@tangle-network/agent-eval 0.135.4 → 0.137.0

package/dist/{default-registry-Brxr728w.d.ts → default-registry-Dc5D_Loc.d.ts}

@@ -1,9 +1,64 @@
-import { c as CostLedgerHandle } from "./cost-ledger-fGS_u_O1.js";
-import { A as ChatClient } from "./types-Cc3qbqzj.js";
-import { t as TraceAnalysisStore } from "./store-CxJry_cs.js";
-import { c as AnalystRunInputs, i as AnalystFinding, l as AnalystRunResult, n as AnalystContext, p as EvidenceRef, r as AnalystCost, s as AnalystRunEvent, t as Analyst, u as AnalystRunSummary } from "./types-DVjczBM9.js";
+import { N as TraceAnalystSpan, i as AnalystFinding, n as AnalystContext, p as EvidenceRef, r as AnalystCost, t as Analyst, x as TraceAnalysisStore } from "./types-CKswbJGO.js";
+import { n as AnalystRegistry, r as AnalystRegistryOptions } from "./registry-BdM7SuTr.js";
 import { AxAIService, AxFunction } from "@ax-llm/ax";
 import { z } from "zod";
+//#region src/trace-analyst/behavioral-metrics.d.ts
+type SuboptimalCode = 'monotonic-input-growth' | 'output-length-decay' | 'single-tool-dependency' | 'no-self-verification';
+interface SuboptimalSignal {
+  code: SuboptimalCode;
+  severity: 'high' | 'medium' | 'low';
+  /** Human-readable claim, with the backing numbers inlined. */
+  detail: string;
+  /** The exact figures the detector fired on — auditable, no model in the loop. */
+  evidence: Record<string, number | string | boolean>;
+}
+interface BehavioralMetrics {
+  /** The only trace represented by these metrics; null when spans are empty. */
+  traceId: string | null;
+  llmCallCount: number;
+  /** Causally serial LLM timelines. Parallel branches are never joined. */
+  tokenSequences: BehavioralTokenSequence[];
+  /** Token values from the longest serial timeline, retained for convenience. */
+  inputTokenTrajectory: number[];
+  outputTokenTrajectory: number[];
+  toolHistogram: Record<string, number>;
+  totalToolCalls: number;
+  distinctTools: number;
+  /** distinct/total tool calls; 1.0 when there are no tool calls. */
+  toolDiversityRatio: number;
+  hasSelfVerification: boolean;
+  signals: SuboptimalSignal[];
+}
+interface BehavioralTokenSequence {
+  scopeId: string;
+  spanIds: string[];
+  inputTokenTrajectory: Array<number | null>;
+  outputTokenTrajectory: Array<number | null>;
+}
+/**
+ * Reduce a span list to behavioral metrics + fired suboptimality signals.
+ * Pure + deterministic: same spans → same output, on any machine, no model.
+ */
+declare function computeTraceMetrics(spans: readonly TraceAnalystSpan[]): BehavioralMetrics;
+//#endregion
+//#region src/analyst/behavioral-analyst.d.ts
+interface BehavioralAnalystOptions {
+  /** Refuse larger unfiltered datasets instead of sampling them silently. Default 1,000. */
+  maxTraces?: number;
+  /** Evidence locations retained per finding. Counts still cover every analyzed trace. Default 20. */
+  maxEvidenceRefsPerFinding?: number;
+}
+/**
+ * Map computed signals → structured AnalystFindings. Pure: no LLM, no clock
+ * dependence beyond `produced_at` (overridable for deterministic tests).
+ */
+declare function deriveEfficiencyFindings(metrics: BehavioralMetrics, opts?: {
+  analystId?: string;
+  producedAt?: string;
+}): AnalystFinding[];
+/** The deterministic behavioral/efficiency analyst (no LLM, any-model). */
+declare function behavioralAnalyst(options?: BehavioralAnalystOptions): Analyst<TraceAnalysisStore>;
+//#endregion
 //#region src/analyst/finding-signature.d.ts
 declare const ANALYST_SEVERITIES: readonly ['critical', 'high', 'medium', 'low', 'info'];
 declare const RawAnalystEvidenceSchema: z.ZodObject<{
@@ -75,18 +130,6 @@ interface TraceAnalystKindSpec {
   postProcess?: (row: RawAnalystFinding, ctx: AnalystContext) => RawAnalystFinding | null;
   /** Minimum citations per finding. Default 1; rows below it are rejected. */
   minimumEvidenceCitations?: number;
-  /** Optional optimizer hook — populated when a kind wants to fit its prompt against labeled examples. */
-  goldens?: TraceAnalystGolden[];
-}
-/**
- * One labeled example consumed by Ax optimizers (MIPRO / GEPA / Bootstrap).
- * Each input is the same `{question}` an analyst would receive; `expected`
- * is the ground-truth finding set a fitted prompt should produce on this
- * input. Metric: kind-specific (default: F1 on `finding_id` overlap).
- */
-interface TraceAnalystGolden {
-  question: string;
-  expected: ReadonlyArray<Omit<RawAnalystFinding, 'confidence'>>;
 }
 interface CreateTraceAnalystKindOpts {
   /** AxAIService bound at registration time. */
@@ -138,125 +181,6 @@ declare function renderPriorFindings(prior: AnalystContext['priorFindings']): st
 /** Render findings produced earlier in this same registry run. */
 declare function renderUpstreamFindings(upstream: AnalystContext['upstreamFindings']): string;
 //#endregion
-//#region src/analyst/registry.d.ts
-interface AnalystHooks {
-  /** Before analyze() — last chance to mutate ctx (e.g. inject tags, override budget). */
-  onBeforeAnalyze?(args: {
-    analyst: Analyst;
-    ctx: AnalystContext;
-    runId: string;
-  }): void | Promise<void>;
-  /** After every analyst (ok | failed | skipped). Use for telemetry, ingestion, rotation. */
-  onAfterAnalyze?(args: {
-    analyst: Analyst;
-    summary: AnalystRunSummary;
-    findings: AnalystFinding[];
-    runId: string;
-  }): void | Promise<void>;
-  /**
-   * On analyst exception. Hook MAY return findings to convert the
-   * error into structured findings; the summary still reports 'failed'.
-   * Return void to keep the default empty-findings behavior.
-   */
-  onError?(args: {
-    analyst: Analyst;
-    error: Error;
-    runId: string;
-  }): AnalystFinding[] | undefined | Promise<AnalystFinding[] | undefined>;
-  /** Once after registry.run() completes. Use for final aggregation, persistence. */
-  onComplete?(args: {
-    result: AnalystRunResult;
-  }): void | Promise<void>;
-}
-interface BudgetPolicy {
-  /** Overall USD cap across the registry.run(). */
-  totalUsd?: number;
-  /** Per-analyst weight for the default allocator. Missing ids get weight 1. */
-  weights?: Record<string, number>;
-  /**
-   * Custom allocator — receives the analyst, remaining/total budget, and
-   * the count of analysts that will run. Returns the per-analyst budget
-   * (or undefined only when the run has no overall cap). Overrides weights
-   * when set.
-   */
-  allocate?: (args: {
-    analyst: Analyst;
-    totalUsd: number | undefined;
-    remainingUsd: number | undefined;
-    runningCount: number;
-  }) => number | undefined;
-}
-interface AnalystRegistryOptions {
-  /** Shared chat client passed to every LLM analyst via AnalystContext. */
-  chat?: ChatClient;
-  /** Logger callback. Defaults to a no-op. */
-  log?: (msg: string, fields?: Record<string, unknown>) => void;
-  /** Hooks invoked around analyze() — observability + customization seam. */
-  hooks?: AnalystHooks;
-  /** Default budget when run() doesn't override. */
-  defaultBudget?: BudgetPolicy;
-}
-interface RegistryRunOpts {
-  /** Restrict to a subset of registered analysts by id. */
-  only?: string[];
-  /** Skip these analysts even if registered. Useful for cheap iteration. */
-  skip?: string[];
-  /** Budget policy — totalUsd + optional weights/allocator. Falls back to options.defaultBudget. */
-  budget?: BudgetPolicy;
-  /** Active-work cap for the complete registry run. Model receipt settlement may follow. */
-  timeoutMs?: number;
-  /** Abort signal — forwarded into every analyst's context. */
-  signal?: AbortSignal;
-  /** Shared paid-call account forwarded to every analyst. */
-  costLedger?: CostLedgerHandle;
-  /** Attribution phase for calls written to `costLedger`. */
-  costPhase?: string;
-  /** Tags echoed into AnalystContext.tags — useful for tracking environment/version in findings. */
-  tags?: Record<string, string>;
-  /**
-   * Prior-run findings made available as retrieval context to every
-   * analyst via `ctx.priorFindings`. The registry forwards the slice
-   * whose `analyst_id` matches each registered analyst so a kind sees
-   * only its own history. Pass `{ '*': findings }` to broadcast to
-   * every analyst (useful when several kinds share the same historical
-   * context). For findings from this run, use `chainFindings` instead.
-   */
-  priorFindings?: ReadonlyArray<AnalystFinding> | Record<string, ReadonlyArray<AnalystFinding>>;
-  /**
-   * Pass findings produced earlier in this registry run to each later analyst
-   * via `ctx.upstreamFindings`. Registration order is dependency order.
-   * Disabled by default because independent analyst suites must opt in.
-   */
-  chainFindings?: boolean;
-}
-declare class AnalystRegistry {
-  private readonly analysts;
-  private readonly options;
-  constructor(options?: AnalystRegistryOptions);
-  register(analyst: Analyst): void;
-  list(): ReadonlyArray<{
-    id: string;
-    description: string;
-    version: string;
-    cost: Analyst['cost'];
-  }>;
-  run(runId: string, inputs: AnalystRunInputs, runOpts?: RegistryRunOpts): Promise<AnalystRunResult>;
-  /**
-   * Streaming counterpart to `run()`. Emits `AnalystRunEvent` values
-   * in real time — `run-started`, then per-analyst `skipped` /
-   * `started` / `completed`, then a terminal `run-completed` whose
-   * payload is the full `AnalystRunResult`. UIs use this to render
-   * progress; persistence consumers use `run()` and read the result.
-   *
-   * Hooks (`onBeforeAnalyze` / `onAfterAnalyze` / `onError` /
-   * `onComplete`) fire as before — streaming is additive, not a hook
-   * replacement.
-   */
-  runStream(runId: string, inputs: AnalystRunInputs, runOpts?: RegistryRunOpts): AsyncGenerator<AnalystRunEvent, void, void>;
-  private selectAnalysts;
-  private routeInput;
-}
-//#endregion
 //#region src/analyst/default-registry.d.ts
 interface DefaultAnalystRegistryOptions {
   /** Ax service for the agentic RLM kinds. Omit → only the deterministic analyst. */
@@ -267,10 +191,12 @@ interface DefaultAnalystRegistryOptions {
   kinds?: readonly TraceAnalystKindSpec[];
   /** Set false to omit the deterministic behavioral analyst (default: include). */
   includeBehavioral?: boolean;
+  /** Bound deterministic trace scanning without changing the store. */
+  behavioral?: BehavioralAnalystOptions;
   /** Forwarded to the AnalystRegistry constructor (signal, tags, priorFindings). */
   registry?: AnalystRegistryOptions;
 }
 declare function buildDefaultAnalystRegistry(opts?: DefaultAnalystRegistryOptions): AnalystRegistry;
 //#endregion
-export { RawAnalystEvidenceSchema as _, AnalystRegistryOptions as a, evidenceRefsFromRawFinding as b, CreateTraceAnalystKindOpts as c, createTraceAnalystKind as d, renderPriorFindings as f, RawAnalystEvidence as g, RAW_FINDING_SCHEMA_PROMPT as h, AnalystRegistry as i, TraceAnalystGolden as l, ANALYST_SEVERITIES as m, buildDefaultAnalystRegistry as n, BudgetPolicy as o, renderUpstreamFindings as p, AnalystHooks as r, RegistryRunOpts as s, DefaultAnalystRegistryOptions as t, TraceAnalystKindSpec as u, RawAnalystFinding as v, parseRawFinding as x, RawAnalystFindingSchema as y };
-//# sourceMappingURL=default-registry-Brxr728w.d.ts.map
+export { computeTraceMetrics as C, SuboptimalSignal as S, behavioralAnalyst as _, createTraceAnalystKind as a, BehavioralTokenSequence as b, ANALYST_SEVERITIES as c, RawAnalystEvidenceSchema as d, RawAnalystFinding as f, BehavioralAnalystOptions as g, parseRawFinding as h, TraceAnalystKindSpec as i, RAW_FINDING_SCHEMA_PROMPT as l, evidenceRefsFromRawFinding as m, buildDefaultAnalystRegistry as n, renderPriorFindings as o, RawAnalystFindingSchema as p, CreateTraceAnalystKindOpts as r, renderUpstreamFindings as s, DefaultAnalystRegistryOptions as t, RawAnalystEvidence as u, deriveEfficiencyFindings as v, SuboptimalCode as x, BehavioralMetrics as y };
+//# sourceMappingURL=default-registry-Dc5D_Loc.d.ts.map

package/dist/default-registry-Dc5D_Loc.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"default-registry-Dc5D_Loc.d.ts","names":[],"sources":["../src/trace-analyst/behavioral-metrics.ts","../src/analyst/behavioral-analyst.ts","../src/analyst/finding-signature.ts","../src/analyst/kind-factory.ts","../src/analyst/default-registry.ts"],"mappings":";;;;;KAoBY;UAMK;EACf,MAAM;EACN;;EAEA;;EAEA,UAAU;;UAGK;;EAEf;EACA;;EAEA,gBAAgB;;EAEhB;EACA;EACA,eAAe;EACf;EACA;;EAEA;EACA;EACA,SAAS;;UAGM;EACf;EACA;EACA,sBAAsB;EACtB,uBAAuB;;;;;;iBA+DT,oBAAoB,gBAAgB,qBAAqB;;;UC1FxD;;EAEf;;EAEA;;;;;;iBA+Dc,yBACd,SAAS,mBACT;EAAQ;EAAoB;IAC3B;;iBAkCa,kBACd,UAAS,2BACR,QAAQ;;;cCrHE;cAEA,0BAAwB,EAAA;;;GAK1B,EAAA,KAAA;KAEC,qBAAqB,EAAE,aAAa;cAiBnC,yBAAuB,EAAA;;;;;;;;;;;;;;;;;GAKzB,EAAA,KAAA;KAEC,oBAAoB,EAAE,aAAa;;;;;;cAOlC;;iBAYG,2BAA2B,SAAS,oBAAoB;iBAQxD,gBACd,cACA,OAAO,aAAa,SAAS,mCAC5B;;;;;;;UCrCc;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,aAAa,OAAO,uBAAuB;;EAE3C;IAAe;IAAkB;;;EAEjC;;EAEA;;EAEA;;EAEA,MAAM;;EAEN,eAAe,KAAK,mBAAmB,KAAK,mBAAmB;;EAE/D;;UAGe;;EAEf,IAAI;;EAEJ;;EAEA;;;;;;;;EAQA;IAAa;IAAiB;IAAiB;IAAgB,mBAAmB;;;EAElF;;;;;;;;;;iBAWc,uBACd,MAAM,sBACN,MAAM,6BACL,QAAQ;;;;;;;;;;;;;;;iBAgRK,oBAAoB,OAAO;;iBAyB3B,uBAAuB,UAAU;;;UChYhC;;EAEf,KAAK;;EAEL;;EAEA,iBAAiB;;EAEjB;;EAEA,aAAa;;EAEb,WAAW;;iBAGG,4BACd,OAAM,gCACL"}

package/dist/{errors-8YnH8WlF.js → errors-D-LKuDhb.js}

@@ -18,6 +18,12 @@ var ValidationError = class extends AgentEvalError {
 		super("validation", message, options);
 	}
 };
+/** A bounded operation or resource exceeds its documented hard limit. */
+var LimitExceededError = class extends AgentEvalError {
+	constructor(message, options) {
+		super("limit_exceeded", message, options);
+	}
+};
 /** A named resource (run, span, rubric, scenario, dataset row, route) does not exist. */
 var NotFoundError = class extends AgentEvalError {
 	constructor(message, options) {
@@ -59,6 +65,6 @@ var ReplayError = class extends AgentEvalError {
 	}
 };
 //#endregion
-export { NotFoundError as a, VerificationError as c, JudgeError as i, CaptureIntegrityError as n, ReplayError as o, ConfigError as r, ValidationError as s, AgentEvalError as t };
+export { LimitExceededError as a, ValidationError as c, JudgeError as i, VerificationError as l, CaptureIntegrityError as n, NotFoundError as o, ConfigError as r, ReplayError as s, AgentEvalError as t };
 
