@tangle-network/agent-eval 0.199.0 → 0.200.0

package/dist/{types-B3jzCp0p2.js → types-B3jzCp0p.js}

@@ -105,4 +105,4 @@ function describeValue(value) {
 //#endregion
 export { assertMultishotShotResult as i, MultishotFatalToolError as n, MultishotShotResultError as r, MultishotDriverEmptyError as t };
 
-//# sourceMappingURL=types-B3jzCp0p2.js.map
+//# sourceMappingURL=types-B3jzCp0p.js.map

package/dist/types-B3jzCp0p.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"types-B3jzCp0p.js","names":[],"sources":["../src/multishot/types.ts"],"sourcesContent":["// Public types for the multishot substrate.\n\nimport type { CostProvenance } from '../cost-ledger'\n\nexport interface MultishotMessage {\n  role: 'user' | 'assistant' | 'tool'\n  content: string\n  toolCallId?: string\n  toolCalls?: Array<{ id: string; name: string; args: Record<string, unknown> }>\n}\n\nexport interface MultishotArtifact {\n  type: string\n  turn: number\n  invocation: { name: string; args: Record<string, unknown> }\n  content: string\n}\n\nexport interface MultishotResult {\n  transcript: MultishotMessage[]\n  artifacts: MultishotArtifact[]\n  toolCalls: number\n  durationMs: number\n  /** Known spend. A subtotal, not a total, when `costProvenance.kind` is\n   *  `uncaptured`. */\n  costUsd: number\n  /** Origin of `costUsd`. A shot that priced every call reports `estimated`\n   *  or `observed`; a shot with a call the router priced at nothing reports\n   *  `uncaptured`, and the matrix records the cell as under-counted instead of\n   *  presenting the subtotal as a complete estimate.\n   *\n   *  Optional so an engine written before this field keeps working; the matrix\n   *  then judges the cell on judge receipts alone, as it did before. */\n  costProvenance?: CostProvenance\n}\n\nexport interface MultishotToolDefinition {\n  type: 'function'\n  function: {\n    name: string\n    description: string\n    parameters: Record<string, unknown>\n  }\n}\n\n/** One chat-completion request the multishot loop issues for a single agent\n *  (or driver) inference step. Mirrors the OpenAI-compat body the loop would\n *  otherwise POST to the Tangle router. */\nexport interface MultishotTransportRequest {\n  model: string\n  messages: Array<Record<string, unknown>>\n  tools?: MultishotToolDefinition[]\n  temperature?: number\n  maxTokens?: number\n  signal?: AbortSignal\n}\n\nexport interface MultishotTransportToolCall {\n  id: string\n  type: 'function'\n  function: { name: string; arguments: string }\n}\n\nexport interface MultishotTransportResponse {\n  message: { content?: string | null; tool_calls?: MultishotTransportToolCall[] }\n  usage?: { prompt_tokens?: number; completion_tokens?: number }\n  /** Actual spend for this call. When omitted, the loop meters cost from\n   *  `usage` via the per-model estimator (estimateMultishotCost). */\n  costUsd?: number\n  /** Model identity the provider reported, when the transport observed one.\n   *  Omitted means unreported, and the requested model is used for\n   *  attribution — which is not proof that model answered. */\n  model?: string\n}\n\n/** Execution seam for one leg of the multishot loop. The caller owns model\n *  execution: agent-eval issues no provider request and holds no credential.