boolean" ? event.accepted : undefined,
+      attempt: typeof event.attempt === "number" ? event.attempt : undefined,
+    });
+  }
+
+  private record(event: Omit<WorkflowInspectionEvent, "id">): void {
+    this.events.push({ ...event, id: `${this.executionId}:event-${++this.revision}` });
+    if (this.events.length > INSPECTION_LIMITS.events) {
+      this.events.shift();
+      this.truncated = true;
+    }
+  }
+
+  settle(runStatus: ScriptRun["status"]): void {
+    // Stop owns its terminal barrier; draining execution alone is not settlement.
+    if (runStatus === "stopping") return;
+    for (const call of this.calls.values()) {
+      if (call.status === "running" || call.status === "pending" || call.status === "queued") {
+        // A final workflow outcome does not prove unpaired calls succeeded.
+        call.status = runStatus === "stopped" ? "stopped" : "interrupted";
+      }
+    }
+    this.revision++;
+  }
+
+  header(executionGeneration: number): WorkflowInspectionHeader {
+    return {
+      currentPhase: this.currentPhase,
+      currentPhaseKey: this.currentPhaseKey,
+      restored: this.restored && !this.observed,
+      revision: this.revision,
+      executionId: this.executionId,
+      executionGeneration,
+    };
+  }
+
+  snapshot(run: ScriptRun, executionGeneration: number): WorkflowInspection {
+    identity(run.runId);
+    const phases: WorkflowInspectionPhase[] = [];
+    const declaredKeys = new Set<string>();
+    for (const [index, phase] of (run.meta.phases ?? []).slice(0, INSPECTION_LIMITS.phases).entries()) {
+      const key = phaseKey(this.rootFrameId, phase.title);
+      // Repeated literal declarations do not invent additional progress units.
+      if (declaredKeys.has(key)) continue;
+      declaredKeys.add(key);
+      phases.push({
+        id: `${this.rootFrameId}:declared-${index}`,
+        frameId: this.rootFrameId,
+        phaseKey: key,
+        title: text(phase.title) ?? "",
+        detail: text(phase.detail),
+        declared: true,
+        entered: this.phases.has(key),
+      });
+    }
+    for (const [key, phase] of this.phases) {
+      if (!declaredKeys.has(key)) phases.push({ ...phase });
+    }
+    const calls = new Map<string, WorkflowInspectionCall>();
+    for (const [nodeId, durableStatus] of Object.entries(run.callStatus).slice(0, INSPECTION_LIMITS.calls)) {
+      const result = run.callResults[nodeId];
+      calls.set(`${this.rootFrameId}:${nodeId}`, {
+        id: `${this.rootFrameId}:${nodeId}`,
+        frameId: this.rootFrameId,
+        index: Number(nodeId),
+        kind: "unknown",
+        status: status(durableStatus),
+        observed: false,
+        journalOnly: true,
+        tier: text(result?.tier ?? run.callTiers[nodeId]?.tier),
+        // A terminal result without managed identity may be local work. Never attach a prior child's id.
+        agentId: identity(result === undefined ? run.agentIds[nodeId] : result.agentId),
+        attemptId: identity(run.attemptIds[nodeId]),
+        outputFile: text(result?.outputFile),
+        tokenCount: result?.tokenCount,
+        compactionCount: result?.compactionCount ?? run.compactions[nodeId],
+        resultPreview: preview(result?.result),
+        errorPreview: text(result?.error, INSPECTION_LIMITS.preview),
+      });
+    }
+    for (const [id, observed] of this.calls) {
+      const durable = calls.get(id);
+      // Replayed calls may reference their retained historical child/results, never an invented live dispatch.
+      const replayFacts =
+        observed.replayed && durable?.status === "completed"
+          ? {
+              agentId: durable.agentId,
+              attemptId: durable.attemptId,
+              outputFile: durable.outputFile,
+              compactionCount: durable.compactionCount,
+              resultPreview: durable.resultPreview,
+            }
+          : {};
+      const merged = { ...replayFacts, ...observed };
+      if (observed.replayed && observed.resultPreview === undefined) merged.resultPreview = replayFacts.resultPreview;
+      calls.set(id, merged);
+    }
+    const boundedCalls = [...calls.values()].slice(-INSPECTION_LIMITS.calls);
+    return structuredClone({
+      runId: run.runId,
+      executionId: this.executionId,
+      executionGeneration,
+      revision: this.revision,
+      name: text(run.meta.name) ?? "",
+      status: run.status,
+      observationSource: this.observed ? "execution" : "journal-only",
+      restored: this.restored && !this.observed,
+      currentPhase: this.currentPhase,
+      currentPhaseKey: this.currentPhaseKey,
+      phases: phases.slice(0, INSPECTION_LIMITS.phases),
+      calls: boundedCalls,
+      tasks: [...this.tasks.values()],
+      events: this.events,
+      logicalCalls: this.logicalCalls,
+      replayedCalls: this.replayedCalls,
+      physicalDispatches: this.physicalDispatches,
+      truncated:
+        this.truncated ||
+        calls.size > INSPECTION_LIMITS.calls ||
+        phases.length > INSPECTION_LIMITS.phases ||
+        (run.meta.phases?.length ?? 0) > INSPECTION_LIMITS.phases,
+      resultPreview: preview(run.finalResult),
+      errorPreview: text(run.error, INSPECTION_LIMITS.preview),
+    });
+  }
+}