-//# sourceMappingURL=errors-8YnH8WlF.js.map
+//# sourceMappingURL=errors-D-LKuDhb.js.map

package/dist/errors-D-LKuDhb.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"errors-D-LKuDhb.js","names":[],"sources":["../src/errors.ts"],"sourcesContent":["/**\n * Error taxonomy for `@tangle-network/agent-eval`.\n *\n * Every error this package throws as part of its *public contract* extends\n * `AgentEvalError`. Consumers can pattern-match by `instanceof <Subclass>` or\n * by the stable string `code` carried on the base class.\n *\n * The codes are stable across minor versions; new codes can be added, but\n * existing codes never change meaning. New subclasses are non-breaking.\n *\n * Internal invariant guards (`throw new Error('this should never happen')`)\n * remain plain `Error`s on purpose — they're programmer-mistake assertions,\n * not consumer-catchable contract failures.\n */\n\nexport type AgentEvalErrorCode =\n  | 'validation'\n  | 'limit_exceeded'\n  | 'not_found'\n  | 'config'\n  | 'capture_integrity'\n  | 'judge'\n  | 'verification'\n  | 'replay'\n  | 'backend_integrity'\n  | 'profile_matrix'\n\n/**\n * Base class for every contract error this package throws — carries the stable\n * string `code` taxonomy so consumers can `instanceof`-match or switch on `code`.\n */\nexport class AgentEvalError extends Error {\n  /** Stable string code. Survives minification; safe to switch on. */\n  readonly code: AgentEvalErrorCode\n\n  constructor(code: AgentEvalErrorCode, message: string, options?: { cause?: unknown }) {\n    super(message, options)\n    this.name = this.constructor.name\n    this.code = code\n  }\n}\n\n/** Caller passed invalid arguments (out of range, mutually-exclusive options, bad shape). */\nexport class ValidationError extends AgentEvalError {\n  constructor(message: string, options?: { cause?: unknown }) {\n    super('validation', message, options)\n  }\n}\n\n/** A bounded operation or resource exceeds its documented hard limit. */\nexport class LimitExceededError extends AgentEvalError {\n  constructor(message: string, options?: { cause?: unknown }) {\n    super('limit_exceeded', message, options)\n  }\n}\n\n/** A named resource (run, span, rubric, scenario, dataset row, route) does not exist. */\nexport class NotFoundError extends AgentEvalError {\n  constructor(message: string, options?: { cause?: unknown }) {\n    super('not_found', message, options)\n  }\n}\n\n/** Configuration missing or malformed (`HOME` unset, required image not supplied, env var absent). */\nexport class ConfigError extends AgentEvalError {\n  constructor(message: string, options?: { cause?: unknown }) {\n    super('config', message, options)\n  }\n}\n\n/**\n * A run is missing the artifacts a launch-grade check requires:\n * raw HTTP capture absent, no LLM spans, route assertion failed, run-end\n * assertion tripped. Block ship on this; do not catch and move on.\n */\nexport class CaptureIntegrityError extends AgentEvalError {\n  constructor(message: string, options?: { cause?: unknown }) {\n    super('capture_integrity', message, options)\n  }\n}\n\n/** A judge call failed in a way that's not retryable: schema parse failure, bad rubric, conflicting dimensions. */\nexport class JudgeError extends AgentEvalError {\n  constructor(message: string, options?: { cause?: unknown }) {\n    super('judge', message, options)\n  }\n}\n\n/** A verifier signalled a hard failure (compile, test, schema) — distinct from a low judge score. */\nexport class VerificationError extends AgentEvalError {\n  constructor(message: string, options?: { cause?: unknown }) {\n    super('verification', message, options)\n  }\n}\n\n/** Replay cache cannot satisfy a request: miss with no fallback, sink lacks list(), unsupported URL. */\nexport class ReplayError extends AgentEvalError {\n  constructor(message: string, options?: { cause?: unknown }) {\n    super('replay', message, options)\n  }\n}\n"],"mappings":";;;;;AA+BA,IAAa,iBAAb,cAAoC,MAAM;;CAExC;CAEA,YAAY,MAA0B,SAAiB,SAA+B;EACpF,MAAM,SAAS,OAAO;EACtB,KAAK,OAAO,KAAK,YAAY;EAC7B,KAAK,OAAO;CACd;AACF;;AAGA,IAAa,kBAAb,cAAqC,eAAe;CAClD,YAAY,SAAiB,SAA+B;EAC1D,MAAM,cAAc,SAAS,OAAO;CACtC;AACF;;AAGA,IAAa,qBAAb,cAAwC,eAAe;CACrD,YAAY,SAAiB,SAA+B;EAC1D,MAAM,kBAAkB,SAAS,OAAO;CAC1C;AACF;;AAGA,IAAa,gBAAb,cAAmC,eAAe;CAChD,YAAY,SAAiB,SAA+B;EAC1D,MAAM,aAAa,SAAS,OAAO;CACrC;AACF;;AAGA,IAAa,cAAb,cAAiC,eAAe;CAC9C,YAAY,SAAiB,SAA+B;EAC1D,MAAM,UAAU,SAAS,OAAO;CAClC;AACF;;;;;;AAOA,IAAa,wBAAb,cAA2C,eAAe;CACxD,YAAY,SAAiB,SAA+B;EAC1D,MAAM,qBAAqB,SAAS,OAAO;CAC7C;AACF;;AAGA,IAAa,aAAb,cAAgC,eAAe;CAC7C,YAAY,SAAiB,SAA+B;EAC1D,MAAM,SAAS,SAAS,OAAO;CACjC;AACF;;AAGA,IAAa,oBAAb,cAAuC,eAAe;CACpD,YAAY,SAAiB,SAA+B;EAC1D,MAAM,gBAAgB,SAAS,OAAO;CACxC;AACF;;AAGA,IAAa,cAAb,cAAiC,eAAe;CAC9C,YAAY,SAAiB,SAA+B;EAC1D,MAAM,UAAU,SAAS,OAAO;CAClC;AACF"}

package/dist/{errors-CEk209JS.d.ts → errors-DkfjIDvD.d.ts}

@@ -13,7 +13,7 @@
  * remain plain `Error`s on purpose — they're programmer-mistake assertions,
  * not consumer-catchable contract failures.
  */
-type AgentEvalErrorCode = 'validation' | 'not_found' | 'config' | 'capture_integrity' | 'judge' | 'verification' | 'replay' | 'backend_integrity' | 'profile_matrix';
+type AgentEvalErrorCode = 'validation' | 'limit_exceeded' | 'not_found' | 'config' | 'capture_integrity' | 'judge' | 'verification' | 'replay' | 'backend_integrity' | 'profile_matrix';
 /**
  * Base class for every contract error this package throws — carries the stable
  * string `code` taxonomy so consumers can `instanceof`-match or switch on `code`.
@@ -31,6 +31,12 @@ declare class ValidationError extends AgentEvalError {
     cause?: unknown;
   });
 }
+/** A bounded operation or resource exceeds its documented hard limit. */
+declare class LimitExceededError extends AgentEvalError {
+  constructor(message: string, options?: {
+    cause?: unknown;
+  });
+}
 /** A named resource (run, span, rubric, scenario, dataset row, route) does not exist. */
 declare class NotFoundError extends AgentEvalError {
   constructor(message: string, options?: {
@@ -72,5 +78,5 @@ declare class ReplayError extends AgentEvalError {
   });
 }
 //#endregion
-export { JudgeError as a, ValidationError as c, ConfigError as i, VerificationError as l, AgentEvalErrorCode as n, NotFoundError as o, CaptureIntegrityError as r, ReplayError as s, AgentEvalError as t };
-//# sourceMappingURL=errors-CEk209JS.d.ts.map
+export { JudgeError as a, ReplayError as c, ConfigError as i, ValidationError as l, AgentEvalErrorCode as n, LimitExceededError as o, CaptureIntegrityError as r, NotFoundError as s, AgentEvalError as t, VerificationError as u };
+//# sourceMappingURL=errors-DkfjIDvD.d.ts.map

package/dist/errors-DkfjIDvD.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"errors-DkfjIDvD.d.ts","names":[],"sources":["../src/errors.ts"],"mappings":";;;;;;;;;;;;;;;KAeY;;;;;cAgBC,uBAAuB;;WAEzB,MAAM;EAEf,YAAY,MAAM,oBAAoB,iBAAiB;IAAY;;;;cAQxD,wBAAwB;EACnC,YAAY,iBAAiB;IAAY;;;;cAM9B,2BAA2B;EACtC,YAAY,iBAAiB;IAAY;;;;cAM9B,sBAAsB;EACjC,YAAY,iBAAiB;IAAY;;;;cAM9B,oBAAoB;EAC/B,YAAY,iBAAiB;IAAY;;;;;;;;cAU9B,8BAA8B;EACzC,YAAY,iBAAiB;IAAY;;;;cAM9B,mBAAmB;EAC9B,YAAY,iBAAiB;IAAY;;;;cAM9B,0BAA0B;EACrC,YAAY,iBAAiB;IAAY;;;;cAM9B,oBAAoB;EAC/B,YAAY,iBAAiB;IAAY"}

package/dist/{eval-campaign-Cc8WZJ6b.js → eval-campaign-CHqfLnff.js}

@@ -1,10 +1,10 @@
-import { f as verifyAgentProfileCell, h as hashJson, p as canonicalize, s as buildAgentProfileCell } from "./agent-profile-cell-OhuTee9n.js";
+import { f as verifyAgentProfileCell, h as hashJson, p as canonicalize, s as buildAgentProfileCell } from "./agent-profile-cell-CbfBm2g6.js";
 import { t as FileSystemRawProviderSink } from "./raw-provider-sink-BQd7mzyT.js";
-import { a as assertLlmRoute } from "./llm-client-DHx8pzyJ.js";
+import { a as assertLlmRoute } from "./llm-client-DKB25jV8.js";
 import { t as TraceEmitter } from "./emitter-CPBAhxum.js";
-import { s as validateRunRecord } from "./run-record-BIwU2wdV.js";
-import { i as researchReport } from "./summary-report-BEk8OFLs.js";
-import { n as assertRunCaptured, t as RunIntegrityError } from "./integrity-BzRbCHzi.js";
+import { s as validateRunRecord } from "./run-record-vRgqWmJw.js";
+import { i as researchReport } from "./summary-report-9A5y7EsK.js";
+import { n as assertRunCaptured, t as RunIntegrityError } from "./integrity-fdt8XPAv.js";
 //#region src/eval-campaign.ts
 /**
 * EvalCampaign — opinionated matrix runner that wires the four
@@ -346,4 +346,4 @@ function defaultRunId(params) {
 //#endregion
 export { runEvalCampaign as t };
 
