@tangle-network/agent-eval 0.150.2 → 0.161.1

package/dist/{counterfactual--bpysZF0.d.ts → counterfactual-BaFUWK3H.d.ts}

@@ -1,6 +1,6 @@
-import { s as TraceStore } from "./store-CT9YIIve.js";
-import { i as TraceEmitter } from "./emitter-DGQGoLyj.js";
-import { n as TrajectoryStep, t as Trajectory } from "./trajectory-YC15QDYQ.js";
+import { s as TraceStore } from "./store-B06JdC56.js";
+import { i as TraceEmitter } from "./emitter-D_jYSGRd.js";
+import { n as TrajectoryStep, t as Trajectory } from "./trajectory-Bi157Gun.js";
 //#region src/counterfactual.d.ts
 type CounterfactualMutation = {
   kind: 'swap-model';
@@ -57,4 +57,4 @@ interface CounterfactualRunner {
 declare function runCounterfactual(store: TraceStore, originalRunId: string, mutation: CounterfactualMutation, runner: CounterfactualRunner): Promise<CounterfactualResult>;
 //#endregion
 export { runCounterfactual as a, CounterfactualRunner as i, CounterfactualMutation as n, CounterfactualResult as r, CounterfactualContext as t };
-//# sourceMappingURL=counterfactual--bpysZF0.d.ts.map
+//# sourceMappingURL=counterfactual-BaFUWK3H.d.ts.map

package/dist/{counterfactual--bpysZF0.d.ts.map → counterfactual-BaFUWK3H.d.ts.map}

@@ -1 +1 @@
-{"version":3,"file":"counterfactual--bpysZF0.d.ts","names":[],"sources":["../src/counterfactual.ts"],"mappings":";;;;KAoBY;EACN;EAAoB;EAAY;;EAChC;EAA0B;EAAY;;EACtC;EAAwB;;EACxB;EAA+B;EAAY;;EAE3C;EACA;EACA;EACA,QAAQ,MAAM,mBAAmB;;UAGtB;EACf;EACA,oBAAoB;;;EAGpB,QAAQ;EACR,UAAU;;;;EAIV,aAAa;;UAGE;EACf;EACA;EACA,UAAU;;EAEV;IACE;IACA;IACA;;;UAIa;;;;;;EAMf,cAAc,KAAK,uBAAuB,SAAS,iBAAiB;;iBAGhD,kBACpB,OAAO,YACP,uBACA,UAAU,wBACV,QAAQ,uBACP,QAAQ"}
+{"version":3,"file":"counterfactual-BaFUWK3H.d.ts","names":[],"sources":["../src/counterfactual.ts"],"mappings":";;;;KAoBY;EACN;EAAoB;EAAY;;EAChC;EAA0B;EAAY;;EACtC;EAAwB;;EACxB;EAA+B;EAAY;;EAE3C;EACA;EACA;EACA,QAAQ,MAAM,mBAAmB;;UAGtB;EACf;EACA,oBAAoB;;;EAGpB,QAAQ;EACR,UAAU;;;;EAIV,aAAa;;UAGE;EACf;EACA;EACA,UAAU;;EAEV;IACE;IACA;IACA;;;UAIa;;;;;;EAMf,cAAc,KAAK,uBAAuB,SAAS,iBAAiB;;iBAGhD,kBACpB,OAAO,YACP,uBACA,UAAU,wBACV,QAAQ,uBACP,QAAQ"}

package/dist/{counterfactual-lDfCx0Uz.js → counterfactual-D_VWavVm.js}

@@ -1,5 +1,5 @@
 import { o as NotFoundError, s as ValidationError } from "./errors-Dngq5h35.js";
-import { t as TraceEmitter } from "./emitter-CPBAhxum.js";
+import { t as TraceEmitter } from "./emitter-BpYFQPj4.js";
 import { t as buildTrajectory } from "./trajectory-D_7rLrvE.js";
 //#region src/counterfactual.ts
 /**
@@ -95,4 +95,4 @@ function applyMutation(step, mutation) {
 //#endregion
 export { runCounterfactual as t };
 
-//# sourceMappingURL=counterfactual-lDfCx0Uz.js.map
+//# sourceMappingURL=counterfactual-D_VWavVm.js.map