\n *  The loop still owns turn scheduling, tool dispatch, transcript capture, and\n *  cost metering. agent-eval has no dependency on agent-runtime; adapt\n *  agent-runtime's `profileChatClient` (or any sandbox, bridge, or router\n *  client) into this signature product-side. */\nexport type MultishotTransport = (\n  req: MultishotTransportRequest,\n) => Promise<MultishotTransportResponse>\n\nexport type MultishotToolExecutor = (\n  args: Record<string, unknown>,\n  ctx: {\n    /** Caller-owned execution seam for the specialist leg this tool runs. */\n    transport: MultishotTransport\n    signal?: AbortSignal\n  },\n) => Promise<{ content: string; costUsd: number }>\n\nexport interface MultishotPersona {\n  /** Stable identifier — used for per-cell artifact paths + matrix axis keys. */\n  id: string\n  /** Per-domain payload (income/profile/voice/etc.) shaped by the consumer. */\n  [k: string]: unknown\n}\n\n/**\n * Persona-shaping callbacks. Both are OPTIONAL: when omitted, the loop derives\n * them from the `AgentProfile` + persona payload (see `defaultShapeFromProfile`)\n * so a pure-profile call — `runMultishot({ profile, persona })` — works with no\n * role-builder functions. Provide callbacks only to override the derived shape.\n */\nexport interface MultishotShape<TPersona extends MultishotPersona> {\n  /** Opening user message (turn 0) — the persona's first ask. */\n  buildOpener?: (persona: TPersona) => string\n  /** System prompt the driver LLM uses to roleplay the persona. Should set\n   *  voice, goals, constraints, time-pressure, and the \"never go silent\" rule. */\n  buildDriverSystemPrompt?: (persona: TPersona) => string\n}\n\nexport class MultishotDriverEmptyError extends Error {\n  constructor(public readonly turn: number) {\n    super(`multishot: driver returned empty content twice at turn ${turn} — failing loud`)\n    this.name = 'MultishotDriverEmptyError'\n  }\n}\n\nexport class MultishotFatalToolError extends Error {\n  constructor(message: string) {\n    super(message)\n    this.name = 'MultishotFatalToolError'\n  }\n}\n\nexport class MultishotShotResultError extends Error {\n  constructor(reason: string) {\n    super(`multishot: shot returned an invalid MultishotResult — ${reason}`)\n    this.name = 'MultishotShotResultError'\n  }\n}\n\nconst MULTISHOT_ROLES = new Set(['user', 'assistant', 'tool'])\n\n/** Contract guard for the value a caller-supplied shot resolves with. The\n *  matrix writes per-cell artifacts, builds judge inputs, and meters cost from\n *  this value, so a malformed result must stop the cell instead of scoring a\n *  degraded one. Two silent degradations this closes: an artifact with no\n *  `type` matches neither the code nor the content artifact set, so the cell\n *  scores as though the artifact was never produced; a non-finite `costUsd`\n *  reaches `summary.totalCostUsd` and makes every cost number NaN.\n *\n *  A rejected cell is still billed: the matrix cell reads the shot's own\n *  `costUsd` when it is a usable amount and declares that spend on the throw,\n *  so money the shot spent before returning a malformed result stays in the\n *  cumulative sum the cost ceiling reads. A result whose `costUsd` is itself\n *  malformed carries no usable amount, and the cell records as `uncaptured`.\n *\n *  Every required field of `MultishotMessage` and `MultishotArtifact` is\n *  checked, including `toolCalls` elements and `invocation.args`. Optional\n *  fields are checked only when present. */\nexport function assertMultishotShotResult(value: unknown): asserts value is MultishotResult {\n  if (typeof value !