-//# sourceMappingURL=eval-campaign-Cc8WZJ6b.js.map
+//# sourceMappingURL=eval-campaign-CHqfLnff.js.map

package/dist/{eval-campaign-Cc8WZJ6b.js.map → eval-campaign-CHqfLnff.js.map}

@@ -1 +1 @@
-{"version":3,"file":"eval-campaign-Cc8WZJ6b.js","names":[],"sources":["../src/eval-campaign.ts"],"sourcesContent":["/**\n * EvalCampaign — opinionated matrix runner that wires the four\n * capture-integrity directives by construction.\n *\n * The canonical benchmark shape — matrix runner → for each\n * (variant, scenario, seed) → start a TraceEmitter → call LLMs → end the\n * run → analyze — has a bug class at the integration boundary: raw\n * events not captured, route silently wrong, integrity not asserted,\n * analyst never run. The directives in `SKILL.md § Capture integrity`\n * are the mitigations.\n *\n * `EvalCampaign` is the structural fix — consumers don't wire the\n * integrity surface themselves; the campaign owns it. Specifically:\n *\n *   - calls `assertLlmRoute` once at preflight before any work runs\n *   - constructs a per-run `TraceStore` and `RawProviderSink` via factories\n *   - constructs the `TraceEmitter` with `onRunComplete: [analyst hook]`\n *   - hands the runner an `LlmClientOptions` pre-wired with the sink and\n *     trace context — the runner can't accidentally call an LLM without\n *     capturing the raw HTTP envelope\n *   - calls `assertRunCaptured` after every `endRun` and routes failures\n *     through a configurable policy (`throw` / `mark_failed` / `log`)\n *   - assembles per-run `RunRecord`s and runs `researchReport` at the end\n *     so the campaign artifact is launch-decision-grade by default\n *   - embeds the campaign fingerprint (a SHA-256 over the canonicalised\n *     run set) and optional `preregistrationHash` in the report\n *\n * The runner contract is intentionally narrow: produce a `CampaignRunOutcome`\n * given a fully-wired `CampaignRunContext`. Everything orchestration-shaped\n * lives in the campaign. This is the inversion-of-control point — consumers\n * stop writing matrix runners and start writing scenario-runners.\n *\n * Out of scope for v1 (tracked in `docs/research-report-methodology.md`):\n *\n *   - Distributed/cluster execution (concurrency is local async)\n *   - Adaptive sampling / sequential interim looks\n *   - Resume from partial state across crashes\n *   - LLM-call retry beyond what `LlmClient` already does\n */\n\nimport {\n  type AgentProfileCell,\n  type AgentProfileCellInput,\n  buildAgentProfileCell,\n  verifyAgentProfileCell,\n} from './agent-profile-cell'\nimport { assertLlmRoute, type LlmClientOptions, type LlmRouteRequirements } from './llm-client'\nimport { canonicalize, hashJson } from './pre-registration'\nimport type {\n  JudgeScoresRecord,\n  RunCostProvenance,\n  RunJudgeMetadata,\n  RunOutcome,\n  RunRecord,\n  RunSplitTag,\n  RunTaskFailure,\n  RunTokenUsage,\n} from './run-record'\nimport { validateRunRecord } from './run-record'\nimport { type ResearchReport, type ResearchReportOptions, researchReport } from './summary-report'\nimport type { RunCompleteHook } from './trace/emitter'\nimport { TraceEmitter } from './trace/emitter'\nimport {\n  assertRunCaptured,\n  RunIntegrityError,\n  type RunIntegrityExpectations,\n  type RunIntegrityReport,\n} from './trace/integrity'\nimport { FileSystemRawProviderSink, type RawProviderSink } from './trace/raw-provider-sink'\nimport type { TraceStore } from './trace/store'\n\n// ── Public types ─────────────────────────────────────────────────────────\n\nexport interface CampaignVariant<V> {\n  id: string\n  payload: V\n}\n\nexport interface CampaignScenario {\n  scenarioId: string\n  /** Free-form metadata propagated to runs and reports. */\n  tags?: Record<string, string>\n}\n\nexport interface CampaignRunContext<V> {\n  /** Stable run id. The campaign generates this; the runner does not. */\n  runId: string\n  /** Logical experiment id (campaignId by default; overridable per-run via opts). */\n  experimentId: string\n  variant: V\n  variantId: string\n  scenarioId: string\n  scenarioTags: Record<string, string>\n  seed: number\n  splitTag: RunSplitTag\n  /**\n   * The TraceEmitter for this run, with `onRunComplete` hooks pre-wired\n   * (analyst auto-execution if configured, plus integrity check). The\n   * runner MUST call `emitter.startRun` before doing any work and either\n   * `emitter.endRun` or `emitter.abortRun` before returning.\n   */\n  emitter: TraceEmitter\n  store: TraceStore\n  rawSink: RawProviderSink\n  /**\n   * Pre-wired LLM client options — `rawSink` and `traceContext` are populated\n   * so any `callLlm(req, ctx.llmOpts)` automatically captures raw HTTP. The\n   * runner can spread additional fields if needed.\n   */\n  llmOpts: LlmClientOptions\n}\n\ninterface CampaignRunOutcomeFields {\n  /** Did the run pass? Mirrors `RunOutcome.pass` semantics. */\n  pass: boolean\n  /** Score for the run on its split. Maps to `searchScore` or `holdoutScore`. */\n  score: number\n  /** Cost in USD, or null when the runner could not capture it. */\n  costUsd: number | null\n  /** Source of the cost amount. */\n  costProvenance: RunCostProvenance\n  tokenUsage: RunTokenUsage\n  /** Snapshot model id (e.g. `claude-sonnet-4-6@2025-04-15`). */\n  model: string\n  /** sha256 of the effective prompt sent to the model. */\n  promptHash: string\n  /** sha256 of the effective config (model, temperature, tools, judges, splits). */\n  configHash: string\n  /** Optional extra numeric metrics to land in `outcome.raw`. */\n  raw?: Record<string, number>\n  /** Optional judge metadata when a judge was used. */\n  judgeMetadata?: RunJudgeMetadata\n  /**\n   * Optional per-judge / per-dim breakdown for ensemble-judged runs.\n   * Propagated to `outcome.judgeScores` on the resulting `RunRecord`.\n   * Single-judge or scalar-only runs leave this unset.\n   */\n  judgeScores?: JudgeScoresRecord\n  /**\n   * Agent profile cell observed by the runner. When supplied, it overrides\n   * `EvalCampaignOptions.agentProfile` for this run and must match the\n   * outcome's `model` and `promptHash`.\n   */\n  agentProfile?: AgentProfileCell | AgentProfileCellInput\n}\n\n/** Campaign result with the same task-failure invariant as `RunRecord`. */\nexport type CampaignRunOutcome = CampaignRunOutcomeFields & RunTaskFailure\n\nexport type CampaignRunner<V> = (ctx: CampaignRunContext<V>) => Promise<CampaignRunOutcome>\n\nexport type CampaignIntegrityPolicy = 'throw' | 'mark_failed' | 'log'\n\nexport interface EvalCampaignOptions<V> {\n  /**\n   * Stable id for the campaign. Used as the default `experimentId` on\n   * every run, and folded into the campaign fingerprint.\n   */\n  campaignId: string\n  variants: CampaignVariant<V>[]\n  scenarios: CampaignScenario[]\n  /** Default `[0, 1, 2]`. */\n  seeds?: number[]\n  /** Default `'holdout'` — the split that anchors a launch decision. */\n  splitTag?: RunSplitTag\n  /** Git SHA the campaign is run against. Mandatory; `RunRecord` rejects unset. */\n  commitSha: string\n  /**\n   * LLM client config. Augmented per-run with `rawSink` and `traceContext`\n   * before being passed to the runner. The campaign asserts this config\n   * matches `routeRequirements` once at preflight.\n   */\n  llmOpts: LlmClientOptions\n  /**\n   * Default `{ requireExplicitBaseUrl: true, requireAuth: true }` — fail\n   * loud if the campaign would silently fall back to the public router or\n   * run unauthenticated. Override with an empty object to disable.\n   */\n  routeRequirements?: LlmRouteRequirements\n  /**\n   * Per-run TraceStore factory. Common shape: a fresh store per run keyed\n   * on `runId`. Implementations that share a store across the campaign\n   * are valid — the campaign only writes through `emitter`.\n   */\n  storeFactory: (params: CampaignFactoryParams) => TraceStore\n  /**\n   * Per-run RawProviderSink factory. Defaults to `FileSystemRawProviderSink`\n   * rooted at `${workDir}/raw-events/${runId}` if `workDir` is supplied;\n   * otherwise required. Forensic capture is non-negotiable in a campaign\n   * run — pass `NoopRawProviderSink` explicitly if you want to opt out.\n   */\n  rawSinkFactory?: (params: CampaignFactoryParams) => RawProviderSink\n  /**\n   * Filesystem root for default `rawSinkFactory`. Ignored if\n   * `rawSinkFactory` is supplied.\n   */\n  workDir?: string\n  /**\n   * Extra `onRunComplete` hooks the campaign appends (after its own\n   * integrity-check hook). Pass `traceAnalystOnRunComplete(...)` here.\n   */\n  onRunComplete?: RunCompleteHook[]\n  /**\n   * Per-run integrity expectations. Defaults to:\n   *   `{ llmSpansMin: 1, requireRawCoverageOfLlmSpans: true, requireOutcome: true }`.\n   * Override (e.g. `{ llmSpansMin: 0 }`) for runs that don't call LLMs.\n   */\n  integrity?: RunIntegrityExpectations\n  /** Behaviour when integrity fails. Default `'mark_failed'`. */\n  onIntegrityFailure?: CampaignIntegrityPolicy\n  /**\n   * Per-run runner. Receives a fully-wired context; produces an outcome\n   * the campaign converts into a `RunRecord`.\n   */\n  runner: CampaignRunner<V>\n  /**\n   * If set, the campaign computes `researchReport` at the end. `comparator`\n   * is a `variantId`. Other fields are forwarded verbatim.\n   */\n  report?: { comparator?: string } & Omit<\n    ResearchReportOptions,\n    'comparator' | 'preregistrationHash' | 'generatedAt'\n  >\n  /**\n   * Hash of a signed `HypothesisManifest` (see `pre-registration.ts`).\n   * Embedded in the campaign fingerprint and the research report.\n   */\n  preregistrationHash?: string\n  /** Local concurrency. Default `1` (sequential). */\n  concurrency?: number\n  /**\n   * Override the time source. Tests pass a mock to make wallMs deterministic.\n   */\n  now?: () => number\n  /** Override the runId generator. Tests pin this. */\n  runId?: (params: CampaignFactoryParams) => string\n  /**\n   * Agent profile cell for campaign runs. Static profiles can pass an object;\n   * routers or variant-specific harnesses can pass a factory. The campaign\n   * stamps the built cell onto every `RunRecord` and rejects profile/model or\n   * profile/prompt contradictions.\n   */\n  agentProfile?:\n    | AgentProfileCell\n    | AgentProfileCellInput\n    | ((\n        params: CampaignFactoryParams & {\n          variant: V\n          scenarioTags: Record<string, string>\n        },\n      ) =>\n        | AgentProfileCell\n        | AgentProfileCellInput\n        | Promise<AgentProfileCell | AgentProfileCellInput>)\n}\n\nexport interface CampaignFactoryParams {\n  campaignId: string\n  runId: string\n  variantId: string\n  scenarioId: string\n  seed: number\n}\n\nexport interface FailedRun {\n  runId: string\n  variantId: string\n  scenarioId: string\n  seed: number\n  reason: string\n  error?: string\n}\n\nexport interface EvalCampaignResult {\n  campaignId: string\n  /** SHA-256 over canonicalised `(variantIds, scenarioIds, seeds, comparator, splitTag, baseUrl, provider, preregistrationHash)`. */\n  campaignFingerprint: string\n  preregistrationHash: string | null\n  /** Successful runs only. Failed runs land in `failedRuns`. */\n  runs: RunRecord[]\n  /** Integrity reports for every successful run. */\n  integrityReports: RunIntegrityReport[]\n  failedRuns: FailedRun[]\n  /** Computed when `report` is set on options. */\n  report?: ResearchReport\n  startedAt: string\n  endedAt: string\n}\n\n// ── Implementation ───────────────────────────────────────────────────────\n\nconst DEFAULT_INTEGRITY: RunIntegrityExpectations = {\n  llmSpansMin: 1,\n  requireRawCoverageOfLlmSpans: true,\n  requireOutcome: true,\n}\n\nconst DEFAULT_ROUTE: LlmRouteRequirements = {\n  requireExplicitBaseUrl: true,\n  requireAuth: true,\n}\n\nexport async function runEvalCampaign<V>(\n  opts: EvalCampaignOptions<V>,\n): Promise<EvalCampaignResult> {\n  // ── Preflight ──────────────────────────────────────────────────────\n  assertLlmRoute(opts.llmOpts, opts.routeRequirements ?? DEFAULT_ROUTE)\n\n  if (opts.variants.length === 0) {\n    throw new Error('runEvalCampaign: variants must be non-empty.')\n  }\n  if (opts.scenarios.length === 0) {\n    throw new Error('runEvalCampaign: scenarios must be non-empty.')\n  }\n  const variantIds = new Set<string>()\n  for (const v of opts.variants) {\n    if (variantIds.has(v.id)) {\n      throw new Error(`runEvalCampaign: duplicate variant id \"${v.id}\".`)\n    }\n    variantIds.add(v.id)\n  }\n  const scenarioIds = new Set<string>()\n  for (const s of opts.scenarios) {\n    if (scenarioIds.has(s.scenarioId)) {\n      throw new Error(`runEvalCampaign: duplicate scenarioId \"${s.scenarioId}\".`)\n    }\n    scenarioIds.add(s.scenarioId)\n  }\n  if (opts.report?.comparator && !variantIds.has(opts.report.comparator)) {\n    throw new Error(\n      `runEvalCampaign: report.comparator \"${opts.report.comparator}\" is not a configured variantId.`,\n    )\n  }\n  if (!opts.commitSha) {\n    throw new Error('runEvalCampaign: commitSha is required (every RunRecord needs it).')\n  }\n\n  const seeds = opts.seeds ?? [0, 1, 2]\n  const splitTag: RunSplitTag = opts.splitTag ?? 'holdout'\n  const concurrency = Math.max(1, opts.concurrency ?? 1)\n  const integrity = { ...DEFAULT_INTEGRITY, ...(opts.integrity ?? {}) }\n  const onIntegrityFailure: CampaignIntegrityPolicy = opts.onIntegrityFailure ?? 