package/dist/{counterfactual-lDfCx0Uz.js.map → counterfactual-D_VWavVm.js.map}

@@ -1 +1 @@
-{"version":3,"file":"counterfactual-lDfCx0Uz.js","names":[],"sources":["../src/counterfactual.ts"],"sourcesContent":["/**\n * Counterfactual replay — \"what would have happened if we'd changed\n * exactly one thing at turn N?\"\n *\n * The framework does NOT drive the agent — it sets up the replay\n * context (prior spans, prior state, mutation spec) and records the\n * resulting divergence. Consumers supply an `executeFrom(ctx)` callback\n * that runs their agent starting from turn N with the mutation applied.\n *\n * Counterfactual runs are recorded as a new Run with `layer='meta'` and\n * `parentRunId = originalRunId`, so downstream diff + correlation\n * pipelines see them natively.\n */\n\nimport { NotFoundError, ValidationError } from './errors'\nimport { TraceEmitter } from './trace/emitter'\nimport type { LlmSpan, Span, ToolSpan } from './trace/schema'\nimport type { TraceStore } from './trace/store'\nimport { buildTrajectory, type Trajectory, type TrajectoryStep } from './trajectory'\n\nexport type CounterfactualMutation =\n  | { kind: 'swap-model'; at: number; newModel: string }\n  | { kind: 'swap-tool-result'; at: number; newResult: unknown }\n  | { kind: 'truncate-after'; at: number }\n  | { kind: 'inject-system-message'; at: number; content: string }\n  | {\n      kind: 'custom'\n      at: number\n      describe: string\n      apply: (step: TrajectoryStep) => TrajectoryStep\n    }\n\nexport interface CounterfactualContext {\n  originalRunId: string\n  originalTrajectory: Trajectory\n  /** Steps up to (but not including) the mutation point — the prefix the\n   *  replayed agent inherits as its prior conversation/tool history. */\n  prefix: TrajectoryStep[]\n  mutation: CounterfactualMutation\n  /** Pre-applied mutation on the step at `mutation.at`. Consumers use this\n   *  as the FIRST step the replayed agent emits (they decide whether to\n   *  re-emit it or continue from there). */\n  mutatedStep: TrajectoryStep\n}\n\nexport interface CounterfactualResult {\n  counterfactualRunId: string\n  originalRunId: string\n  mutation: CounterfactualMutation\n  /** Structured delta summary — caller can extend via scoring. */\n  delta: {\n    originalOutcomeScore: number | null\n    counterfactualOutcomeScore: number | null\n    deltaScore: number | null\n  }\n}\n\nexport interface CounterfactualRunner {\n  /**\n   * Execute the agent from `ctx.prefix` with the mutation applied.\n   * MUST emit spans into the provided emitter so they become part of\n   * the counterfactual run. MUST call emitter.endRun() with a verdict.\n   */\n  executeFrom: (ctx: CounterfactualContext, emitter: TraceEmitter) => Promise<void>\n}\n\nexport async function runCounterfactual(\n  store: TraceStore,\n  originalRunId: string,\n  mutation: CounterfactualMutation,\n  runner: CounterfactualRunner,\n): Promise<CounterfactualResult> {\n  const originalRun = await store.getRun(originalRunId)\n  if (!originalRun) throw new NotFoundError(`counterfactual: run ${originalRunId} not found`)\n  const trajectory = await buildTrajectory(store, originalRunId)\n  if (mutation.at < 0 || mutation.at >= trajectory.steps.length) {\n    throw new ValidationError(\n      `counterfactual: mutation.at=${mutation.at} out of range [0, ${trajectory.steps.length})`,\n    )\n  }\n  const targetStep = trajectory.steps[mutation.at]!\n  const mutatedStep = applyMutation(targetStep, mutation)\n\n  const cfEmitter = new TraceEmitter(store)\n  await cfEmitter.startRun({\n    scenarioId: originalRun.scenarioId,\n    variantId: originalRun.variantId\n      ? `${originalRun.variantId}+cf:${mutation.kind}@${mutation.at}`\n      : `cf:${mutation.kind}@${mutation.at}`,\n    projectId: originalRun.projectId,\n    parentRunId: originalRunId,\n    layer: 'meta',\n    tags: { counterfactual: 'true', mutationKind: mutation.kind, mutationAt: String(mutation.at) },\n  })\n\n  await runner.executeFrom(\n    {\n      originalRunId,\n      originalTrajectory: trajectory,\n      prefix: trajectory.steps.slice(0, mutation.at),\n      mutation,\n      mutatedStep,\n    },\n    cfEmitter,\n  )\n\n  const counterfactual = await store.getRun(cfEmitter.runId)\n  const delta = {\n    originalOutcomeScore: originalRun.outcome?.score ?? null,\n    counterfactualOutcomeScore: counterfactual?.outcome?.score ?? null,\n    deltaScore:\n      originalRun.outcome?.score !== undefined && counterfactual?.outcome?.score !== undefined\n        ? counterfactual.outcome.score - originalRun.outcome.score\n        : null,\n  }\n  return { counterfactualRunId: cfEmitter.runId, originalRunId, mutation, delta }\n}\n\nfunction applyMutation(step: TrajectoryStep, mutation: CounterfactualMutation): TrajectoryStep {\n  if (mutation.kind === 'swap-model' && step.span.kind === 'llm') {\n    const llm = step.span as LlmSpan\n    return { ...step, span: { ...llm, model: mutation.newModel } }\n  }\n  if (mutation.kind === 'swap-tool-result' && step.span.kind === 'tool') {\n    const tool = step.span as ToolSpan\n    return { ...step, span: { ...tool, result: mutation.newResult } }\n  }\n  if (mutation.kind === 'inject-system-message' && step.span.kind === 'llm') {\n    const llm = step.span as LlmSpan\n    return {\n      ...step,\n      span: {\n        ...llm,\n        messages: [{ role: 'system', content: mutation.content }, ...llm.messages],\n      },\n    }\n  }\n  if (mutation.kind === 'custom') return mutation.apply(step)\n  // swap-tool-result on non-tool span / swap-model on non-llm / truncate-after: no step-level change.\n  return step\n}\n\n/**\n * Aggregate a batch of counterfactuals into a simple attribution table:\n * which mutation kinds move outcomes most? (Useful when you run a grid\n * over the same trajectory — swap-model at every llm span, swap-tool\n * at every tool span — and want a ranked summary.)