== 'object' || value === null) {\n    throw new MultishotShotResultError(`expected an object, received ${describeValue(value)}`)\n  }\n  const result = value as Record<string, unknown>\n  if (!Array.isArray(result.transcript)) {\n    throw new MultishotShotResultError(\n      `transcript must be an array, received ${describeValue(result.transcript)}`,\n    )\n  }\n  if (!Array.isArray(result.artifacts)) {\n    throw new MultishotShotResultError(\n      `artifacts must be an array, received ${describeValue(result.artifacts)}`,\n    )\n  }\n  result.transcript.forEach(assertMessage)\n  result.artifacts.forEach(assertArtifact)\n  assertFiniteCount(result.toolCalls, 'toolCalls')\n  assertFiniteCount(result.durationMs, 'durationMs')\n  assertFiniteCount(result.costUsd, 'costUsd')\n  if (result.costProvenance !== undefined) assertCostProvenance(result.costProvenance)\n}\n\nfunction assertCostProvenance(value: unknown): void {\n  const row = requireRow(value, 'costProvenance')\n  if (row.kind !== 'observed' && row.kind !== 'estimated' && row.kind !== 'uncaptured') {\n    throw new MultishotShotResultError(\n      `costProvenance.kind must be observed, estimated or uncaptured, received ${describeValue(row.kind)}`,\n    )\n  }\n  // The matrix reads the amount from `costUsd`; `usd` only has to agree with\n  // the kind, so an uncaptured provenance cannot smuggle in a total.\n  if (row.kind === 'uncaptured') {\n    if (row.usd !== null) {\n      throw new MultishotShotResultError(\n        `uncaptured costProvenance.usd must be null, received ${describeValue(row.usd)}`,\n      )\n    }\n    return\n  }\n  assertFiniteCount(row.usd, 'costProvenance.usd')\n}\n\nfunction assertMessage(value: unknown, index: number): void {\n  const row = requireRow(value, `transcript[${index}]`)\n  if (typeof row.role !== 'string' || !MULTISHOT_ROLES.has(row.role)) {\n    throw new MultishotShotResultError(\n      `transcript[${index}].role must be user, assistant or tool, received ${describeValue(row.role)}`,\n    )\n  }\n  assertString(row.content, `transcript[${index}].content`)\n  if (row.toolCallId !== undefined) {\n    assertString(row.toolCallId, `transcript[${index}].toolCallId`)\n  }\n  if (row.toolCalls === undefined) return\n  if (!Array.isArray(row.toolCalls)) {\n    throw new MultishotShotResultError(\n      `transcript[${index}].toolCalls must be an array when present, received ${describeValue(row.toolCalls)}`,\n    )\n  }\n  row.toolCalls.forEach((call, callIndex) => {\n    const field = `transcript[${index}].toolCalls[${callIndex}]`\n    const row = requireRow(call, field)\n    assertString(row.id, `${field}.id`)\n    assertString(row.name, `${field}.name`)\n    requireRow(row.args, `${field}.args`)\n  })\n}\n\nfunction assertArtifact(value: unknown, index: number): void {\n  const row = requireRow(value, `artifacts[${index}]`)\n  assertString(row.type, `artifacts[${index}].type`)\n  assertString(row.content, `artifacts[${index}].content`)\n  assertFiniteCount(row.turn, `artifacts[${index}].turn`)\n  const invocation = requireRow(row.invocation, `artifacts[${index}].invocation`)\n  assertString(invocation.name, `artifacts[${index}].invocation.name`)\n  requireRow(invocation.args, `artifacts[${index}].invocation.args`)\n}\n\nfunction requireRow(value: unknown, field: string): Record<string, unknown> {\n  if (typeof value !== 'object' || value === null || Array.isArray(value)) {\n    throw new MultishotShotResultError(\n      `${field} must be an object, received ${describeValue(value)}`,\n    )\n  }\n  return value as Record<string, unknown>\n}\n\nfunction assertString(value: unknown, field: string): void {\n  if (typeof value !