'mark_failed'\n  const now = opts.now ?? (() => Date.now())\n  const baseUrl = (opts.llmOpts.baseUrl ?? '').replace(/\\/+$/, '')\n  const provider = opts.llmOpts.provider ?? null\n  const preregistrationHash = opts.preregistrationHash ?? null\n\n  const rawSinkFactory = opts.rawSinkFactory ?? defaultRawSinkFactory(opts.workDir)\n\n  // ── Fingerprint ────────────────────────────────────────────────────\n  const campaignFingerprint = await hashJson(\n    canonicalize({\n      campaignId: opts.campaignId,\n      variants: opts.variants.map((v) => v.id).sort(),\n      scenarios: opts.scenarios.map((s) => s.scenarioId).sort(),\n      seeds: [...seeds].sort((a, b) => a - b),\n      splitTag,\n      comparator: opts.report?.comparator ?? null,\n      baseUrl,\n      provider,\n      preregistrationHash,\n    }),\n  )\n\n  // ── Plan the matrix ────────────────────────────────────────────────\n  type Cell = { variant: CampaignVariant<V>; scenario: CampaignScenario; seed: number }\n  const cells: Cell[] = []\n  for (const variant of opts.variants) {\n    for (const scenario of opts.scenarios) {\n      for (const seed of seeds) {\n        cells.push({ variant, scenario, seed })\n      }\n    }\n  }\n\n  const startedAt = new Date(now()).toISOString()\n  const runs: RunRecord[] = []\n  const integrityReports: RunIntegrityReport[] = []\n  const failedRuns: FailedRun[] = []\n\n  // ── Execute (bounded-concurrency worker pool) ──────────────────────\n  // A genuine (non-CellExecutionError) error from any worker is a bug, not a\n  // run-level failure. We must NOT let `Promise.all` reject mid-flight and\n  // orphan the other workers' in-progress runs (emitters never finalized,\n  // sinks/handles leak, partial work silently discarded). Instead: capture the\n  // first genuine error, stop dispatching new cells so in-flight workers wind\n  // down, finalize every still-open run, then re-throw the aggregated error.\n  let cursor = 0\n  let aborting = false\n  const genuineErrors: unknown[] = []\n  // Emitters for runs that are currently mid-flight, keyed by runId. A worker\n  // registers its emitter before invoking the runner and removes it once the\n  // run is finalized (via endRun/abortRun, success or failure). Anything left\n  // here after the pool settles is an orphan we must abort.\n  const openRuns = new Map<string, TraceEmitter>()\n\n  async function worker(): Promise<void> {\n    while (!aborting) {\n      const i = cursor++\n      if (i >= cells.length) return\n      const cell = cells[i]!\n      try {\n        const result = await runOneCell(cell)\n        runs.push(result.record)\n        integrityReports.push(result.integrity)\n      } catch (err) {\n        if (err instanceof CellExecutionError) {\n          failedRuns.push(err.failed)\n          if (err.integrity) integrityReports.push(err.integrity)\n        } else {\n          // Genuine bug — not a runner failure, not an integrity failure.\n          // Capture it and stop dispatching so peers wind down gracefully;\n          // re-thrown after the pool settles. Do not surface here — that would\n          // reject Promise.all and orphan the other workers' runs.\n          genuineErrors.push(err)\n          aborting = true\n          return\n        }\n      }\n    }\n  }\n\n  async function runOneCell(\n    cell: Cell,\n  ): Promise<{ record: RunRecord; integrity: RunIntegrityReport }> {\n    const runId = (opts.runId ?? defaultRunId)({\n      campaignId: opts.campaignId,\n      runId: '', // unused by default generator\n      variantId: cell.variant.id,\n      scenarioId: cell.scenario.scenarioId,\n      seed: cell.seed,\n    })\n    const factoryParams: CampaignFactoryParams = {\n      campaignId: opts.campaignId,\n      runId,\n      variantId: cell.variant.id,\n      scenarioId: cell.scenario.scenarioId,\n      seed: cell.seed,\n    }\n    const store = opts.storeFactory(factoryParams)\n    const rawSink = rawSinkFactory(factoryParams)\n\n    const emitter = new TraceEmitter(store, {\n      runId,\n      now: opts.now,\n      onRunComplete: opts.onRunComplete,\n    })\n    // Track this run as open so a genuine error elsewhere in the pool can\n    // finalize it instead of orphaning it. Removed in the finally below.\n    openRuns.set(runId, emitter)\n\n    const llmOpts: LlmClientOptions = {\n      ...opts.llmOpts,\n      rawSink,\n      traceContext: { runId },\n    }\n\n    const ctx: CampaignRunContext<V> = {\n      runId,\n      experimentId: opts.campaignId,\n      variant: cell.variant.payload,\n      variantId: cell.variant.id,\n      scenarioId: cell.scenario.scenarioId,\n      scenarioTags: cell.scenario.tags ?? {},\n      seed: cell.seed,\n      splitTag,\n      emitter,\n      store,\n      rawSink,\n      llmOpts,\n    }\n\n    try {\n      const wallStart = now()\n      let outcome: CampaignRunOutcome\n      try {\n        outcome = await opts.runner(ctx)\n      } catch (err) {\n        const message = err instanceof Error ? err.message : String(err)\n        // The runner threw mid-execution. Abort the run so the emitter\n        // finalizes. The only benign abortRun failure is \"the runner never\n        // started the run\" (nothing to finalize). A store-write failure (disk\n        // full, FS error) is a genuine diagnostic — surface it rather than\n        // masking it as a plain runner failure.\n        await finalizeAbort(emitter, runId, message)\n        throw new CellExecutionError({\n          runId,\n          variantId: cell.variant.id,\n          scenarioId: cell.scenario.scenarioId,\n          seed: cell.seed,\n          reason: 'runner_threw',\n          error: message,\n        })\n      }\n      const wallMs = now() - wallStart\n\n      const integrityReport = await assertRunCaptured(store, runId, { ...integrity, rawSink })\n      if (!integrityReport.ok) {\n        switch (onIntegrityFailure) {\n          case 'throw':\n            throw new RunIntegrityError(integrityReport)\n          case 'mark_failed':\n            throw new CellExecutionError(\n              {\n                runId,\n                variantId: cell.variant.id,\n                scenarioId: cell.scenario.scenarioId,\n                seed: cell.seed,\n                reason: 'integrity_failed',\n                error: integrityReport.issues.map((i) => i.code).join(', '),\n              },\n              integrityReport,\n            )\n          case 'log':\n            // Caller wants the run admitted with a flagged report; fall through.\n            break\n        }\n      }\n\n      const recordOutcome: RunOutcome = {\n        raw: outcome.raw ?? {},\n      }\n      if (splitTag === 'holdout') recordOutcome.holdoutScore = outcome.score\n      else recordOutcome.searchScore = outcome.score\n      if (outcome.judgeScores !== undefined) recordOutcome.judgeScores = outcome.judgeScores\n\n      const record: RunRecord = {\n        runId,\n        experimentId: opts.campaignId,\n        candidateId: cell.variant.id,\n        seed: cell.seed,\n        model: outcome.model,\n        promptHash: outcome.promptHash,\n        configHash: outcome.configHash,\n        commitSha: opts.commitSha,\n        wallMs,\n        costUsd: outcome.costUsd,\n        costProvenance: outcome.costProvenance,\n        tokenUsage: outcome.tokenUsage,\n        terminalOutcome: 'succeeded',\n        judgeMetadata: outcome.judgeMetadata,\n        outcome: recordOutcome,\n        ...(outcome.failureClass ? { failureClass: outcome.failureClass } : {}),\n        failureMode: outcome.failureMode,\n        splitTag,\n        scenarioId: cell.scenario.scenarioId,\n      }\n      const profileSource =\n        outcome.agentProfile ??\n        (typeof opts.agentProfile === 'function'\n          ? await opts.agentProfile({\n              campaignId: opts.campaignId,\n              runId,\n              variantId: cell.variant.id,\n              scenarioId: cell.scenario.scenarioId,\n              seed: cell.seed,\n              variant: cell.variant.payload,\n              scenarioTags: cell.scenario.tags ?? {},\n            })\n          : opts.agentProfile)\n      if (profileSource !== undefined) {\n        const agentProfile = await resolveAgentProfileCell(profileSource)\n        assertAgentProfileMatchesRun(agentProfile, outcome.model, outcome.promptHash)\n        record.agentProfile = agentProfile\n      }\n      return { record: validateRunRecord(record), integrity: integrityReport }\n    } finally {\n      // This run's worker has finished with it (success, run-level failure, or\n      // genuine error). It is no longer the pool's job to finalize — drop it so\n      // the post-settle sweep doesn't double-abort a finalized run.\n      openRuns.delete(runId)\n    }\n  }\n\n  const workers = Array.from({ length: Math.min(concurrency, cells.length) }, () => worker())\n  // allSettled (not all): a genuine error in one worker must not reject the\n  // pool mid-flight and orphan the others. Each worker captures its own\n  // genuine error into `genuineErrors` and returns; we re-throw below.\n  await Promise.allSettled(workers)\n\n  // Finalize any run still open after the pool wound down. With the\n  // stop-dispatch flag these are the runs that were mid-flight in peer workers\n  // when the first genuine error fired — abort them so their emitters finalize\n  // (hooks fire, store records a terminal status) instead of leaking.\n  for (const [runId, emitter] of openRuns) {\n    await finalizeAbort(emitter, runId, 'campaign aborted: genuine error in a sibling run')\n  }\n  openRuns.clear()\n\n  if (genuineErrors.length > 0) {\n    throw genuineErrors.length === 1\n      ? genuineErrors[0]\n      : new AggregateError(\n          genuineErrors,\n          `runEvalCampaign: ${genuineErrors.length} runs failed with genuine (non-run-level) errors`,\n        )\n  }\n\n  // ── Optional research report ───────────────────────────────────────\n  let report: ResearchReport | undefined\n  if (opts.report) {\n    const reportOpts: ResearchReportOptions = {\n      ...opts.report,\n      comparator: opts.report.comparator,\n      split: splitTag === 'dev' ? 'search' : splitTag,\n      generatedAt: new Date(now()).toISOString(),\n      preregistrationHash: preregistrationHash ?? undefined,\n    }\n    report = await researchReport(runs, reportOpts)\n  }\n\n  const endedAt = new Date(now()).toISOString()\n\n  return {\n    campaignId: opts.campaignId,\n    campaignFingerprint,\n    preregistrationHash,\n    runs,\n    integrityReports,\n    failedRuns,\n    report,\n    startedAt,\n    endedAt,\n  }\n}\n\n// ── Internal ─────────────────────────────────────────────────────────────\n\nclass CellExecutionError extends Error {\n  readonly failed: FailedRun\n  readonly integrity?: RunIntegrityReport\n  constructor(failed: FailedRun, integrity?: RunIntegrityReport) {\n    super(`cell ${failed.variantId}/${failed.scenarioId}@${failed.seed} failed: ${failed.reason}`)\n    this.failed = failed\n    this.integrity = integrity\n  }\n}\n\n/**\n * Abort a run whose owning work threw or was orphaned by a sibling's genuine\n * error, finalizing the emitter so hooks fire and the store records a terminal\n * status. Safe to call unconditionally: it only aborts runs that are still\n * `running`. Two no-op cases are intentional and benign:\n *\n *   - the run was never started (absent from the store) — nothing to finalize\n *   - the run is already terminal (`completed` / `failed` / `aborted`) —\n *     finalizing again would overwrite the real outcome (e.g. flip a passed run\n *     to `aborted` with `{ pass: false, notes: reason }`), so we leave it alone\n *\n * Any OTHER failure of the store read or the abort write (disk full, FS fault,\n * backend down) is a genuine diagnostic and propagates rather than being\n * swallowed.\n */\nexport async function finalizeAbort(\n  emitter: TraceEmitter,\n  runId: string,\n  reason: string,\n): Promise<void> {\n  const existing = await emitter.traceStore.getRun(runId)\n  if (existing === undefined) return // run never started; nothing to abort\n  if (existing.status !== 'running') return // already finalized; never overwrite a real outcome\n  await emitter.abortRun(reason)\n}\n\nfunction defaultRawSinkFactory(workDir: string | undefined) {\n  return (params: CampaignFactoryParams): RawProviderSink => {\n    if (!workDir) {\n      throw new Error(\n        'runEvalCampaign: rawSinkFactory not supplied and workDir not set. Pass either to enable raw provider capture, or pass `new NoopRawProviderSink()` via rawSinkFactory to opt out explicitly.',\n      )\n    }\n    return new FileSystemRawProviderSink({\n      dir: `${workDir}/raw-events/${params.runId}`,\n    })\n  }\n}\n\nasync function resolveAgentProfileCell(\n  input: AgentProfileCell | AgentProfileCellInput,\n): Promise<AgentProfileCell> {\n  if (isAgentProfileCell(input)) {\n    if (!(await verifyAgentProfileCell(input))) {\n      throw new Error(`runEvalCampaign: agentProfile.cellId does not match its content`)\n    }\n    return input\n  }\n  return buildAgentProfileCell(input)\n}\n\nfunction isAgentProfileCell(\n  input: AgentProfileCell | AgentProfileCellInput,\n): input is AgentProfileCell {\n  return 'schemaVersion' in input && 'cellId' in input\n}\n\nfunction assertAgentProfileMatchesRun(\n  profile: AgentProfileCell,\n  model: string,\n  promptHash: string,\n): void {\n  if (profile.model !== undefined && profile.model !== model) {\n    throw new Error(\n      `runEvalCampaign: agentProfile.model \"${profile.model}\" does not match outcome.model \"${model}\"`,\n    )\n  }\n  if (profile.promptHash !== undefined && profile.promptHash !== promptHash) {\n    throw new Error(\n      `runEvalCampaign: agentProfile.promptHash \"${profile.promptHash}\" does not match outcome.promptHash \"${promptHash}\"`,\n    )\n  }\n}\n\nfunction defaultRunId(params: CampaignFactoryParams): string {\n  // Stable across re-runs: fingerprint of (campaignId, variantId, scenarioId, seed).