\n */\nexport function attributeCounterfactuals(results: CounterfactualResult[]): Array<{\n  mutationKind: CounterfactualMutation['kind']\n  n: number\n  meanAbsDelta: number\n  meanSignedDelta: number\n}> {\n  const grouped = new Map<string, CounterfactualResult[]>()\n  for (const r of results) {\n    const arr = grouped.get(r.mutation.kind) ?? []\n    arr.push(r)\n    grouped.set(r.mutation.kind, arr)\n  }\n  const out: Array<{\n    mutationKind: CounterfactualMutation['kind']\n    n: number\n    meanAbsDelta: number\n    meanSignedDelta: number\n  }> = []\n  for (const [kind, items] of grouped) {\n    const deltas = items\n      .map((i) => i.delta.deltaScore)\n      .filter((d): d is number => typeof d === 'number')\n    if (deltas.length === 0) continue\n    const meanAbs = deltas.reduce((a, b) => a + Math.abs(b), 0) / deltas.length\n    const meanSigned = deltas.reduce((a, b) => a + b, 0) / deltas.length\n    out.push({\n      mutationKind: kind as CounterfactualMutation['kind'],\n      n: deltas.length,\n      meanAbsDelta: meanAbs,\n      meanSignedDelta: meanSigned,\n    })\n  }\n  return out.sort((a, b) => b.meanAbsDelta - a.meanAbsDelta)\n}\n\n// Re-export Span type for consumer ergonomics.\nexport type { Span }\n"],"mappings":";;;;;;;;;;;;;;;;;AAkEA,eAAsB,kBACpB,OACA,eACA,UACA,QAC+B;CAC/B,MAAM,cAAc,MAAM,MAAM,OAAO,aAAa;CACpD,IAAI,CAAC,aAAa,MAAM,IAAI,cAAc,uBAAuB,cAAc,WAAW;CAC1F,MAAM,aAAa,MAAM,gBAAgB,OAAO,aAAa;CAC7D,IAAI,SAAS,KAAK,KAAK,SAAS,MAAM,WAAW,MAAM,QACrD,MAAM,IAAI,gBACR,+BAA+B,SAAS,GAAG,oBAAoB,WAAW,MAAM,OAAO,EACzF;CAEF,MAAM,aAAa,WAAW,MAAM,SAAS;CAC7C,MAAM,cAAc,cAAc,YAAY,QAAQ;CAEtD,MAAM,YAAY,IAAI,aAAa,KAAK;CACxC,MAAM,UAAU,SAAS;EACvB,YAAY,YAAY;EACxB,WAAW,YAAY,YACnB,GAAG,YAAY,UAAU,MAAM,SAAS,KAAK,GAAG,SAAS,OACzD,MAAM,SAAS,KAAK,GAAG,SAAS;EACpC,WAAW,YAAY;EACvB,aAAa;EACb,OAAO;EACP,MAAM;GAAE,gBAAgB;GAAQ,cAAc,SAAS;GAAM,YAAY,OAAO,SAAS,EAAE;EAAE;CAC/F,CAAC;CAED,MAAM,OAAO,YACX;EACE;EACA,oBAAoB;EACpB,QAAQ,WAAW,MAAM,MAAM,GAAG,SAAS,EAAE;EAC7C;EACA;CACF,GACA,SACF;CAEA,MAAM,iBAAiB,MAAM,MAAM,OAAO,UAAU,KAAK;CACzD,MAAM,QAAQ;EACZ,sBAAsB,YAAY,SAAS,SAAS;EACpD,4BAA4B,gBAAgB,SAAS,SAAS;EAC9D,YACE,YAAY,SAAS,UAAU,KAAA,KAAa,gBAAgB,SAAS,UAAU,KAAA,IAC3E,eAAe,QAAQ,QAAQ,YAAY,QAAQ,QACnD;CACR;CACA,OAAO;EAAE,qBAAqB,UAAU;EAAO;EAAe;EAAU;CAAM;AAChF;AAEA,SAAS,cAAc,MAAsB,UAAkD;CAC7F,IAAI,SAAS,SAAS,gBAAgB,KAAK,KAAK,SAAS,OAAO;EAC9D,MAAM,MAAM,KAAK;EACjB,OAAO;GAAE,GAAG;GAAM,MAAM;IAAE,GAAG;IAAK,OAAO,SAAS;GAAS;EAAE;CAC/D;CACA,IAAI,SAAS,SAAS,sBAAsB,KAAK,KAAK,SAAS,QAAQ;EACrE,MAAM,OAAO,KAAK;EAClB,OAAO;GAAE,GAAG;GAAM,MAAM;IAAE,GAAG;IAAM,QAAQ,SAAS;GAAU;EAAE;CAClE;CACA,IAAI,SAAS,SAAS,2BAA2B,KAAK,KAAK,SAAS,OAAO;EACzE,MAAM,MAAM,KAAK;EACjB,OAAO;GACL,GAAG;GACH,MAAM;IACJ,GAAG;IACH,UAAU,CAAC;KAAE,MAAM;KAAU,SAAS,SAAS;IAAQ,GAAG,GAAG,IAAI,QAAQ;GAC3E;EACF;CACF;CACA,IAAI,SAAS,SAAS,UAAU,OAAO,SAAS,MAAM,IAAI;CAE1D,OAAO;AACT"}
+{"version":3,"file":"counterfactual-D_VWavVm.js","names":[],"sources":["../src/counterfactual.ts"],"sourcesContent":["/**\n * Counterfactual replay — \"what would have happened if we'd changed\n * exactly one thing at turn N?\"\n *\n * The framework does NOT drive the agent — it sets up the replay\n * context (prior spans, prior state, mutation spec) and records the\n * resulting divergence. Consumers supply an `executeFrom(ctx)` callback\n * that runs their agent starting from turn N with the mutation applied.\n *\n * Counterfactual runs are recorded as a new Run with `layer='meta'` and\n * `parentRunId = originalRunId`, so downstream diff + correlation\n * pipelines see them natively.\n */\n\nimport { NotFoundError, ValidationError } from './errors'\nimport { TraceEmitter } from './trace/emitter'\nimport type { LlmSpan, Span, ToolSpan } from './trace/schema'\nimport type { TraceStore } from './trace/store'\nimport { buildTrajectory, type Trajectory, type TrajectoryStep } from './trajectory'\n\nexport type CounterfactualMutation =\n  | { kind: 'swap-model'; at: number; newModel: string }\n  | { kind: 'swap-tool-result'; at: number; newResult: unknown }\n  | { kind: 'truncate-after'; at: number }\n  | { kind: 'inject-system-message'; at: number; content: string }\n  | {\n      kind: 'custom'\n      at: number\n      describe: string\n      apply: (step: TrajectoryStep) => TrajectoryStep\n    }\n\nexport interface CounterfactualContext {\n  originalRunId: string\n  originalTrajectory: Trajectory\n  /** Steps up to (but not including) the mutation point — the prefix the\n   *  replayed agent inherits as its prior conversation/tool history. */\n  prefix: TrajectoryStep[]\n  mutation: CounterfactualMutation\n  /** Pre-applied mutation on the step at `mutation.at`. Consumers use this\n   *  as the FIRST step the replayed agent emits (they decide whether to\n   *  re-emit it or continue from there). */\n  mutatedStep: TrajectoryStep\n}\n\nexport interface CounterfactualResult {\n  counterfactualRunId: string\n  originalRunId: string\n  mutation: CounterfactualMutation\n  /** Structured delta summary — caller can extend via scoring. */\n  delta: {\n    originalOutcomeScore: number | null\n    counterfactualOutcomeScore: number | null\n    deltaScore: number | null\n  }\n}\n\nexport interface CounterfactualRunner {\n  /**\n   * Execute the agent from `ctx.prefix` with the mutation applied.\n   * MUST emit spans into the provided emitter so they become part of\n   * the counterfactual run. MUST call emitter.endRun() with a verdict.