package/src/journal.ts

@@ -154,7 +154,7 @@ export interface ScriptRun {
   callResults: Record<string, CallResult>;
   /** Call index → resolved tier identity for live and stopped calls. */
   callTiers: Record<string, CallTierIdentity>;
-  /** Call index → latest accepted nested workflow result generation. */
+  /** Monotonic nested-result watermarks; retained even when a managed call invalidates replay. */
   workflowResultGenerations: Record<string, number>;
   compactions: Record<string, number>;
   startedAt: number;
@@ -537,6 +537,8 @@ function parseTerminalResult(raw: unknown, runId: string): RecoveryTerminalResul
 }
 
 function parseRunRecovery(raw: Record<string, unknown>, run: ScriptRun, timestamp: number): RunRecoveryEvent {
+  if (run.status === "completed" || run.nonResumable || run.terminalIntent === "stop")
+    throw new Error("cannot recover a completed or non-resumable workflow run");
   if (raw.status !== "interrupted") throw new Error("run recovery status is invalid");
   if (typeof raw.branchGeneration !== "number" || !Number.isInteger(raw.branchGeneration) || raw.branchGeneration < 0) {
     throw new Error("run recovery branchGeneration is invalid");
@@ -1140,6 +1142,8 @@ export function applyRecoveryEvent(run: ScriptRun, event: RunRecoveryEvent | Att
   if (event.kind === "run_recovery") {
     run.status = "interrupted";
     run.terminalIntent = undefined;
+    run.error = undefined;
+    run.finalResult = undefined;
     for (const rotation of event.rotations) applyRecoveryRotation(run, rotation);
     applyRecoveryTerminalResults(run, event.terminalResults ?? []);
   } else {
@@ -1218,6 +1222,7 @@ function applyJournalEvent(runs: Map<string, ScriptRun>, event: JournalEvent): v
       break;
     case "workflow_transition":
       run.status = event.status;
+      if (event.status === "running") run.error = undefined;
       if (event.error !== undefined) run.error = event.error;
       if (event.finalResult !== undefined) run.finalResult = structuredClone(event.finalResult);
       if (event.terminalIntent !== undefined) run.terminalIntent = event.terminalIntent;
@@ -1353,9 +1358,10 @@ export function snapshotRun(run: ScriptRun): ScriptRun {
  * The runtime keys its cache by `${runId}:${callIndex}` (the same namespacing
  * the store deltas use), so a nested workflow() call with its own runId can
  * never replay the parent's entry. Every `call_result` or `workflow_result`
- * event that carries a `callHash` is a replayable fact. Entries from another
- * schema or without a hash are not replayable; they are ignored here even when
- * `replayJournal` retains them as non-replayable lifecycle history.
+ * event that carries a `callHash` is a replayable candidate only while it is
+ * the aggregate's final completed result. Entries from another schema or
+ * without a hash are not replayable, but accepted lifecycle history can still
+ * supersede an older candidate.
  */
 export function buildResumeJournal(entries: readonly SessionEntryLike[]): Map<
   string,
@@ -1381,7 +1387,7 @@ export function buildResumeJournal(entries: readonly SessionEntryLike[]): Map<
     agentCount?: number;
   };
   const journal = new Map<string, ResumeEntry>();
-  const journalRunIds = new Map<string, string>();
+  const journalSources = new Map<string, { runId: string; nodeId: string; result: CallResult }>();
 
   // Reuse the authoritative replay validator so stale call results from an
   // older attempt, terminal-after-lifecycle facts, and quarantined runs never
@@ -1404,6 +1410,10 @@ export function buildResumeJournal(entries: readonly SessionEntryLike[]): Map<
           : {}),
       };
       const key = `${event.runId}:${event.nodeId}`;
+      // Parsing resolves the result's tier before application, so the
+      // aggregate retains this exact accepted result object. Record even an
+      // identical nested retry: it becomes the aggregate's latest identity.
+      journalSources.set(key, { runId: event.runId, nodeId: event.nodeId, result: event.result });
       const existing = journal.get(key);
       if (candidate.generation !== undefined && existing?.generation !== undefined) {
         if (candidate.generation < existing.generation) return;
@@ -1419,19 +1429,36 @@ export function buildResumeJournal(entries: readonly SessionEntryLike[]): Map<
         }
       }
       journal.set(key, candidate);
-      journalRunIds.set(key, event.runId);
     },
   });
 
-  // replayJournal may quarantine a run after an accepted fact (for example,
-  // when a later contradictory event invalidates the aggregate). Do not retain
-  // cache entries belonging to that run.
-  for (const [key, runId] of journalRunIds) {
-    if (!replayedRuns.has(runId)) journal.delete(key);
+  // Reconcile with authoritative final state, not a second event acceptance
+  // parser. Generation rotations clear results; non-replayable replacements
+  // (including recovery terminal results) replace their identity; removal and
+  // quarantine discard the run. Rejected stale events change none of these.
+  for (const [key, source] of journalSources) {
+    const run = replayedRuns.get(source.runId);
+    if (run?.callStatus[source.nodeId] !== "completed" || run.callResults[source.nodeId] !== source.result) {
+      journal.delete(key);
+    }
   }
   return journal;
 }
 
+/**
+ * Copy accepted aggregate watermarks for WorkflowRunOptions.resumeWorkflowGenerations.
+ * Not a result cache: kind changes, rotations and failed results must not erase
+ * an earlier nested generation or make its old value eligible for replay.
+ * Supply a run returned by replayJournal (or the engine's accepted live aggregate).
+ */
+export function buildWorkflowGenerations(
+  run: Pick<ScriptRun, "workflowResultGenerations">,
+): ReadonlyMap<number, number> {
+  return new Map(
+    Object.entries(run.workflowResultGenerations).map(([index, generation]) => [Number(index), generation]),
+  );
+}
+
 /** Derive a stable recovery id from a recovery event's content. */
 export function deriveRecoveryId(event: { runId: string; rotations: unknown[]; timestamp: number }): string {
   const digest = createHash("sha256")