\n  // Caller can override via opts.runId for non-deterministic IDs.\n  const base = `${params.campaignId}::${params.variantId}::${params.scenarioId}::${params.seed}`\n  // Lightweight hex: we don't need crypto-grade here, just stability + uniqueness.\n  let h1 = 0x811c9dc5\n  let h2 = 0x12345678\n  for (let i = 0; i < base.length; i++) {\n    const c = base.charCodeAt(i)\n    h1 = Math.imul(h1 ^ c, 0x01000193) >>> 0\n    h2 = Math.imul(h2 ^ c, 0x9e3779b1) >>> 0\n  }\n  return `run-${h1.toString(16).padStart(8, '0')}${h2.toString(16).padStart(8, '0')}`\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAmSA,MAAM,oBAA8C;CAClD,aAAa;CACb,8BAA8B;CAC9B,gBAAgB;AAClB;AAEA,MAAM,gBAAsC;CAC1C,wBAAwB;CACxB,aAAa;AACf;AAEA,eAAsB,gBACpB,MAC6B;CAE7B,eAAe,KAAK,SAAS,KAAK,qBAAqB,aAAa;CAEpE,IAAI,KAAK,SAAS,WAAW,GAC3B,MAAM,IAAI,MAAM,8CAA8C;CAEhE,IAAI,KAAK,UAAU,WAAW,GAC5B,MAAM,IAAI,MAAM,+CAA+C;CAEjE,MAAM,6BAAa,IAAI,IAAY;CACnC,KAAK,MAAM,KAAK,KAAK,UAAU;EAC7B,IAAI,WAAW,IAAI,EAAE,EAAE,GACrB,MAAM,IAAI,MAAM,0CAA0C,EAAE,GAAG,GAAG;EAEpE,WAAW,IAAI,EAAE,EAAE;CACrB;CACA,MAAM,8BAAc,IAAI,IAAY;CACpC,KAAK,MAAM,KAAK,KAAK,WAAW;EAC9B,IAAI,YAAY,IAAI,EAAE,UAAU,GAC9B,MAAM,IAAI,MAAM,0CAA0C,EAAE,WAAW,GAAG;EAE5E,YAAY,IAAI,EAAE,UAAU;CAC9B;CACA,IAAI,KAAK,QAAQ,cAAc,CAAC,WAAW,IAAI,KAAK,OAAO,UAAU,GACnE,MAAM,IAAI,MACR,uCAAuC,KAAK,OAAO,WAAW,iCAChE;CAEF,IAAI,CAAC,KAAK,WACR,MAAM,IAAI,MAAM,oEAAoE;CAGtF,MAAM,QAAQ,KAAK,SAAS;EAAC;EAAG;EAAG;CAAC;CACpC,MAAM,WAAwB,KAAK,YAAY;CAC/C,MAAM,cAAc,KAAK,IAAI,GAAG,KAAK,eAAe,CAAC;CACrD,MAAM,YAAY;EAAE,GAAG;EAAmB,GAAI,KAAK,aAAa,CAAC;CAAG;CACpE,MAAM,qBAA8C,KAAK,sBAAsB;CAC/E,MAAM,MAAM,KAAK,cAAc,KAAK,IAAI;CACxC,MAAM,WAAW,KAAK,QAAQ,WAAW,GAAA,CAAI,QAAQ,QAAQ,EAAE;CAC/D,MAAM,WAAW,KAAK,QAAQ,YAAY;CAC1C,MAAM,sBAAsB,KAAK,uBAAuB;CAExD,MAAM,iBAAiB,KAAK,kBAAkB,sBAAsB,KAAK,OAAO;CAGhF,MAAM,sBAAsB,MAAM,SAChC,aAAa;EACX,YAAY,KAAK;EACjB,UAAU,KAAK,SAAS,KAAK,MAAM,EAAE,EAAE,CAAC,CAAC,KAAK;EAC9C,WAAW,KAAK,UAAU,KAAK,MAAM,EAAE,UAAU,CAAC,CAAC,KAAK;EACxD,OAAO,CAAC,GAAG,KAAK,CAAC,CAAC,MAAM,GAAG,MAAM,IAAI,CAAC;EACtC;EACA,YAAY,KAAK,QAAQ,cAAc;EACvC;EACA;EACA;CACF,CAAC,CACH;CAIA,MAAM,QAAgB,CAAC;CACvB,KAAK,MAAM,WAAW,KAAK,UACzB,KAAK,MAAM,YAAY,KAAK,WAC1B,KAAK,MAAM,QAAQ,OACjB,MAAM,KAAK;EAAE;EAAS;EAAU;CAAK,CAAC;CAK5C,MAAM,YAAY,IAAI,KAAK,IAAI,CAAC,CAAC,CAAC,YAAY;CAC9C,MAAM,OAAoB,CAAC;CAC3B,MAAM,mBAAyC,CAAC;CAChD,MAAM,aAA0B,CAAC;CASjC,IAAI,SAAS;CACb,IAAI,WAAW;CACf,MAAM,gBAA2B,CAAC;CAKlC,MAAM,2BAAW,IAAI,IAA0B;CAE/C,eAAe,SAAwB;EACrC,OAAO,CAAC,UAAU;GAChB,MAAM,IAAI;GACV,IAAI,KAAK,MAAM,QAAQ;GACvB,MAAM,OAAO,MAAM;GACnB,IAAI;IACF,MAAM,SAAS,MAAM,WAAW,IAAI;IACpC,KAAK,KAAK,OAAO,MAAM;IACvB,iBAAiB,KAAK,OAAO,SAAS;GACxC,SAAS,KAAK;IACZ,IAAI,eAAe,oBAAoB;KACrC,WAAW,KAAK,IAAI,MAAM;KAC1B,IAAI,IAAI,WAAW,iBAAiB,KAAK,IAAI,SAAS;IACxD,OAAO;KAKL,cAAc,KAAK,GAAG;KACtB,WAAW;KACX;IACF;GACF;EACF;CACF;CAEA,eAAe,WACb,MAC+D;EAC/D,MAAM,SAAS,KAAK,SAAS,aAAA,CAAc;GACzC,YAAY,KAAK;GACjB,OAAO;GACP,WAAW,KAAK,QAAQ;GACxB,YAAY,KAAK,SAAS;GAC1B,MAAM,KAAK;EACb,CAAC;EACD,MAAM,gBAAuC;GAC3C,YAAY,KAAK;GACjB;GACA,WAAW,KAAK,QAAQ;GACxB,YAAY,KAAK,SAAS;GAC1B,MAAM,KAAK;EACb;EACA,MAAM,QAAQ,KAAK,aAAa,aAAa;EAC7C,MAAM,UAAU,eAAe,aAAa;EAE5C,MAAM,UAAU,IAAI,aAAa,OAAO;GACtC;GACA,KAAK,KAAK;GACV,eAAe,KAAK;EACtB,CAAC;EAGD,SAAS,IAAI,OAAO,OAAO;EAE3B,MAAM,UAA4B;GAChC,GAAG,KAAK;GACR;GACA,cAAc,EAAE,MAAM;EACxB;EAEA,MAAM,MAA6B;GACjC;GACA,cAAc,KAAK;GACnB,SAAS,KAAK,QAAQ;GACtB,WAAW,KAAK,QAAQ;GACxB,YAAY,KAAK,SAAS;GAC1B,cAAc,KAAK,SAAS,QAAQ,CAAC;GACrC,MAAM,KAAK;GACX;GACA;GACA;GACA;GACA;EACF;EAEA,IAAI;GACF,MAAM,YAAY,IAAI;GACtB,IAAI;GACJ,IAAI;IACF,UAAU,MAAM,KAAK,OAAO,GAAG;GACjC,SAAS,KAAK;IACZ,MAAM,UAAU,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;IAM/D,MAAM,cAAc,SAAS,OAAO,OAAO;IAC3C,MAAM,IAAI,mBAAmB;KAC3B;KACA,WAAW,KAAK,QAAQ;KACxB,YAAY,KAAK,SAAS;KAC1B,MAAM,KAAK;KACX,QAAQ;KACR,OAAO;IACT,CAAC;GACH;GACA,MAAM,SAAS,IAAI,IAAI;GAEvB,MAAM,kBAAkB,MAAM,kBAAkB,OAAO,OAAO;IAAE,GAAG;IAAW;GAAQ,CAAC;GACvF,IAAI,CAAC,gBAAgB,IACnB,QAAQ,oBAAR;IACE,KAAK,SACH,MAAM,IAAI,kBAAkB,eAAe;IAC7C,KAAK,eACH,MAAM,IAAI,mBACR;KACE;KACA,WAAW,KAAK,QAAQ;KACxB,YAAY,KAAK,SAAS;KAC1B,MAAM,KAAK;KACX,QAAQ;KACR,OAAO,gBAAgB,OAAO,KAAK,MAAM,EAAE,IAAI,CAAC,CAAC,KAAK,IAAI;IAC5D,GACA,eACF;IACF,KAAK,OAEH;GACJ;GAGF,MAAM,gBAA4B,EAChC,KAAK,QAAQ,OAAO,CAAC,EACvB;GACA,IAAI,aAAa,WAAW,cAAc,eAAe,QAAQ;QAC5D,cAAc,cAAc,QAAQ;GACzC,IAAI,QAAQ,gBAAgB,KAAA,GAAW,cAAc,cAAc,QAAQ;GAE3E,MAAM,SAAoB;IACxB;IACA,cAAc,KAAK;IACnB,aAAa,KAAK,QAAQ;IAC1B,MAAM,KAAK;IACX,OAAO,QAAQ;IACf,YAAY,QAAQ;IACpB,YAAY,QAAQ;IACpB,WAAW,KAAK;IAChB;IACA,SAAS,QAAQ;IACjB,gBAAgB,QAAQ;IACxB,YAAY,QAAQ;IACpB,iBAAiB;IACjB,eAAe,QAAQ;IACvB,SAAS;IACT,GAAI,QAAQ,eAAe,EAAE,cAAc,QAAQ,aAAa,IAAI,CAAC;IACrE,aAAa,QAAQ;IACrB;IACA,YAAY,KAAK,SAAS;GAC5B;GACA,MAAM,gBACJ,QAAQ,iBACP,OAAO,KAAK,iBAAiB,aAC1B,MAAM,KAAK,aAAa;IACtB,YAAY,KAAK;IACjB;IACA,WAAW,KAAK,QAAQ;IACxB,YAAY,KAAK,SAAS;IAC1B,MAAM,KAAK;IACX,SAAS,KAAK,QAAQ;IACtB,cAAc,KAAK,SAAS,QAAQ,CAAC;GACvC,CAAC,IACD,KAAK;GACX,IAAI,kBAAkB,KAAA,GAAW;IAC/B,MAAM,eAAe,MAAM,wBAAwB,aAAa;IAChE,6BAA6B,cAAc,QAAQ,OAAO,QAAQ,UAAU;IAC5E,OAAO,eAAe;GACxB;GACA,OAAO;IAAE,QAAQ,kBAAkB,MAAM;IAAG,WAAW;GAAgB;EACzE,UAAU;GAIR,SAAS,OAAO,KAAK;EACvB;CACF;CAEA,MAAM,UAAU,MAAM,KAAK,EAAE,QAAQ,KAAK,IAAI,aAAa,MAAM,MAAM,EAAE,SAAS,OAAO,CAAC;CAI1F,MAAM,QAAQ,WAAW,OAAO;CAMhC,KAAK,MAAM,CAAC,OAAO,YAAY,UAC7B,MAAM,cAAc,SAAS,OAAO,kDAAkD;CAExF,SAAS,MAAM;CAEf,IAAI,cAAc,SAAS,GACzB,MAAM,cAAc,WAAW,IAC3B,cAAc,KACd,IAAI,eACF,eACA,oBAAoB,cAAc,OAAO,iDAC3C;CAIN,IAAI;CACJ,IAAI,KAAK,QAQP,SAAS,MAAM,eAAe,MAAM;EANlC,GAAG,KAAK;EACR,YAAY,KAAK,OAAO;EACxB,OAAO,aAAa,QAAQ,WAAW;EACvC,aAAa,IAAI,KAAK,IAAI,CAAC,CAAC,CAAC,YAAY;EACzC,qBAAqB,uBAAuB,KAAA;CAED,CAAC;CAGhD,MAAM,UAAU,IAAI,KAAK,IAAI,CAAC,CAAC,CAAC,YAAY;CAE5C,OAAO;EACL,YAAY,KAAK;EACjB;EACA;EACA;EACA;EACA;EACA;EACA;EACA;CACF;AACF;AAIA,IAAM,qBAAN,cAAiC,MAAM;CACrC;CACA;CACA,YAAY,QAAmB,WAAgC;EAC7D,MAAM,QAAQ,OAAO,UAAU,GAAG,OAAO,WAAW,GAAG,OAAO,KAAK,WAAW,OAAO,QAAQ;EAC7F,KAAK,SAAS;EACd,KAAK,YAAY;CACnB;AACF;;;;;;;;;;;;;;;;AAiBA,eAAsB,cACpB,SACA,OACA,QACe;CACf,MAAM,WAAW,MAAM,QAAQ,WAAW,OAAO,KAAK;CACtD,IAAI,aAAa,KAAA,GAAW;CAC5B,IAAI,SAAS,WAAW,WAAW;CACnC,MAAM,QAAQ,SAAS,MAAM;AAC/B;AAEA,SAAS,sBAAsB,SAA6B;CAC1D,QAAQ,WAAmD;EACzD,IAAI,CAAC,SACH,MAAM,IAAI,MACR,6LACF;EAEF,OAAO,IAAI,0BAA0B,EACnC,KAAK,GAAG,QAAQ,cAAc,OAAO,QACvC,CAAC;CACH;AACF;AAEA,eAAe,wBACb,OAC2B;CAC3B,IAAI,mBAAmB,KAAK,GAAG;EAC7B,IAAI,CAAE,MAAM,uBAAuB,KAAK,GACtC,MAAM,IAAI,MAAM,iEAAiE;EAEnF,OAAO;CACT;CACA,OAAO,sBAAsB,KAAK;AACpC;AAEA,SAAS,mBACP,OAC2B;CAC3B,OAAO,mBAAmB,SAAS,YAAY;AACjD;AAEA,SAAS,6BACP,SACA,OACA,YACM;CACN,IAAI,QAAQ,UAAU,KAAA,KAAa,QAAQ,UAAU,OACnD,MAAM,IAAI,MACR,wCAAwC,QAAQ,MAAM,kCAAkC,MAAM,EAChG;CAEF,IAAI,QAAQ,eAAe,KAAA,KAAa,QAAQ,eAAe,YAC7D,MAAM,IAAI,MACR,6CAA6C,QAAQ,WAAW,uCAAuC,WAAW,EACpH;AAEJ;AAEA,SAAS,aAAa,QAAuC;CAG3D,MAAM,OAAO,GAAG,OAAO,WAAW,IAAI,OAAO,UAAU,IAAI,OAAO,WAAW,IAAI,OAAO;CAExF,IAAI,KAAK;CACT,IAAI,KAAK;CACT,KAAK,IAAI,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK;EACpC,MAAM,IAAI,KAAK,WAAW,CAAC;EAC3B,KAAK,KAAK,KAAK,KAAK,GAAG,QAAU,MAAM;EACvC,KAAK,KAAK,KAAK,KAAK,GAAG,UAAU,MAAM;CACzC;CACA,OAAO,OAAO,GAAG,SAAS,EAAE,CAAC,CAAC,SAAS,GAAG,GAAG,IAAI,GAAG,SAAS,EAAE,CAAC,CAAC,SAAS,GAAG,GAAG;AAClF"}
+{"version":3,"file":"eval-campaign-CHqfLnff.js","names":[],"sources":["../src/eval-campaign.ts"],"sourcesContent":["/**\n * EvalCampaign — opinionated matrix runner that wires the four\n * capture-integrity directives by construction.\n *\n * The canonical benchmark shape — matrix runner → for each\n * (variant, scenario, seed) → start a TraceEmitter → call LLMs → end the\n * run → analyze — has a bug class at the integration boundary: raw\n * events not captured, route silently wrong, integrity not asserted,\n * analyst never run. The directives in `SKILL.md § Capture integrity`\n * are the mitigations.\n *\n * `EvalCampaign` is the structural fix — consumers don't wire the\n * integrity surface themselves; the campaign owns it. Specifically:\n *\n *   - calls `assertLlmRoute` once at preflight before any work runs\n *   - constructs a per-run `TraceStore` and `RawProviderSink` via factories\n *   - constructs the `TraceEmitter` with `onRunComplete: [analyst hook]`\n *   - hands the runner an `LlmClientOptions` pre-wired with the sink and\n *     trace context — the runner can't accidentally call an LLM without\n *     capturing the raw HTTP envelope\n *   - calls `assertRunCaptured` after every `endRun` and routes failures\n *     through a configurable policy (`throw` / `mark_failed` / `log`)\n *   - assembles per-run `RunRecord`s and runs `researchReport` at the end\n *     so the campaign artifact is launch-decision-grade by default\n *   - embeds the campaign fingerprint (a SHA-256 over the canonicalised\n *     run set) and optional `preregistrationHash` in the report\n *\n * The runner contract is intentionally narrow: produce a `CampaignRunOutcome`\n * given a fully-wired `CampaignRunContext`. Everything orchestration-shaped\n * lives in the campaign. This is the inversion-of-control point — consumers\n * stop writing matrix runners and start writing scenario-runners.\n *\n * Out of scope for v1 (tracked in `docs/research-report-methodology.md`):\n *\n *   - Distributed/cluster execution (concurrency is local async)\n *   - Adaptive sampling / sequential interim looks\n *   - Resume from partial state across crashes\n *   - LLM-call retry beyond what `LlmClient` already does\n */\n\nimport {\n  type AgentProfileCell,\n  type AgentProfileCellInput,\n  buildAgentProfileCell,\n  verifyAgentProfileCell,\n} from './agent-profile-cell'\nimport { assertLlmRoute, type LlmClientOptions, type LlmRouteRequirements } from './llm-client'\nimport { canonicalize, hashJson } from './pre-registration'\nimport type {\n  JudgeScoresRecord,\n  RunCostProvenance,\n  RunJudgeMetadata,\n  RunOutcome,\n  RunRecord,\n  RunSplitTag,\n  RunTaskFailure,\n  RunTokenUsage,\n} from './run-record'\nimport { validateRunRecord } from './run-record'\nimport { type ResearchReport, type ResearchReportOptions, researchReport } from './summary-report'\nimport type { RunCompleteHook } from './trace/emitter'\nimport { TraceEmitter } from './trace/emitter'\nimport {\n  assertRunCaptured,\n  RunIntegrityError,\n  type RunIntegrityExpectations,\n  type RunIntegrityReport,\n} from './trace/integrity'\nimport { FileSystemRawProviderSink, type RawProviderSink } from './trace/raw-provider-sink'\nimport type { TraceStore } from './trace/store'\n\n// ── Public types ─────────────────────────────────────────────────────────\n\nexport interface CampaignVariant<V> {\n  id: string\n  payload: V\n}\n\nexport interface CampaignScenario {\n  scenarioId: string\n  /** Free-form metadata propagated to runs and reports. */\n  tags?: Record<string, string>\n}\n\nexport interface CampaignRunContext<V> {\n  /** Stable run id. The campaign generates this; the runner does not. */\n  runId: string\n  /** Logical experiment id (campaignId by default; overridable per-run via opts). */\n  experimentId: string\n  variant: V\n  variantId: string\n  scenarioId: string\n  scenarioTags: Record<string, string>\n  seed: number\n  splitTag: RunSplitTag\n  /**\n   * The TraceEmitter for this run, with `onRunComplete` hooks pre-wired\n   * (analyst auto-execution if configured, plus integrity check). The\n   * runner MUST call `emitter.startRun` before doing any work and either\n   * `emitter.endRun` or `emitter.abortRun` before returning.\n   */\n  emitter: TraceEmitter\n  store: TraceStore\n  rawSink: RawProviderSink\n  /**\n   * Pre-wired LLM client options — `rawSink` and `traceContext` are populated\n   * so any `callLlm(req, ctx.llmOpts)` automatically captures raw HTTP. The\n   * runner can spread additional fields if needed.\n   */\n  llmOpts: LlmClientOptions\n}\n\ninterface CampaignRunOutcomeFields {\n  /** Did the run pass? Mirrors `RunOutcome.pass` semantics. */\n  pass: boolean\n  /** Score for the run on its split. Maps to `searchScore` or `holdoutScore`. */\n  score: number\n  /** Cost in USD, or null when the runner could not capture it. */\n  costUsd: number | null\n  /** Source of the cost amount. */\n  costProvenance: RunCostProvenance\n  tokenUsage: RunTokenUsage\n  /** Snapshot model id (e.g. `claude-sonnet-4-6@2025-04-15`). */\n  model: string\n  /** sha256 of the effective prompt sent to the model. */\n  promptHash: string\n  /** sha256 of the effective config (model, temperature, tools, judges, splits). */\n  configHash: string\n  /** Optional extra numeric metrics to land in `outcome.raw`. */\n  raw?: Record<string, number>\n  /** Optional judge metadata when a judge was used. */\n  judgeMetadata?: RunJudgeMetadata\n  /**\n   * Optional per-judge / per-dim breakdown for ensemble-judged runs.\n   * Propagated to `outcome.judgeScores` on the resulting `RunRecord`.\n   * Single-judge or scalar-only runs leave this unset.\n   */\n  judgeScores?: JudgeScoresRecord\n  /**\n   * Agent profile cell observed by the runner. When supplied, it overrides\n   * `EvalCampaignOptions.agentProfile` for this run and must match the\n   * outcome's `model` and `promptHash`.\n   */\n  agentProfile?: AgentProfileCell | AgentProfileCellInput\n}\n\n/** Campaign result with the same task-failure invariant as `RunRecord`. */\nexport type CampaignRunOutcome = CampaignRunOutcomeFields & RunTaskFailure\n\nexport type CampaignRunner<V> = (ctx: CampaignRunContext<V>) => Promise<CampaignRunOutcome>\n\nexport type CampaignIntegrityPolicy = 'throw' | 'mark_failed' | 'log'\n\nexport interface EvalCampaignOptions<V> {\n  /**\n   * Stable id for the campaign. Used as the default `experimentId` on\n   * every run, and folded into the campaign fingerprint.