\n   */\n  executeFrom: (ctx: CounterfactualContext, emitter: TraceEmitter) => Promise<void>\n}\n\nexport async function runCounterfactual(\n  store: TraceStore,\n  originalRunId: string,\n  mutation: CounterfactualMutation,\n  runner: CounterfactualRunner,\n): Promise<CounterfactualResult> {\n  const originalRun = await store.getRun(originalRunId)\n  if (!originalRun) throw new NotFoundError(`counterfactual: run ${originalRunId} not found`)\n  const trajectory = await buildTrajectory(store, originalRunId)\n  if (mutation.at < 0 || mutation.at >= trajectory.steps.length) {\n    throw new ValidationError(\n      `counterfactual: mutation.at=${mutation.at} out of range [0, ${trajectory.steps.length})`,\n    )\n  }\n  const targetStep = trajectory.steps[mutation.at]!\n  const mutatedStep = applyMutation(targetStep, mutation)\n\n  const cfEmitter = new TraceEmitter(store)\n  await cfEmitter.startRun({\n    scenarioId: originalRun.scenarioId,\n    variantId: originalRun.variantId\n      ? `${originalRun.variantId}+cf:${mutation.kind}@${mutation.at}`\n      : `cf:${mutation.kind}@${mutation.at}`,\n    projectId: originalRun.projectId,\n    parentRunId: originalRunId,\n    layer: 'meta',\n    tags: { counterfactual: 'true', mutationKind: mutation.kind, mutationAt: String(mutation.at) },\n  })\n\n  await runner.executeFrom(\n    {\n      originalRunId,\n      originalTrajectory: trajectory,\n      prefix: trajectory.steps.slice(0, mutation.at),\n      mutation,\n      mutatedStep,\n    },\n    cfEmitter,\n  )\n\n  const counterfactual = await store.getRun(cfEmitter.runId)\n  const delta = {\n    originalOutcomeScore: originalRun.outcome?.score ?? null,\n    counterfactualOutcomeScore: counterfactual?.outcome?.score ?? null,\n    deltaScore:\n      originalRun.outcome?.score !== undefined && counterfactual?.outcome?.score !== undefined\n        ? counterfactual.outcome.score - originalRun.outcome.score\n        : null,\n  }\n  return { counterfactualRunId: cfEmitter.runId, originalRunId, mutation, delta }\n}\n\nfunction applyMutation(step: TrajectoryStep, mutation: CounterfactualMutation): TrajectoryStep {\n  if (mutation.kind === 'swap-model' && step.span.kind === 'llm') {\n    const llm = step.span as LlmSpan\n    return { ...step, span: { ...llm, model: mutation.newModel } }\n  }\n  if (mutation.kind === 'swap-tool-result' && step.span.kind === 'tool') {\n    const tool = step.span as ToolSpan\n    return { ...step, span: { ...tool, result: mutation.newResult } }\n  }\n  if (mutation.kind === 'inject-system-message' && step.span.kind === 'llm') {\n    const llm = step.span as LlmSpan\n    return {\n      ...step,\n      span: {\n        ...llm,\n        messages: [{ role: 'system', content: mutation.content }, ...llm.messages],\n      },\n    }\n  }\n  if (mutation.kind === 'custom') return mutation.apply(step)\n  // swap-tool-result on non-tool span / swap-model on non-llm / truncate-after: no step-level change.\n  return step\n}\n\n/**\n * Aggregate a batch of counterfactuals into a simple attribution table:\n * which mutation kinds move outcomes most? (Useful when you run a grid\n * over the same trajectory — swap-model at every llm span, swap-tool\n * at every tool span — and want a ranked summary.)\n */\nexport function attributeCounterfactuals(results: CounterfactualResult[]): Array<{\n  mutationKind: CounterfactualMutation['kind']\n  n: number\n  meanAbsDelta: number\n  meanSignedDelta: number\n}> {\n  const grouped = new Map<string, CounterfactualResult[]>()\n  for (const r of results) {\n    const arr = grouped.get(r.mutation.kind) ?? []\n    arr.push(r)\n    grouped.set(r.mutation.kind, arr)\n  }\n  const out: Array<{\n    mutationKind: CounterfactualMutation['kind']\n    n: number\n    meanAbsDelta: number\n    meanSignedDelta: number\n  }> = []\n  for (const [kind, items] of grouped) {\n    const deltas = items\n      .map((i) => i.delta.deltaScore)\n      .filter((d): d is number => typeof d === 'number')\n    if (deltas.length === 0) continue\n    const meanAbs = deltas.reduce((a, b) => a + Math.abs(b), 0) / deltas.length\n    const meanSigned = deltas.reduce((a, b) => a + b, 0) / deltas.length\n    out.push({\n      mutationKind: kind as CounterfactualMutation['kind'],\n      n: deltas.length,\n      meanAbsDelta: meanAbs,\n      meanSignedDelta: meanSigned,\n    })\n  }\n  return out.sort((a, b) => b.meanAbsDelta - a.meanAbsDelta)\n}\n\n// Re-export Span type for consumer ergonomics.\nexport type { Span }\n"],"mappings":";;;;;;;;;;;;;;;;;AAkEA,eAAsB,kBACpB,OACA,eACA,UACA,QAC+B;CAC/B,MAAM,cAAc,MAAM,MAAM,OAAO,aAAa;CACpD,IAAI,CAAC,aAAa,MAAM,IAAI,cAAc,uBAAuB,cAAc,WAAW;CAC1F,MAAM,aAAa,MAAM,gBAAgB,OAAO,aAAa;CAC7D,IAAI,SAAS,KAAK,KAAK,SAAS,MAAM,WAAW,MAAM,QACrD,MAAM,IAAI,gBACR,+BAA+B,SAAS,GAAG,oBAAoB,WAAW,MAAM,OAAO,EACzF;CAEF,MAAM,aAAa,WAAW,MAAM,SAAS;CAC7C,MAAM,cAAc,cAAc,YAAY,QAAQ;CAEtD,MAAM,YAAY,IAAI,aAAa,KAAK;CACxC,MAAM,UAAU,SAAS;EACvB,YAAY,YAAY;EACxB,WAAW,YAAY,YACnB,GAAG,YAAY,UAAU,MAAM,SAAS,KAAK,GAAG,SAAS,OACzD,MAAM,SAAS,KAAK,GAAG,SAAS;EACpC,WAAW,YAAY;EACvB,aAAa;EACb,OAAO;EACP,MAAM;GAAE,gBAAgB;GAAQ,cAAc,SAAS;GAAM,YAAY,OAAO,SAAS,EAAE;EAAE;CAC/F,CAAC;CAED,MAAM,OAAO,YACX;EACE;EACA,oBAAoB;EACpB,QAAQ,WAAW,MAAM,MAAM,GAAG,SAAS,EAAE;EAC7C;EACA;CACF,GACA,SACF;CAEA,MAAM,iBAAiB,MAAM,MAAM,OAAO,UAAU,KAAK;CACzD,MAAM,QAAQ;EACZ,sBAAsB,YAAY,SAAS,SAAS;EACpD,4BAA4B,gBAAgB,SAAS,SAAS;EAC9D,YACE,YAAY,SAAS,UAAU,KAAA,KAAa,gBAAgB,SAAS,UAAU,KAAA,IAC3E,eAAe,QAAQ,QAAQ,YAAY,QAAQ,QACnD;CACR;CACA,OAAO;EAAE,qBAAqB,UAAU;EAAO;EAAe;EAAU;CAAM;AAChF;AAEA,SAAS,cAAc,MAAsB,UAAkD;CAC7F,IAAI,SAAS,SAAS,gBAAgB,KAAK,KAAK,SAAS,OAAO;EAC9D,MAAM,MAAM,KAAK;EACjB,OAAO;GAAE,GAAG;GAAM,MAAM;IAAE,GAAG;IAAK,OAAO,SAAS;GAAS;EAAE;CAC/D;CACA,IAAI,SAAS,SAAS,sBAAsB,KAAK,KAAK,SAAS,QAAQ;EACrE,MAAM,OAAO,KAAK;EAClB,OAAO;GAAE,GAAG;GAAM,MAAM;IAAE,GAAG;IAAM,QAAQ,SAAS;GAAU;EAAE;CAClE;CACA,IAAI,SAAS,SAAS,2BAA2B,KAAK,KAAK,SAAS,OAAO;EACzE,MAAM,MAAM,KAAK;EACjB,OAAO;GACL,GAAG;GACH,MAAM;IACJ,GAAG;IACH,UAAU,CAAC;KAAE,MAAM;KAAU,SAAS,SAAS;IAAQ,GAAG,GAAG,IAAI,QAAQ;GAC3E;EACF;CACF;CACA,IAAI,SAAS,SAAS,UAAU,OAAO,SAAS,MAAM,IAAI;CAE1D,OAAO;AACT"}