\n   */\n  campaignId: string\n  variants: CampaignVariant<V>[]\n  scenarios: CampaignScenario[]\n  /** Default `[0, 1, 2]`. */\n  seeds?: number[]\n  /** Default `'holdout'` — the split that anchors a launch decision. */\n  splitTag?: RunSplitTag\n  /** Git SHA the campaign is run against. Mandatory; `RunRecord` rejects unset. */\n  commitSha: string\n  /**\n   * LLM client config. Augmented per-run with `rawSink` and `traceContext`\n   * before being passed to the runner. The campaign asserts this config\n   * matches `routeRequirements` once at preflight.\n   */\n  llmOpts: LlmClientOptions\n  /**\n   * Default `{ requireExplicitBaseUrl: true, requireAuth: true }` — fail\n   * loud if the campaign would silently fall back to the public router or\n   * run unauthenticated. Override with an empty object to disable.\n   */\n  routeRequirements?: LlmRouteRequirements\n  /**\n   * Per-run TraceStore factory. Common shape: a fresh store per run keyed\n   * on `runId`. Implementations that share a store across the campaign\n   * are valid — the campaign only writes through `emitter`.\n   */\n  storeFactory: (params: CampaignFactoryParams) => TraceStore\n  /**\n   * Per-run RawProviderSink factory. Defaults to `FileSystemRawProviderSink`\n   * rooted at `${workDir}/raw-events/${runId}` if `workDir` is supplied;\n   * otherwise required. Forensic capture is non-negotiable in a campaign\n   * run — pass `NoopRawProviderSink` explicitly if you want to opt out.\n   */\n  rawSinkFactory?: (params: CampaignFactoryParams) => RawProviderSink\n  /**\n   * Filesystem root for default `rawSinkFactory`. Ignored if\n   * `rawSinkFactory` is supplied.\n   */\n  workDir?: string\n  /**\n   * Extra `onRunComplete` hooks the campaign appends (after its own\n   * integrity-check hook). Pass `traceAnalystOnRunComplete(...)` here.\n   */\n  onRunComplete?: RunCompleteHook[]\n  /**\n   * Per-run integrity expectations. Defaults to:\n   *   `{ llmSpansMin: 1, requireRawCoverageOfLlmSpans: true, requireOutcome: true }`.\n   * Override (e.g. `{ llmSpansMin: 0 }`) for runs that don't call LLMs.\n   */\n  integrity?: RunIntegrityExpectations\n  /** Behaviour when integrity fails. Default `'mark_failed'`. */\n  onIntegrityFailure?: CampaignIntegrityPolicy\n  /**\n   * Per-run runner. Receives a fully-wired context; produces an outcome\n   * the campaign converts into a `RunRecord`.\n   */\n  runner: CampaignRunner<V>\n  /**\n   * If set, the campaign computes `researchReport` at the end. `comparator`\n   * is a `variantId`. Other fields are forwarded verbatim.\n   */\n  report?: { comparator?: string } & Omit<\n    ResearchReportOptions,\n    'comparator' | 'preregistrationHash' | 'generatedAt'\n  >\n  /**\n   * Hash of a signed `HypothesisManifest` (see `pre-registration.ts`).\n   * Embedded in the campaign fingerprint and the research report.\n   */\n  preregistrationHash?: string\n  /** Local concurrency. Default `1` (sequential). */\n  concurrency?: number\n  /**\n   * Override the time source. Tests pass a mock to make wallMs deterministic.\n   */\n  now?: () => number\n  /** Override the runId generator. Tests pin this. */\n  runId?: (params: CampaignFactoryParams) => string\n  /**\n   * Agent profile cell for campaign runs. Static profiles can pass an object;\n   * routers or variant-specific harnesses can pass a factory. The campaign\n   * stamps the built cell onto every `RunRecord` and rejects profile/model or\n   * profile/prompt contradictions.\n   */\n  agentProfile?:\n    | AgentProfileCell\n    | AgentProfileCellInput\n    | ((\n        params: CampaignFactoryParams & {\n          variant: V\n          scenarioTags: Record<string, string>\n        },\n      ) =>\n        | AgentProfileCell\n        | AgentProfileCellInput\n        | Promise<AgentProfileCell | AgentProfileCellInput>)\n}\n\nexport interface CampaignFactoryParams {\n  campaignId: string\n  runId: string\n  variantId: string\n  scenarioId: string\n  seed: number\n}\n\nexport interface FailedRun {\n  runId: string\n  variantId: string\n  scenarioId: string\n  seed: number\n  reason: string\n  error?: string\n}\n\nexport interface EvalCampaignResult {\n  campaignId: string\n  /** SHA-256 over canonicalised `(variantIds, scenarioIds, seeds, comparator, splitTag, baseUrl, provider, preregistrationHash)`. */\n  campaignFingerprint: string\n  preregistrationHash: string | null\n  /** Successful runs only. Failed runs land in `failedRuns`. */\n  runs: RunRecord[]\n  /** Integrity reports for every successful run. */\n  integrityReports: RunIntegrityReport[]\n  failedRuns: FailedRun[]\n  /** Computed when `report` is set on options. */\n  report?: ResearchReport\n  startedAt: string\n  endedAt: string\n}\n\n// ── Implementation ───────────────────────────────────────────────────────\n\nconst DEFAULT_INTEGRITY: RunIntegrityExpectations = {\n  llmSpansMin: 1,\n  requireRawCoverageOfLlmSpans: true,\n  requireOutcome: true,\n}\n\nconst DEFAULT_ROUTE: LlmRouteRequirements = {\n  requireExplicitBaseUrl: true,\n  requireAuth: true,\n}\n\nexport async function runEvalCampaign<V>(\n  opts: EvalCampaignOptions<V>,\n): Promise<EvalCampaignResult> {\n  // ── Preflight ──────────────────────────────────────────────────────\n  assertLlmRoute(opts.llmOpts, opts.routeRequirements ?? DEFAULT_ROUTE)\n\n  if (opts.variants.length === 0) {\n    throw new Error('runEvalCampaign: variants must be non-empty.')\n  }\n  if (opts.scenarios.length === 0) {\n    throw new Error('runEvalCampaign: scenarios must be non-empty.')\n  }\n  const variantIds = new Set<string>()\n  for (const v of opts.variants) {\n    if (variantIds.has(v.id)) {\n      throw new Error(`runEvalCampaign: duplicate variant id \"${v.id}\".`)\n    }\n    variantIds.add(v.id)\n  }\n  const scenarioIds = new Set<string>()\n  for (const s of opts.scenarios) {\n    if (scenarioIds.has(s.scenarioId)) {\n      throw new Error(`runEvalCampaign: duplicate scenarioId \"${s.scenarioId}\".`)\n    }\n    scenarioIds.add(s.scenarioId)\n  }\n  if (opts.report?.comparator && !variantIds.has(opts.report.comparator)) {\n    throw new Error(\n      `runEvalCampaign: report.comparator \"${opts.report.comparator}\" is not a configured variantId.`,\n    )\n  }\n  if (!opts.commitSha) {\n    throw new Error('runEvalCampaign: commitSha is required (every RunRecord needs it).')\n  }\n\n  const seeds = opts.seeds ?? [0, 1, 2]\n  const splitTag: RunSplitTag = opts.splitTag ?? 'holdout'\n  const concurrency = Math.max(1, opts.concurrency ?? 1)\n  const integrity = { ...DEFAULT_INTEGRITY, ...(opts.integrity ?? {}) }\n  const onIntegrityFailure: CampaignIntegrityPolicy = opts.onIntegrityFailure ?? 'mark_failed'\n  const now = opts.now ?? (() => Date.now())\n  const baseUrl = (opts.llmOpts.baseUrl ?? '').replace(/\\/+$/, '')\n  const provider = opts.llmOpts.provider ?? null\n  const preregistrationHash = opts.preregistrationHash ?? null\n\n  const rawSinkFactory = opts.rawSinkFactory ?? defaultRawSinkFactory(opts.workDir)\n\n  // ── Fingerprint ────────────────────────────────────────────────────\n  const campaignFingerprint = await hashJson(\n    canonicalize({\n      campaignId: opts.campaignId,\n      variants: opts.variants.map((v) => v.id).sort(),\n      scenarios: opts.scenarios.map((s) => s.scenarioId).sort(),\n      seeds: [...seeds].sort((a, b) => a - b),\n      splitTag,\n      comparator: opts.report?.comparator ?? null,\n      baseUrl,\n      provider,\n      preregistrationHash,\n    }),\n  )\n\n  // ── Plan the matrix ────────────────────────────────────────────────\n  type Cell = { variant: CampaignVariant<V>; scenario: CampaignScenario; seed: number }\n  const cells: Cell[] = []\n  for (const variant of opts.variants) {\n    for (const scenario of opts.scenarios) {\n      for (const seed of seeds) {\n        cells.push({ variant, scenario, seed })\n      }\n    }\n  }\n\n  const startedAt = new Date(now()).toISOString()\n  const runs: RunRecord[] = []\n  const integrityReports: RunIntegrityReport[] = []\n  const failedRuns: FailedRun[] = []\n\n  // ── Execute (bounded-concurrency worker pool) ──────────────────────\n  // A genuine (non-CellExecutionError) error from any worker is a bug, not a\n  // run-level failure. We must NOT let `Promise.all` reject mid-flight and\n  // orphan the other workers' in-progress runs (emitters never finalized,\n  // sinks/handles leak, partial work silently discarded). Instead: capture the\n  // first genuine error, stop dispatching new cells so in-flight workers wind\n  // down, finalize every still-open run, then re-throw the aggregated error.\n  let cursor = 0\n  let aborting = false\n  const genuineErrors: unknown[] = []\n  // Emitters for runs that are currently mid-flight, keyed by runId. A worker\n  // registers its emitter before invoking the runner and removes it once the\n  // run is finalized (via endRun/abortRun, success or failure). Anything left\n  // here after the pool settles is an orphan we must abort.\n  const openRuns = new Map<string, TraceEmitter>()\n\n  async function worker(): Promise<void> {\n    while (!aborting) {\n      const i = cursor++\n      if (i >= cells.length) return\n      const cell = cells[i]!\n      try {\n        const result = await runOneCell(cell)\n        runs.push(result.record)\n        integrityReports.push(result.integrity)\n      } catch (err) {\n        if (err instanceof CellExecutionError) {\n          failedRuns.push(err.failed)\n          if (err.integrity) integrityReports.push(err.integrity)\n        } else {\n          // Genuine bug — not a runner failure, not an integrity failure.\n          // Capture it and stop dispatching so peers wind down gracefully;\n          // re-thrown after the pool settles. Do not surface here — that would\n          // reject Promise.all and orphan the other workers' runs.\n          genuineErrors.push(err)\n          aborting = true\n          return\n        }\n      }\n    }\n  }\n\n  async function runOneCell(\n    cell: Cell,\n  ): Promise<{ record: RunRecord; integrity: RunIntegrityReport }> {\n    const runId = (opts.runId ?? defaultRunId)({\n      campaignId: opts.campaignId,\n      runId: '', // unused by default generator\n      variantId: cell.variant.id,\n      scenarioId: cell.scenario.scenarioId,\n      seed: cell.seed,\n    })\n    const factoryParams: CampaignFactoryParams = {\n      campaignId: opts.campaignId,\n      runId,\n      variantId: cell.variant.id,\n      scenarioId: cell.scenario.scenarioId,\n      seed: cell.seed,\n    }\n    const store = opts.storeFactory(factoryParams)\n    const rawSink = rawSinkFactory(factoryParams)\n\n    const emitter = new TraceEmitter(store, {\n      runId,\n      now: opts.now,\n      onRunComplete: opts.onRunComplete,\n    })\n    // Track this run as open so a genuine error elsewhere in the pool can\n    // finalize it instead of orphaning it. Removed in the finally below.\n    openRuns.set(runId, emitter)\n\n    const llmOpts: LlmClientOptions = {\n      ...opts.llmOpts,\n      rawSink,\n      traceContext: { runId },\n    }\n\n    const ctx: CampaignRunContext<V> = {\n      runId,\n      experimentId: opts.campaignId,\n      variant: cell.variant.payload,\n      variantId: cell.variant.id,\n      scenarioId: cell.scenario.scenarioId,\n      scenarioTags: cell.scenario.tags ?? {},\n      seed: cell.seed,\n      splitTag,\n      emitter,\n      store,\n      rawSink,\n      llmOpts,\n    }\n\n    try {\n      const wallStart = now()\n      let outcome: CampaignRunOutcome\n      try {\n        outcome = await opts.runner(ctx)\n      } catch (err) {\n        const message = err instanceof Error ? err.message : String(err)\n        // The runner threw mid-execution. Abort the run so the emitter\n        // finalizes. The only benign abortRun failure is \"the runner never\n        // started the run\" (nothing to finalize). A store-write failure (disk\n        // full, FS error) is a genuine diagnostic — surface it rather than\n        // masking it as a plain runner failure.\n        await finalizeAbort(emitter, runId, message)\n        throw new CellExecutionError({\n          runId,\n          variantId: cell.variant.id,\n          scenarioId: cell.scenario.scenarioId,\n          seed: cell.seed,\n          reason: 'runner_threw',\n          error: message,\n        })\n      }\n      const wallMs = now() - wallStart\n\n      const integrityReport = await assertRunCaptured(store, runId, { ...integrity, rawSink })\n      if (!integrityReport.ok) {\n        switch (onIntegrityFailure) {\n          case 'throw':\n            throw new RunIntegrityError(integrityReport)\n          case 'mark_failed':\n            throw new CellExecutionError(\n              {\n                runId,\n                variantId: cell.variant.id,\n                scenarioId: cell.scenario.scenarioId,\n                seed: cell.seed,\n                reason: 'integrity_failed',\n                error: integrityReport.issues.map((i) => i.code).join(', '),\n              },\n              integrityReport,\n            )\n          case 'log':\n            // Caller wants the run admitted with a flagged report; fall through.\n            break\n        }\n      }\n\n      const recordOutcome: RunOutcome = {\n        raw: outcome.raw ?? {},\n      }\n      if (splitTag === 'holdout') recordOutcome.holdoutScore = outcome.score\n      else recordOutcome.searchScore = outcome.score\n      if (outcome.judgeScores !== undefined) recordOutcome.judgeScores = outcome.judgeScores\n\n      const record: RunRecord = {\n        runId,\n        experimentId: opts.campaignId,\n        candidateId: cell.variant.id,\n        seed: cell.seed,\n        model: outcome.model,\n        promptHash: outcome.promptHash,\n        configHash: outcome.configHash,\n        commitSha: opts.commitSha,\n        wallMs,\n        costUsd: outcome.costUsd,\n        costProvenance: outcome.costProvenance,\n        tokenUsage: outcome.tokenUsage,\n        terminalOutcome: 'succeeded',\n        judgeMetadata: outcome.judgeMetadata,\n        outcome: recordOutcome,\n        ...