package/dist/{dataset-CJjKqQfA.d.ts → dataset-DQqhOCPt.d.ts}

@@ -49,7 +49,7 @@ interface DatasetProvenance {
 interface DatasetManifest {
   name: string;
   provenance: DatasetProvenance;
-  /** sha256 hex over canonicalized scenarios. */
+  /** sha256 hex over the RFC 8785 canonical JSON of the id-sorted scenarios. */
   contentHash: string;
   scenarioCount: number;
   splitCounts: Record<DatasetSplit, number>;
@@ -98,12 +98,13 @@ declare class Dataset {
   add(scenario: DatasetScenario): void;
   remove(scenarioId: string): void;
   /**
-   * Stable JSON-Lines serialization — deterministic byte-for-byte.
-   * Write to disk for contamination-verifiable archives.
+   * Stable JSON-Lines serialization — one RFC 8785 canonical row per scenario,
+   * sorted by id, deterministic byte-for-byte. Write to disk for
+   * contamination-verifiable archives.
    */
   toJsonl(): string;
   static fromJsonl(jsonl: string, manifest: Omit<DatasetManifest, 'contentHash' | 'scenarioCount' | 'splitCounts'>): Dataset;
 }
 //#endregion
 export { DatasetSplit as i, DatasetManifest as n, DatasetScenario as r, Dataset as t };
-//# sourceMappingURL=dataset-CJjKqQfA.d.ts.map
+//# sourceMappingURL=dataset-DQqhOCPt.d.ts.map

package/dist/{dataset-CJjKqQfA.d.ts.map → dataset-DQqhOCPt.d.ts.map}

@@ -1 +1 @@
-{"version":3,"file":"dataset-CJjKqQfA.d.ts","names":[],"sources":["../src/dataset.ts"],"mappings":";;;;;;;;;;;;;;;;;KAiBY;KACA;UAEK;EACf;;EAEA;EACA,QAAQ;EACR,aAAa;;EAEb;;;;;;;;;;EAUA;EACA,OAAO;;UAGQ;EACf;EACA;EACA;EACA;EACA;;EAEA;;UAGe;EACf;EACA,YAAY;;EAEZ;EACA;EACA,aAAa,OAAO;;UAGL;EACf,QAAQ;EACR,aAAa;;EAEb;EACA;;EAEA,UAAU,UAAU;;EAEpB;;cAcW;WACF;WACA,YAAY;UACb;UACA;EAER,YAAY;IACV;IACA,YAAY;IACZ,WAAW;IACX;;;EASF,gBAAgB;MAIZ;;;;;EAQJ,MAAM,UAAS,eAAoB;;;;;EAqB7B,YAAY,QAAQ;;EAgB1B,MAAM,YAAW;IAAU;IAAc;OAA0B;EAWnE;EAIA,IAAI,UAAU;EAQd,OAAO;;;;;EAWP;SAQO,UACL,eACA,UAAU,KAAK,oEACd"}
+{"version":3,"file":"dataset-DQqhOCPt.d.ts","names":[],"sources":["../src/dataset.ts"],"mappings":";;;;;;;;;;;;;;;;;KAiBY;KACA;UAEK;EACf;;EAEA;EACA,QAAQ;EACR,aAAa;;EAEb;;;;;;;;;;EAUA;EACA,OAAO;;UAGQ;EACf;EACA;EACA;EACA;EACA;;EAEA;;UAGe;EACf;EACA,YAAY;;EAEZ;EACA;EACA,aAAa,OAAO;;UAGL;EACf,QAAQ;EACR,aAAa;;EAEb;EACA;;EAEA,UAAU,UAAU;;EAEpB;;cAeW;WACF;WACA,YAAY;UACb;UACA;EAER,YAAY;IACV;IACA,YAAY;IACZ,WAAW;IACX;;;EASF,gBAAgB;MAIZ;;;;;EAQJ,MAAM,UAAS,eAAoB;;;;;EAqB7B,YAAY,QAAQ;;EAgB1B,MAAM,YAAW;IAAU;IAAc;OAA0B;EAWnE;EAIA,IAAI,UAAU;EAQd,OAAO;;;;;;EAYP;SAQO,UACL,eACA,UAAU,KAAK,oEACd"}

package/dist/{default-registry-Cw0Ohdoj.d.ts → default-registry-G9CKMNkc.d.ts}

@@ -1,8 +1,8 @@
-import { i as AnalystFinding, n as AnalystContext, w as TraceAnalysisStore } from "./types-B2NsbrNy.js";
-import { l as ExactCapableAnalyst } from "./exact-types-ccQAyut1.js";
-import { n as AnalystRegistry, r as AnalystRegistryOptions } from "./registry-BQwrSYpC.js";
-import { a as TraceAnalystLimits, d as RawAnalystFinding, i as TraceAnalysisEngineResult, n as TraceAnalysisEngine } from "./engine-BLzhNzoY.js";
-import { t as TraceToolGroupName } from "./tool-groups-B4tqh8jB.js";
+import { i as AnalystFinding, n as AnalystContext, w as TraceAnalysisStore } from "./types-D9ssmxKL.js";
+import { l as ExactCapableAnalyst } from "./exact-types-qnexxJ1Z.js";
+import { n as AnalystRegistry, r as AnalystRegistryOptions } from "./registry-8You7OK1.js";
+import { c as RawAnalystFinding, i as TraceAnalystLimits, r as TraceAnalysisEngineResult, t as TraceAnalysisEngine } from "./engine-Cu5qD5Fc.js";
+import { t as TraceToolGroupName } from "./tool-groups-Ci8i9ErB.js";
 //#region src/trace-analyst/behavioral-metrics.d.ts
 type SuboptimalCode = 'monotonic-input-growth' | 'output-length-decay' | 'single-tool-dependency' | 'no-self-verification';
 interface SuboptimalSignal {
@@ -92,7 +92,6 @@ declare function runTraceAnalyst(args: {
 /** Adapt a research definition to the common Analyst registry contract. */
 declare function createTraceAnalyst(definition: TraceAnalystDefinition, options: CreateTraceAnalystOptions): ExactCapableAnalyst<TraceAnalysisStore>;
 declare function renderPriorFindings(prior: AnalystContext['priorFindings']): string;
-declare function renderUpstreamFindings(upstream: AnalystContext['upstreamFindings']): string;
 //#endregion
 //#region src/analyst/default-registry.d.ts
 interface DefaultAnalystRegistryOptions {
@@ -110,5 +109,5 @@ interface DefaultAnalystRegistryOptions {
 }
 declare function buildDefaultAnalystRegistry(options?: DefaultAnalystRegistryOptions): AnalystRegistry;
 //#endregion
-export { createTraceAnalyst as a, runTraceAnalyst as c, deriveEfficiencyFindings as d, TraceAnalystDefinition as i, BehavioralAnalystOptions as l, buildDefaultAnalystRegistry as n, renderPriorFindings as o, CreateTraceAnalystOptions as r, renderUpstreamFindings as s, DefaultAnalystRegistryOptions as t, behavioralAnalyst as u };
-//# sourceMappingURL=default-registry-Cw0Ohdoj.d.ts.map
+export { createTraceAnalyst as a, BehavioralAnalystOptions as c, TraceAnalystDefinition as i, behavioralAnalyst as l, buildDefaultAnalystRegistry as n, renderPriorFindings as o, CreateTraceAnalystOptions as r, runTraceAnalyst as s, DefaultAnalystRegistryOptions as t, deriveEfficiencyFindings as u };
+//# sourceMappingURL=default-registry-G9CKMNkc.d.ts.map

package/dist/{default-registry-Cw0Ohdoj.d.ts.map → default-registry-G9CKMNkc.d.ts.map}

@@ -1 +1 @@
-{"version":3,"file":"default-registry-Cw0Ohdoj.d.ts","names":[],"sources":["../src/trace-analyst/behavioral-metrics.ts","../src/analyst/behavioral-analyst.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;;;;UC1BR;;EAEf;;EAEA;;;;;;iBA+Dc,yBACd,SAAS,mBACT;EAAQ;EAAoB;IAC3B;;iBAkCa,kBACd,UAAS,2BACR,oBAAoB;;;;UCrHN;EACf;EACA;EACA;EACA;;EAEA,sBAAsB,SAAS;;EAE/B;;EAEA,WAAW;;EAEX,kBACE,OAAO,oBACP,SAAS,wCACe;EAC1B,SAAS,QAAQ;EACjB,eAAe,KAAK,mBAAmB,SAAS,mBAAmB;EACnE;EACA;;UAGe;EACf,QAAQ;;EAER;EACA;;;iBAIoB,gBAAgB;EACpC,YAAY;EACZ,QAAQ;EACR,OAAO;EACP,SAAS;EACT;IACE,QAAQ;;iBAgFI,mBACd,YAAY,wBACZ,SAAS,4BACR,oBAAoB;iBA+QP,oBAAoB,OAAO;iBAe3B,uBAAuB,UAAU;;;UCnahC;;;;;;EAMf,SAAS;EACT,uBAAuB;;EAEvB;EACA,aAAa;EACb,WAAW;;iBAGG,4BACd,UAAS,gCACR"}
+{"version":3,"file":"default-registry-G9CKMNkc.d.ts","names":[],"sources":["../src/trace-analyst/behavioral-metrics.ts","../src/analyst/behavioral-analyst.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;;;;UC1BR;;EAEf;;EAEA;;;;;;iBA+Dc,yBACd,SAAS,mBACT;EAAQ;EAAoB;IAC3B;;iBAkCa,kBACd,UAAS,2BACR,oBAAoB;;;;UCrHN;EACf;EACA;EACA;EACA;;EAEA,sBAAsB,SAAS;;EAE/B;;EAEA,WAAW;;EAEX,kBACE,OAAO,oBACP,SAAS,wCACe;EAC1B,SAAS,QAAQ;EACjB,eAAe,KAAK,mBAAmB,SAAS,mBAAmB;EACnE;EACA;;UAGe;EACf,QAAQ;;EAER;EACA;;;iBAIoB,gBAAgB;EACpC,YAAY;EACZ,QAAQ;EACR,OAAO;EACP,SAAS;EACT;IACE,QAAQ;;iBAgFI,mBACd,YAAY,wBACZ,SAAS,4BACR,oBAAoB;iBA+QP,oBAAoB,OAAO;;;UCpZ1B;;;;;;EAMf,SAAS;EACT,uBAAuB;;EAEvB;EACA,aAAa;EACb,WAAW;;iBAGG,4BACd,UAAS,gCACR"}

package/dist/{define-agent-eval-CEQWL9Hy.d.ts → define-agent-eval-Dx1JnPEa.d.ts}

@@ -1,10 +1,10 @@
 import { c as CostLedgerHandle, f as CostLedgerSummary, m as CostReceipt } from "./cost-ledger-DbQdN3nO.js";
-import { _ as ProposalFinding } from "./types-B2NsbrNy.js";
-import { H as SurfaceProposer, R as Scenario, S as JudgeConfig, a as CampaignResult, d as DispatchContext, j as MutableSurface, k as LabeledScenarioStore, p as Gate } from "./types-DdFNuyxQ.js";
-import { o as InsightReport } from "./insight-report-BeT8KCgI.js";
-import { C as RunOptimizationOptions, Cn as CampaignStorage, E as RunEvalOptions, J as OptimizationMethodProvenance, S as PremeasuredOptimizationBaseline, W as OptimizationMethod, Y as OptimizationMethodResult, b as RunImprovementLoopResult, l as LoopProvenanceRecord } from "./provenance-oA4-zUqm.js";
+import { An as CampaignCellRetryPolicy, C as RunImprovementLoopResult, E as RunOptimizationOptions, Ln as CampaignStorage, T as PremeasuredOptimizationBaseline, at as OptimizationMethod, f as LoopProvenanceRecord, k as RunEvalOptions, lt as OptimizationMethodProvenance, ut as OptimizationMethodResult, wt as SearchHistoryReceipt } from "./transient-failure-DKF5Mofa.js";
+import { _ as ProposalFinding } from "./types-D9ssmxKL.js";
+import { H as SurfaceProposer, R as Scenario, S as JudgeConfig, a as CampaignResult, d as DispatchContext, j as MutableSurface, k as LabeledScenarioStore, p as Gate } from "./types-D4s7Z6nq.js";
+import { o as InsightReport } from "./insight-report-DRe8LB6d.js";
 import { o as PowerPreflight } from "./pareto-BqNW3LJR.js";
-import { n as HostedTenant } from "./client-kPQYT_56.js";
+import { n as HostedTenant } from "./client-L9VVPkim.js";
 //#region src/contract/self-improve.d.ts
 interface SelfImproveBudget {
   /** Hard spend cap across the full run. Each paid call reserves its enforced
@@ -166,6 +166,13 @@ interface SelfImproveOptions<TScenario extends Scenario, TArtifact> {
   /** Per-cell agent dispatch deadline, applied to baseline, candidate, and
    *  held-out campaigns. Default 600_000 ms. Set 0 to disable. */
   dispatchTimeoutMs?: number;
+  /** Bounded in-run retry of failed cells — same as
+   *  `RunCampaignOptions.cellRetry`, applied to baseline, candidate, and
+   *  held-out campaigns. Pair with `transientDispatchFailure()` so a
+   *  transport hiccup (a router 503, a dropped stream) is re-dispatched in
+   *  the same slot instead of leaving holdout coverage incomplete. Absent by
+   *  default: a failed cell is final. */
+  cellRetry?: CampaignCellRetryPolicy;
   /** Streaming hook — fires on baseline + each generation + gate decision.
    *  Consumer routes events wherever (UI, dashboard, logs). */
   onProgress?: (event: SelfImproveProgressEvent) => void;
@@ -225,6 +232,16 @@ interface SelfImproveOptions<TScenario extends Scenario, TArtifact> {
    *  winner-selection and the ship-gate rank on the identical metric and cannot
    *  invert. See `RunOptimizationOptions.selectionRankKey`. */
   selectionRankKey?: RunOptimizationOptions<TScenario, TArtifact>['selectionRankKey'];
+  /** Override which scored surface each generation MUTATES. Defaults to the
+   *  global incumbent (hill-climb; the recorded lineage is a chain). Pass
+   *  `crowdedFrontierParent({ seed })` or a custom policy to draw the parent
+   *  from the Pareto frontier; promotion still compares against the incumbent.
+   *  Proposer mode only. See `RunOptimizationOptions.selectParent`. */
+  selectParent?: RunOptimizationOptions<TScenario, TArtifact>['selectParent'];
+  /** Record this run's candidate search into a durable `SearchLedger` and
+   *  return the bounded receipt on `searchHistory`. See
+   *  `RunOptimizationOptions.searchLedger`. */
+  searchLedger?: RunOptimizationOptions<TScenario, TArtifact>['searchLedger'];
 }
 interface SelfImproveResult<TScenario extends Scenario, TArtifact> {
   /** Composite mean across all scenarios, baseline run. When
@@ -275,6 +292,9 @@ interface SelfImproveResult<TScenario extends Scenario, TArtifact> {
   /** Run-wide receipts across proposal, search, holdout, judging, analysis,
    *  and promotion work, with phase and actor attribution. */
   receipts: CostReceipt[];
+  /** Bounded proof envelope over this run's canonical search ledger. Present
+   *  only when `searchLedger` was supplied. */
+  searchHistory?: SearchHistoryReceipt;
   /** Exact external method and source identity, when `method` was used. */
   optimization?: {
     name: string;
@@ -384,4 +404,4 @@ interface DefinedAgentEval<TScenario extends Scenario, TArtifact> {
 declare function defineAgentEval<TScenario extends Scenario, TArtifact>(defaults: DefineAgentEvalOptions<TScenario, TArtifact>): DefinedAgentEval<TScenario, TArtifact>;
 //#endregion
 export { DefinedAgentEval as a, SelfImproveOptions as c, SelfImproveRunError as d, selfImprove as f, DefineAgentEvalOptions as i, SelfImproveProgressEvent as l, AgentEvalEvaluateOptions as n, defineAgentEval as o, AgentEvalImproveOptions as r, SelfImproveBudget as s, AgentEvalAgent as t, SelfImproveResult as u };
-//# sourceMappingURL=define-agent-eval-CEQWL9Hy.d.ts.map
+//# sourceMappingURL=define-agent-eval-Dx1JnPEa.d.ts.map