(outcome.failureClass ? { failureClass: outcome.failureClass } : {}),\n        failureMode: outcome.failureMode,\n        splitTag,\n        scenarioId: cell.scenario.scenarioId,\n      }\n      const profileSource =\n        outcome.agentProfile ??\n        (typeof opts.agentProfile === 'function'\n          ? await opts.agentProfile({\n              campaignId: opts.campaignId,\n              runId,\n              variantId: cell.variant.id,\n              scenarioId: cell.scenario.scenarioId,\n              seed: cell.seed,\n              variant: cell.variant.payload,\n              scenarioTags: cell.scenario.tags ?? {},\n            })\n          : opts.agentProfile)\n      if (profileSource !== undefined) {\n        const agentProfile = await resolveAgentProfileCell(profileSource)\n        assertAgentProfileMatchesRun(agentProfile, outcome.model, outcome.promptHash)\n        record.agentProfile = agentProfile\n      }\n      return { record: validateRunRecord(record), integrity: integrityReport }\n    } finally {\n      // This run's worker has finished with it (success, run-level failure, or\n      // genuine error). It is no longer the pool's job to finalize — drop it so\n      // the post-settle sweep doesn't double-abort a finalized run.\n      openRuns.delete(runId)\n    }\n  }\n\n  const workers = Array.from({ length: Math.min(concurrency, cells.length) }, () => worker())\n  // allSettled (not all): a genuine error in one worker must not reject the\n  // pool mid-flight and orphan the others. Each worker captures its own\n  // genuine error into `genuineErrors` and returns; we re-throw below.\n  await Promise.allSettled(workers)\n\n  // Finalize any run still open after the pool wound down. With the\n  // stop-dispatch flag these are the runs that were mid-flight in peer workers\n  // when the first genuine error fired — abort them so their emitters finalize\n  // (hooks fire, store records a terminal status) instead of leaking.\n  for (const [runId, emitter] of openRuns) {\n    await finalizeAbort(emitter, runId, 'campaign aborted: genuine error in a sibling run')\n  }\n  openRuns.clear()\n\n  if (genuineErrors.length > 0) {\n    throw genuineErrors.length === 1\n      ? genuineErrors[0]\n      : new AggregateError(\n          genuineErrors,\n          `runEvalCampaign: ${genuineErrors.length} runs failed with genuine (non-run-level) errors`,\n        )\n  }\n\n  // ── Optional research report ───────────────────────────────────────\n  let report: ResearchReport | undefined\n  if (opts.report) {\n    const reportOpts: ResearchReportOptions = {\n      ...opts.report,\n      comparator: opts.report.comparator,\n      split: splitTag === 'dev' ? 'search' : splitTag,\n      generatedAt: new Date(now()).toISOString(),\n      preregistrationHash: preregistrationHash ?? undefined,\n    }\n    report = await researchReport(runs, reportOpts)\n  }\n\n  const endedAt = new Date(now()).toISOString()\n\n  return {\n    campaignId: opts.campaignId,\n    campaignFingerprint,\n    preregistrationHash,\n    runs,\n    integrityReports,\n    failedRuns,\n    report,\n    startedAt,\n    endedAt,\n  }\n}\n\n// ── Internal ─────────────────────────────────────────────────────────────\n\nclass CellExecutionError extends Error {\n  readonly failed: FailedRun\n  readonly integrity?: RunIntegrityReport\n  constructor(failed: FailedRun, integrity?: RunIntegrityReport) {\n    super(`cell ${failed.variantId}/${failed.scenarioId}@${failed.seed} failed: ${failed.reason}`)\n    this.failed = failed\n    this.integrity = integrity\n  }\n}\n\n/**\n * Abort a run whose owning work threw or was orphaned by a sibling's genuine\n * error, finalizing the emitter so hooks fire and the store records a terminal\n * status. Safe to call unconditionally: it only aborts runs that are still\n * `running`. Two no-op cases are intentional and benign:\n *\n *   - the run was never started (absent from the store) — nothing to finalize\n *   - the run is already terminal (`completed` / `failed` / `aborted`) —\n *     finalizing again would overwrite the real outcome (e.g. flip a passed run\n *     to `aborted` with `{ pass: false, notes: reason }`), so we leave it alone\n *\n * Any OTHER failure of the store read or the abort write (disk full, FS fault,\n * backend down) is a genuine diagnostic and propagates rather than being\n * swallowed.\n */\nexport async function finalizeAbort(\n  emitter: TraceEmitter,\n  runId: string,\n  reason: string,\n): Promise<void> {\n  const existing = await emitter.traceStore.getRun(runId)\n  if (existing === undefined) return // run never started; nothing to abort\n  if (existing.status !== 'running') return // already finalized; never overwrite a real outcome\n  await emitter.abortRun(reason)\n}\n\nfunction defaultRawSinkFactory(workDir: string | undefined) {\n  return (params: CampaignFactoryParams): RawProviderSink => {\n    if (!workDir) {\n      throw new Error(\n        'runEvalCampaign: rawSinkFactory not supplied and workDir not set. Pass either to enable raw provider capture, or pass `new NoopRawProviderSink()` via rawSinkFactory to opt out explicitly.',\n      )\n    }\n    return new FileSystemRawProviderSink({\n      dir: `${workDir}/raw-events/${params.runId}`,\n    })\n  }\n}\n\nasync function resolveAgentProfileCell(\n  input: AgentProfileCell | AgentProfileCellInput,\n): Promise<AgentProfileCell> {\n  if (isAgentProfileCell(input)) {\n    if (!(await verifyAgentProfileCell(input))) {\n      throw new Error(`runEvalCampaign: agentProfile.cellId does not match its content`)\n    }\n    return input\n  }\n  return buildAgentProfileCell(input)\n}\n\nfunction isAgentProfileCell(\n  input: AgentProfileCell | AgentProfileCellInput,\n): input is AgentProfileCell {\n  return 'schemaVersion' in input && 'cellId' in input\n}\n\nfunction assertAgentProfileMatchesRun(\n  profile: AgentProfileCell,\n  model: string,\n  promptHash: string,\n): void {\n  if (profile.model !== undefined && profile.model !== model) {\n    throw new Error(\n      `runEvalCampaign: agentProfile.model \"${profile.model}\" does not match outcome.model \"${model}\"`,\n    )\n  }\n  if (profile.promptHash !== undefined && profile.promptHash !== promptHash) {\n    throw new Error(\n      `runEvalCampaign: agentProfile.promptHash \"${profile.promptHash}\" does not match outcome.promptHash \"${promptHash}\"`,\n    )\n  }\n}\n\nfunction defaultRunId(params: CampaignFactoryParams): string {\n  // Stable across re-runs: fingerprint of (campaignId, variantId, scenarioId, seed).\n  // Caller can override via opts.runId for non-deterministic IDs.\n  const base = `${params.campaignId}::${params.variantId}::${params.scenarioId}::${params.seed}`\n  // Lightweight hex: we don't need crypto-grade here, just stability + uniqueness.\n  let h1 = 0x811c9dc5\n  let h2 = 0x12345678\n  for (let i = 0; i < base.length; i++) {\n    const c = base.charCodeAt(i)\n    h1 = Math.imul(h1 ^ c, 0x01000193) >>> 0\n    h2 = Math.imul(h2 ^ c, 0x9e3779b1) >>> 0\n  }\n  return `run-${h1.toString(16).padStart(8, '0')}${h2.toString(16).padStart(8, '0')}`\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAmSA,MAAM,oBAA8C;CAClD,aAAa;CACb,8BAA8B;CAC9B,gBAAgB;AAClB;AAEA,MAAM,gBAAsC;CAC1C,wBAAwB;CACxB,aAAa;AACf;AAEA,eAAsB,gBACpB,MAC6B;CAE7B,eAAe,KAAK,SAAS,KAAK,qBAAqB,aAAa;CAEpE,IAAI,KAAK,SAAS,WAAW,GAC3B,MAAM,IAAI,MAAM,8CAA8C;CAEhE,IAAI,KAAK,UAAU,WAAW,GAC5B,MAAM,IAAI,MAAM,+CAA+C;CAEjE,MAAM,6BAAa,IAAI,IAAY;CACnC,KAAK,MAAM,KAAK,KAAK,UAAU;EAC7B,IAAI,WAAW,IAAI,EAAE,EAAE,GACrB,MAAM,IAAI,MAAM,0CAA0C,EAAE,GAAG,GAAG;EAEpE,WAAW,IAAI,EAAE,EAAE;CACrB;CACA,MAAM,8BAAc,IAAI,IAAY;CACpC,KAAK,MAAM,KAAK,KAAK,WAAW;EAC9B,IAAI,YAAY,IAAI,EAAE,UAAU,GAC9B,MAAM,IAAI,MAAM,0CAA0C,EAAE,WAAW,GAAG;EAE5E,YAAY,IAAI,EAAE,UAAU;CAC9B;CACA,IAAI,KAAK,QAAQ,cAAc,CAAC,WAAW,IAAI,KAAK,OAAO,UAAU,GACnE,MAAM,IAAI,MACR,uCAAuC,KAAK,OAAO,WAAW,iCAChE;CAEF,IAAI,CAAC,KAAK,WACR,MAAM,IAAI,MAAM,oEAAoE;CAGtF,MAAM,QAAQ,KAAK,SAAS;EAAC;EAAG;EAAG;CAAC;CACpC,MAAM,WAAwB,KAAK,YAAY;CAC/C,MAAM,cAAc,KAAK,IAAI,GAAG,KAAK,eAAe,CAAC;CACrD,MAAM,YAAY;EAAE,GAAG;EAAmB,GAAI,KAAK,aAAa,CAAC;CAAG;CACpE,MAAM,qBAA8C,KAAK,sBAAsB;CAC/E,MAAM,MAAM,KAAK,cAAc,KAAK,IAAI;CACxC,MAAM,WAAW,KAAK,QAAQ,WAAW,GAAA,CAAI,QAAQ,QAAQ,EAAE;CAC/D,MAAM,WAAW,KAAK,QAAQ,YAAY;CAC1C,MAAM,sBAAsB,KAAK,uBAAuB;CAExD,MAAM,iBAAiB,KAAK,kBAAkB,sBAAsB,KAAK,OAAO;CAGhF,MAAM,sBAAsB,MAAM,SAChC,aAAa;EACX,YAAY,KAAK;EACjB,UAAU,KAAK,SAAS,KAAK,MAAM,EAAE,EAAE,CAAC,CAAC,KAAK;EAC9C,WAAW,KAAK,UAAU,KAAK,MAAM,EAAE,UAAU,CAAC,CAAC,KAAK;EACxD,OAAO,CAAC,GAAG,KAAK,CAAC,CAAC,MAAM,GAAG,MAAM,IAAI,CAAC;EACtC;EACA,YAAY,KAAK,QAAQ,cAAc;EACvC;EACA;EACA;CACF,CAAC,CACH;CAIA,MAAM,QAAgB,CAAC;CACvB,KAAK,MAAM,WAAW,KAAK,UACzB,KAAK,MAAM,YAAY,KAAK,WAC1B,KAAK,MAAM,QAAQ,OACjB,MAAM,KAAK;EAAE;EAAS;EAAU;CAAK,CAAC;CAK5C,MAAM,YAAY,IAAI,KAAK,IAAI,CAAC,CAAC,CAAC,YAAY;CAC9C,MAAM,OAAoB,CAAC;CAC3B,MAAM,mBAAyC,CAAC;CAChD,MAAM,aAA0B,CAAC;CASjC,IAAI,SAAS;CACb,IAAI,WAAW;CACf,MAAM,gBAA2B,CAAC;CAKlC,MAAM,2BAAW,IAAI,IAA0B;CAE/C,eAAe,SAAwB;EACrC,OAAO,CAAC,UAAU;GAChB,MAAM,IAAI;GACV,IAAI,KAAK,MAAM,QAAQ;GACvB,MAAM,OAAO,MAAM;GACnB,IAAI;IACF,MAAM,SAAS,MAAM,WAAW,IAAI;IACpC,KAAK,KAAK,OAAO,MAAM;IACvB,iBAAiB,KAAK,OAAO,SAAS;GACxC,SAAS,KAAK;IACZ,IAAI,eAAe,oBAAoB;KACrC,WAAW,KAAK,IAAI,MAAM;KAC1B,IAAI,IAAI,WAAW,iBAAiB,KAAK,IAAI,SAAS;IACxD,OAAO;KAKL,cAAc,KAAK,GAAG;KACtB,WAAW;KACX;IACF;GACF;EACF;CACF;CAEA,eAAe,WACb,MAC+D;EAC/D,MAAM,SAAS,KAAK,SAAS,aAAA,CAAc;GACzC,YAAY,KAAK;GACjB,OAAO;GACP,WAAW,KAAK,QAAQ;GACxB,YAAY,KAAK,SAAS;GAC1B,MAAM,KAAK;EACb,CAAC;EACD,MAAM,gBAAuC;GAC3C,YAAY,KAAK;GACjB;GACA,WAAW,KAAK,QAAQ;GACxB,YAAY,KAAK,SAAS;GAC1B,MAAM,KAAK;EACb;EACA,MAAM,QAAQ,KAAK,aAAa,aAAa;EAC7C,MAAM,UAAU,eAAe,aAAa;EAE5C,MAAM,UAAU,IAAI,aAAa,OAAO;GACtC;GACA,KAAK,KAAK;GACV,eAAe,KAAK;EACtB,CAAC;EAGD,SAAS,IAAI,OAAO,OAAO;EAE3B,MAAM,UAA4B;GAChC,GAAG,KAAK;GACR;GACA,cAAc,EAAE,MAAM;EACxB;EAEA,MAAM,MAA6B;GACjC;GACA,cAAc,KAAK;GACnB,SAAS,KAAK,QAAQ;GACtB,WAAW,KAAK,QAAQ;GACxB,YAAY,KAAK,SAAS;GAC1B,cAAc,KAAK,SAAS,QAAQ,CAAC;GACrC,MAAM,KAAK;GACX;GACA;GACA;GACA;GACA;EACF;EAEA,IAAI;GACF,MAAM,YAAY,IAAI;GACtB,IAAI;GACJ,IAAI;IACF,UAAU,MAAM,KAAK,OAAO,GAAG;GACjC,SAAS,KAAK;IACZ,MAAM,UAAU,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;IAM/D,MAAM,cAAc,SAAS,OAAO,OAAO;IAC3C,MAAM,IAAI,mBAAmB;KAC3B;KACA,WAAW,KAAK,QAAQ;KACxB,YAAY,KAAK,SAAS;KAC1B,MAAM,KAAK;KACX,QAAQ;KACR,OAAO;IACT,CAAC;GACH;GACA,MAAM,SAAS,IAAI,IAAI;GAEvB,MAAM,kBAAkB,MAAM,kBAAkB,OAAO,OAAO;IAAE,GAAG;IAAW;GAAQ,CAAC;GACvF,IAAI,CAAC,gBAAgB,IACnB,QAAQ,oBAAR;IACE,KAAK,SACH,MAAM,IAAI,kBAAkB,eAAe;IAC7C,KAAK,eACH,MAAM,IAAI,mBACR;KACE;KACA,WAAW,KAAK,QAAQ;KACxB,YAAY,KAAK,SAAS;KAC1B,MAAM,KAAK;KACX,QAAQ;KACR,OAAO,gBAAgB,OAAO,KAAK,MAAM,EAAE,IAAI,CAAC,CAAC,KAAK,IAAI;IAC5D,GACA,eACF;IACF,KAAK,OAEH;GACJ;GAGF,MAAM,gBAA4B,EAChC,KAAK,QAAQ,OAAO,CAAC,EACvB;GACA,IAAI,aAAa,WAAW,cAAc,eAAe,QAAQ;QAC5D,cAAc,cAAc,QAAQ;GACzC,IAAI,QAAQ,gBAAgB,KAAA,GAAW,cAAc,cAAc,QAAQ;GAE3E,MAAM,SAAoB;IACxB;IACA,cAAc,KAAK;IACnB,aAAa,KAAK,QAAQ;IAC1B,MAAM,KAAK;IACX,OAAO,QAAQ;IACf,YAAY,QAAQ;IACpB,YAAY,QAAQ;IACpB,WAAW,KAAK;IAChB;IACA,SAAS,QAAQ;IACjB,gBAAgB,QAAQ;IACxB,YAAY,QAAQ;IACpB,iBAAiB;IACjB,eAAe,QAAQ;IACvB,SAAS;IACT,GAAI,QAAQ,eAAe,EAAE,cAAc,QAAQ,aAAa,IAAI,CAAC;IACrE,aAAa,QAAQ;IACrB;IACA,YAAY,KAAK,SAAS;GAC5B;GACA,MAAM,gBACJ,QAAQ,iBACP,OAAO,KAAK,iBAAiB,aAC1B,MAAM,KAAK,aAAa;IACtB,YAAY,KAAK;IACjB;IACA,WAAW,KAAK,QAAQ;IACxB,YAAY,KAAK,SAAS;IAC1B,MAAM,KAAK;IACX,SAAS,KAAK,QAAQ;IACtB,cAAc,KAAK,SAAS,QAAQ,CAAC;GACvC,CAAC,IACD,KAAK;GACX,IAAI,kBAAkB,KAAA,GAAW;IAC/B,MAAM,eAAe,MAAM,wBAAwB,aAAa;IAChE,6BAA6B,cAAc,QAAQ,OAAO,QAAQ,UAAU;IAC5E,OAAO,eAAe;GACxB;GACA,OAAO;IAAE,QAAQ,kBAAkB,MAAM;IAAG,WAAW;GAAgB;EACzE,UAAU;GAIR,SAAS,OAAO,KAAK;EACvB;CACF;CAEA,MAAM,UAAU,MAAM,KAAK,EAAE,QAAQ,KAAK,IAAI,aAAa,MAAM,MAAM,EAAE,SAAS,OAAO,CAAC;CAI1F,MAAM,QAAQ,WAAW,OAAO;CAMhC,KAAK,MAAM,CAAC,OAAO,YAAY,UAC7B,MAAM,cAAc,SAAS,OAAO,kDAAkD;CAExF,SAAS,MAAM;CAEf,IAAI,cAAc,SAAS,GACzB,MAAM,cAAc,WAAW,IAC3B,cAAc,KACd,IAAI,eACF,eACA,oBAAoB,cAAc,OAAO,iDAC3C;CAIN,IAAI;CACJ,IAAI,KAAK,QAQP,SAAS,MAAM,eAAe,MAAM;EANlC,GAAG,KAAK;EACR,YAAY,KAAK,OAAO;EACxB,OAAO,aAAa,QAAQ,WAAW;EACvC,aAAa,IAAI,KAAK,IAAI,CAAC,CAAC,CAAC,YAAY;EACzC,qBAAqB,uBAAuB,KAAA;CAED,CAAC;CAGhD,MAAM,UAAU,IAAI,KAAK,IAAI,CAAC,CAAC,CAAC,YAAY;CAE5C,OAAO;EACL,YAAY,KAAK;EACjB;EACA;EACA;EACA;EACA;EACA;EACA;EACA;CACF;AACF;AAIA,IAAM,qBAAN,cAAiC,MAAM;CACrC;CACA;CACA,YAAY,QAAmB,WAAgC;EAC7D,MAAM,QAAQ,OAAO,UAAU,GAAG,OAAO,WAAW,GAAG,OAAO,KAAK,WAAW,OAAO,QAAQ;EAC7F,KAAK,SAAS;EACd,KAAK,YAAY;CACnB;AACF;;;;;;;;;;;;;;;;AAiBA,eAAsB,cACpB,SACA,OACA,QACe;CACf,MAAM,WAAW,MAAM,QAAQ,WAAW,OAAO,KAAK;CACtD,IAAI,aAAa,KAAA,GAAW;CAC5B,IAAI,SAAS,WAAW,WAAW;CACnC,MAAM,QAAQ,SAAS,MAAM;AAC/B;AAEA,SAAS,sBAAsB,SAA6B;CAC1D,QAAQ,WAAmD;EACzD,IAAI,CAAC,SACH,MAAM,IAAI,MACR,6LACF;EAEF,OAAO,IAAI,0BAA0B,EACnC,KAAK,GAAG,QAAQ,cAAc,OAAO,QACvC,CAAC;CACH;AACF;AAEA,eAAe,wBACb,OAC2B;CAC3B,IAAI,mBAAmB,KAAK,GAAG;EAC7B,IAAI,CAAE,MAAM,uBAAuB,KAAK,GACtC,MAAM,IAAI,MAAM,iEAAiE;EAEnF,OAAO;CACT;CACA,OAAO,sBAAsB,KAAK;AACpC;AAEA,SAAS,mBACP,OAC2B;CAC3B,OAAO,mBAAmB,SAAS,YAAY;AACjD;AAEA,SAAS,6BACP,SACA,OACA,YACM;CACN,IAAI,QAAQ,UAAU,KAAA,KAAa,QAAQ,UAAU,OACnD,MAAM,IAAI,MACR,wCAAwC,QAAQ,MAAM,kCAAkC,MAAM,EAChG;CAEF,IAAI,QAAQ,eAAe,KAAA,KAAa,QAAQ,eAAe,YAC7D,MAAM,IAAI,MACR,6CAA6C,QAAQ,WAAW,uCAAuC,WAAW,EACpH;AAEJ;AAEA,SAAS,aAAa,QAAuC;CAG3D,MAAM,OAAO,GAAG,OAAO,WAAW,IAAI,OAAO,UAAU,IAAI,OAAO,WAAW,IAAI,OAAO;CAExF,IAAI,KAAK;CACT,IAAI,KAAK;CACT,KAAK,IAAI,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK;EACpC,MAAM,IAAI,KAAK,WAAW,CAAC;EAC3B,KAAK,KAAK,KAAK,KAAK,GAAG,QAAU,MAAM;EACvC,KAAK,KAAK,KAAK,KAAK,GAAG,UAAU,MAAM;CACzC;CACA,OAAO,OAAO,GAAG,SAAS,EAAE,CAAC,CAAC,SAAS,GAAG,GAAG,IAAI,GAAG,SAAS,EAAE,CAAC,CAAC,SAAS,GAAG,GAAG;AAClF"}

package/dist/{extract-usage-DIQpN-ww.js → extract-usage-p-56bh8q.js}

@@ -1,6 +1,6 @@
-import { s as ValidationError } from "./errors-8YnH8WlF.js";
+import { c as ValidationError } from "./errors-D-LKuDhb.js";
 import { LLM_CACHED_TOKEN_ATTR_KEYS, LLM_CACHE_WRITE_TOKEN_ATTR_KEYS, LLM_COST_ATTR_KEYS, LLM_INPUT_TOKEN_ATTR_KEYS, LLM_OUTPUT_TOKEN_ATTR_KEYS, LLM_REASONING_TOKEN_ATTR_KEYS, RUN_COST_ATTR_KEYS } from "./trace-attributes.js";
-import "./tools-D8yTtNSN.js";
+import "./tools-DZk2Jn64.js";
 import { t as FAILURE_CLASSES } from "./schema-CRhEY1SO.js";
 import { TOKEN_USAGE_INPUT_KEYS, TOKEN_USAGE_OUTPUT_KEYS, firstTokenCount, tokenUsageSource } from "@tangle-network/agent-core/telemetry";
 import { SSEChunkParser } from "@tangle-network/agent-core/sse";
@@ -460,4 +460,4 @@ async function extractUsageFromResponse(response, sseOptions) {
 //#endregion
 export { recordAggregateMeasurements as a, readTaskFailureLabels as i, extractUsageFromResponse as n, summarizeExecutionMeasurements as o, extractUsageFromSse as r, summarizeTraceErrors as s, extractUsage as t };
 
-//# sourceMappingURL=extract-usage-DIQpN-ww.js.map
+//# sourceMappingURL=extract-usage-p-56bh8q.js.map