package/dist/define-agent-eval-Dx1JnPEa.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"define-agent-eval-Dx1JnPEa.d.ts","names":[],"sources":["../src/contract/self-improve.ts","../src/contract/define-agent-eval.ts"],"mappings":";;;;;;;;UAiEiB;;;EAGf;;;;EAIA;;EAEA;;EAEA;;;EAGA;;;EAGA;;;;EAIA;;EAEA,mBAAmB;;;;;;;;;EASnB;;EAEA;;;;;EAKA;;KAGU;EACN;EAA0B;;EAC1B;EAA4B;EAAuB;;EACnD;EAA4B;EAAe;;EAC3C;EAA8B;EAAe;EAAuB;;EAGpE;EAAsB;EAAkB;;EACxC;EAAyB;EAAW;EAAY;EAAa;;UAElD,mBAAmB,kBAAkB,UAAU;;;;;;;;;;;;;;EAc9D,QAAQ,SAAS,gBAAgB,UAAU,WAAW,KAAK,oBAAoB,QAAQ;;;;;;;EAQvF;;;EAIA,WAAW;;EAGX,OAAO,YAAY,WAAW;;;EAI9B,iBAAiB;;EAGjB,SAAS;;;;;;;;;;;;EAaT,sBAAsB,gCAAgC,WAAW;;;;;EAMjE,WAAW,gBAAgB;;;;;;EAO3B,SAAS,mBAAmB,WAAW;;;EAIvC,qBAAqB;;;EAIrB,OAAO,KAAK,WAAW;;;;;;;;EASvB,cAAc,eAAe,gBAAgB,iBAAiB,mBAAmB;;;;EAKjF,UAAU;;;;EAKV;;;EAIA,gBAAgB,QAAQ;;;;;EAMxB,iBAAiB;IACf,UAAU;IACV;IACA;;;;EAKF;;;;;;;EAQA,YAAY;;;EAIZ,cAAc,OAAO;;;;EAKrB;EACA;EACA;;;;;;;;;;;;;;EAeA,eAAe;;;EAIf,eAAe;;;;EAKf,eAAe;;EAGf;;;;;;;;EASA;;;;;;;;;EAUA,oBAAoB,uBAAuB,WAAW;;;EAItD,WAAW;;;;;;;EAQX,mBAAmB,uBAAuB,WAAW;;;;;;EAOrD,eAAe,uBAAuB,WAAW;;;;EAKjD,eAAe,uBAAuB,WAAW;;UAGlC,kBAAkB,kBAAkB,UAAU;;;;EAI7D;IACE;IACA,aAAa;;;;;EAKf;IACE;IACA,aAAa;IACb,SAAS;;;IAGT;;;;IAIA;;;;;;EAMF;;;EAGA;;;;EAIA,YAAY;;EAEZ;;;EAGA;;EAEA;;EAEA;;EAEA,MAAM;;;EAGN,UAAU;;;EAGV,gBAAgB;;EAEhB;IACE;IACA,MAAM;IACN;IACA,aAAa;;;;;;;;EAQf,SAAS;;;;;EAKT,QAAQ;;;;;;EAMR,KAAK,yBAAyB,WAAW;;;cAI9B,4BAA4B;WAC9B,MAAM;WACN,UAAU;EAEnB,YAAY,gBAAgB,QAAQ;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBAgMhB,YAAY,kBAAkB,UAAU,WAC5D,MAAM,mBAAmB,WAAW,aACnC,QAAQ,kBAAkB,WAAW;;;KChkB5B,eAAe,kBAAkB,UAAU,cACrD,SAAS,gBACT,UAAU,WACV,KAAK,oBACF,QAAQ;KAED,uBAAuB,kBAAkB,UAAU,aAAa,mBAC1E,WACA;UAGe,yBAAyB,kBAAkB,UAAU,mBAC5D,KACN,eAAe,WAAW;;EAI5B,YAAY;;EAEZ,UAAU;;EAEV,QAAQ,eAAe,WAAW;;EAElC,QAAQ,YAAY,WAAW;;EAE/B,SAAS,YAAY,WAAW;;EAEhC;;KAGU,wBAAwB,kBAAkB,UAAU,aAAa,KAC3E,QAAQ,mBAAmB,WAAW;EAGtC,SAAS,QAAQ;EACjB,eAAe,QAAQ;;UAGR,iBAAiB,kBAAkB,UAAU;;WAEnD,oBAAoB;;WAEpB,iBAAiB;;;;;EAK1B,SACE,OAAO,yBAAyB,WAAW,aAC1C,QAAQ,eAAe,WAAW;;;;;;;EAOrC,QACE,OAAO,wBAAwB,WAAW,aACzC,QAAQ,kBAAkB,WAAW;;;;;;;;;iBAU1B,gBAAgB,kBAAkB,UAAU,WAC1D,UAAU,uBAAuB,WAAW,aAC3C,iBAAiB,WAAW"}

package/dist/{define-agent-eval-C8V8sMqP.js → define-agent-eval-h-s-sI-v.js}

@@ -6,14 +6,15 @@ import { i as requiredPairedSampleSize, r as pairedMde } from "./power-and-mde-C
 import { r as pairRunRecords } from "./paired-arms-D-XRF_fy.js";
 import { o as pairedTTest, r as pairedBootstrap } from "./paired-tests-BHIhYVdu.js";
 import { r as welchsTTest } from "./baseline-BhPRQBVn.js";
+import "./query-CHmMP42p.js";
 import { r as observedSplitScore } from "./reward-nw2xZGZG.js";
-import { c as validateRunRecord, i as modelHasSnapshot } from "./run-record-D2lDdSAz.js";
-import { r as paretoChart } from "./summary-report-DW2bEpdB.js";
-import { $ as runEval, A as surfaceHash, f as runImprovementLoop, k as surfaceContentHash, nt as resolveRunDir, o as emitLoopProvenance, q as defaultProductionGate, s as loopProvenanceArgsFromResult, y as assertOptimizationResult } from "./llm-judge-BqqMS8t7.js";
-import { c as campaignCellTaskScore, l as campaignCellToRunRecord, o as campaignCellExecutionEvidence, s as campaignCellJudgeDimensions } from "./reward-hacking-C0x0xihA.js";
-import { S as inMemoryCampaignStorage, b as createRunCostLedger, x as fsCampaignStorage } from "./external-optimizer-subprocess-CKNb42oM.js";
+import { c as validateRunRecord, i as modelHasSnapshot } from "./run-record-BC0ebuRP.js";
+import { r as paretoChart } from "./summary-report-BI5hUtvK.js";
+import { N as surfaceContentHash, P as surfaceHash, Y as defaultProductionGate, c as emitLoopProvenance, it as resolveRunDir, l as loopProvenanceArgsFromResult, m as runImprovementLoop, tt as runEval, w as assertOptimizationResult } from "./llm-judge-BhasIPFT.js";
+import { c as campaignCellTaskScore, l as campaignCellToRunRecord, o as campaignCellExecutionEvidence, s as campaignCellJudgeDimensions } from "./reward-hacking-t4lB1yt8.js";
+import { S as inMemoryCampaignStorage, b as createRunCostLedger, x as fsCampaignStorage } from "./external-optimizer-subprocess-BIWbHpgD.js";
 import { t as powerPreflight } from "./power-preflight-DEw-uC7q.js";
-import { t as createHostedClient } from "./client-LIuo-KPv.js";
+import { t as createHostedClient } from "./client-CX7KqIdB.js";
 //#region src/contamination-guard.ts
 function checkCanaries(output, scenarios) {
 	const leaks = [];
@@ -987,7 +988,7 @@ function assertSelfImproveSearchMode(opts) {
 	const budget = opts.budget;
 	if (budget?.generations !== void 0 && budget.generations !== 1) throw new Error("selfImprove: method owns its rounds; budget.generations must be 1 when set");
 	if (budget?.populationSize !== void 0 && budget.populationSize !== 1) throw new Error("selfImprove: method owns its candidates; budget.populationSize must be 1 when set");
-	if (budget?.candidateConcurrency !== void 0 || budget?.maxImprovementShots !== void 0 || opts.analyzeGeneration !== void 0 || opts.findings !== void 0) throw new Error("selfImprove: candidateConcurrency, maxImprovementShots, analyzeGeneration, and findings apply only to proposer mode");
+	if (budget?.candidateConcurrency !== void 0 || budget?.maxImprovementShots !== void 0 || opts.analyzeGeneration !== void 0 || opts.findings !== void 0 || opts.selectParent !== void 0) throw new Error("selfImprove: candidateConcurrency, maxImprovementShots, analyzeGeneration, findings, and selectParent apply only to proposer mode");
 }
 function splitMethodPartitions(searchScenarios, explicitSelection, fraction) {
 	if (!Number.isFinite(fraction) || fraction <= 0 || fraction >= 1) throw new Error("selfImprove: budget.selectionFraction must be in (0, 1)");
@@ -1147,6 +1148,7 @@ async function runSelfImprove(opts, costLedger, startedAt, runDir, storage) {
 					maxConcurrency,
 					reps: budget.reps,
 					dispatchTimeoutMs: opts.dispatchTimeoutMs,
+					cellRetry: opts.cellRetry,
 					expectUsage,
 					costCeiling: budget.dollars
 				}),
@@ -1196,13 +1198,16 @@ async function runSelfImprove(opts, costLedger, startedAt, runDir, storage) {
 		maxConcurrency,
 		cellPlacement: opts.cellPlacement,
 		dispatchTimeoutMs: opts.dispatchTimeoutMs,
+		cellRetry: opts.cellRetry,
 		costLedger,
 		expectUsage,
 		labeledStore: opts.labeledStore,
 		captureSource: opts.captureSource,
 		analyzeGeneration: opts.analyzeGeneration,
 		findings: opts.findings,
-		selectionRankKey: opts.selectionRankKey
+		selectionRankKey: opts.selectionRankKey,
+		selectParent: opts.selectParent,
+		searchLedger: opts.searchLedger
 	});
 	const reportSplit = holdoutDeferred ? "search" : "holdout";
 	const reportBaselineCampaign = holdoutDeferred ? result.baselineCampaign : result.baselineOnHoldout;
@@ -1292,6 +1297,7 @@ async function runSelfImprove(opts, costLedger, startedAt, runDir, storage) {
 			...optimizationResult.durationMs === void 0 ? {} : { durationMs: optimizationResult.durationMs },
 			...optimizationResult.provenance === void 0 ? {} : { provenance: structuredClone(optimizationResult.provenance) }
 		} } : {},
+		...result.searchHistory ? { searchHistory: result.searchHistory } : {},
 		insight,
 		...power ? { power } : {},
 		raw: result
@@ -1491,4 +1497,4 @@ function requirePositiveInteger(value, field) {
 //#endregion
 export { summarizeExecution as a, analyzeRuns as i, SelfImproveRunError as n, checkCanaries as o, selfImprove as r, defineAgentEval as t };
 
-//# sourceMappingURL=define-agent-eval-C8V8sMqP.js.map
+//# sourceMappingURL=define-agent-eval-h-s-sI-v.js.map