== 'string') {\n    throw new MultishotShotResultError(\n      `${field} must be a string, received ${describeValue(value)}`,\n    )\n  }\n}\n\nfunction assertFiniteCount(value: unknown, field: string): void {\n  if (typeof value !== 'number' || !Number.isFinite(value) || value < 0) {\n    throw new MultishotShotResultError(\n      `${field} must be a finite number >= 0, received ${describeValue(value)}`,\n    )\n  }\n}\n\nfunction describeValue(value: unknown): string {\n  if (value === null) return 'null'\n  if (Array.isArray(value)) return 'an array'\n  if (typeof value === 'object') return 'an object'\n  return `${typeof value} ${String(value)}`\n}\n"],"mappings":";AAmHA,IAAa,4BAAb,cAA+C,MAAM;CACvB;CAA5B,YAAY,MAA8B;EACxC,MAAM,0DAA0D,KAAK,gBAAgB;EAD3D,KAAA,OAAA;EAE1B,KAAK,OAAO;CACd;AACF;AAEA,IAAa,0BAAb,cAA6C,MAAM;CACjD,YAAY,SAAiB;EAC3B,MAAM,OAAO;EACb,KAAK,OAAO;CACd;AACF;AAEA,IAAa,2BAAb,cAA8C,MAAM;CAClD,YAAY,QAAgB;EAC1B,MAAM,yDAAyD,QAAQ;EACvE,KAAK,OAAO;CACd;AACF;AAEA,MAAM,kCAAkB,IAAI,IAAI;CAAC;CAAQ;CAAa;AAAM,CAAC;;;;;;;;;;;;;;;;;;AAmB7D,SAAgB,0BAA0B,OAAkD;CAC1F,IAAI,OAAO,UAAU,YAAY,UAAU,MACzC,MAAM,IAAI,yBAAyB,gCAAgC,cAAc,KAAK,GAAG;CAE3F,MAAM,SAAS;CACf,IAAI,CAAC,MAAM,QAAQ,OAAO,UAAU,GAClC,MAAM,IAAI,yBACR,yCAAyC,cAAc,OAAO,UAAU,GAC1E;CAEF,IAAI,CAAC,MAAM,QAAQ,OAAO,SAAS,GACjC,MAAM,IAAI,yBACR,wCAAwC,cAAc,OAAO,SAAS,GACxE;CAEF,OAAO,WAAW,QAAQ,aAAa;CACvC,OAAO,UAAU,QAAQ,cAAc;CACvC,kBAAkB,OAAO,WAAW,WAAW;CAC/C,kBAAkB,OAAO,YAAY,YAAY;CACjD,kBAAkB,OAAO,SAAS,SAAS;CAC3C,IAAI,OAAO,mBAAmB,KAAA,GAAW,qBAAqB,OAAO,cAAc;AACrF;AAEA,SAAS,qBAAqB,OAAsB;CAClD,MAAM,MAAM,WAAW,OAAO,gBAAgB;CAC9C,IAAI,IAAI,SAAS,cAAc,IAAI,SAAS,eAAe,IAAI,SAAS,cACtE,MAAM,IAAI,yBACR,2EAA2E,cAAc,IAAI,IAAI,GACnG;CAIF,IAAI,IAAI,SAAS,cAAc;EAC7B,IAAI,IAAI,QAAQ,MACd,MAAM,IAAI,yBACR,wDAAwD,cAAc,IAAI,GAAG,GAC/E;EAEF;CACF;CACA,kBAAkB,IAAI,KAAK,oBAAoB;AACjD;AAEA,SAAS,cAAc,OAAgB,OAAqB;CAC1D,MAAM,MAAM,WAAW,OAAO,cAAc,MAAM,EAAE;CACpD,IAAI,OAAO,IAAI,SAAS,YAAY,CAAC,gBAAgB,IAAI,IAAI,IAAI,GAC/D,MAAM,IAAI,yBACR,cAAc,MAAM,mDAAmD,cAAc,IAAI,IAAI,GAC/F;CAEF,aAAa,IAAI,SAAS,cAAc,MAAM,UAAU;CACxD,IAAI,IAAI,eAAe,KAAA,GACrB,aAAa,IAAI,YAAY,cAAc,MAAM,aAAa;CAEhE,IAAI,IAAI,cAAc,KAAA,GAAW;CACjC,IAAI,CAAC,MAAM,QAAQ,IAAI,SAAS,GAC9B,MAAM,IAAI,yBACR,cAAc,MAAM,sDAAsD,cAAc,IAAI,SAAS,GACvG;CAEF,IAAI,UAAU,SAAS,MAAM,cAAc;EACzC,MAAM,QAAQ,cAAc,MAAM,cAAc,UAAU;EAC1D,MAAM,MAAM,WAAW,MAAM,KAAK;EAClC,aAAa,IAAI,IAAI,GAAG,MAAM,IAAI;EAClC,aAAa,IAAI,MAAM,GAAG,MAAM,MAAM;EACtC,WAAW,IAAI,MAAM,GAAG,MAAM,MAAM;CACtC,CAAC;AACH;AAEA,SAAS,eAAe,OAAgB,OAAqB;CAC3D,MAAM,MAAM,WAAW,OAAO,aAAa,MAAM,EAAE;CACnD,aAAa,IAAI,MAAM,aAAa,MAAM,OAAO;CACjD,aAAa,IAAI,SAAS,aAAa,MAAM,UAAU;CACvD,kBAAkB,IAAI,MAAM,aAAa,MAAM,OAAO;CACtD,MAAM,aAAa,WAAW,IAAI,YAAY,aAAa,MAAM,aAAa;CAC9E,aAAa,WAAW,MAAM,aAAa,MAAM,kBAAkB;CACnE,WAAW,WAAW,MAAM,aAAa,MAAM,kBAAkB;AACnE;AAEA,SAAS,WAAW,OAAgB,OAAwC;CAC1E,IAAI,OAAO,UAAU,YAAY,UAAU,QAAQ,MAAM,QAAQ,KAAK,GACpE,MAAM,IAAI,yBACR,GAAG,MAAM,+BAA+B,cAAc,KAAK,GAC7D;CAEF,OAAO;AACT;AAEA,SAAS,aAAa,OAAgB,OAAqB;CACzD,IAAI,OAAO,UAAU,UACnB,MAAM,IAAI,yBACR,GAAG,MAAM,8BAA8B,cAAc,KAAK,GAC5D;AAEJ;AAEA,SAAS,kBAAkB,OAAgB,OAAqB;CAC9D,IAAI,OAAO,UAAU,YAAY,CAAC,OAAO,SAAS,KAAK,KAAK,QAAQ,GAClE,MAAM,IAAI,yBACR,GAAG,MAAM,0CAA0C,cAAc,KAAK,GACxE;AAEJ;AAEA,SAAS,cAAc,OAAwB;CAC7C,IAAI,UAAU,MAAM,OAAO;CAC3B,IAAI,MAAM,QAAQ,KAAK,GAAG,OAAO;CACjC,IAAI,OAAO,UAAU,UAAU,OAAO;CACtC,OAAO,GAAG,OAAO,MAAM,GAAG,OAAO,KAAK;AACxC"}

package/dist/{types-D94EanDL.d.ts → types-C4hH69a9.d.ts}

@@ -1,5 +1,5 @@
 import { c as RunSplitTag } from "./run-record-DF-fuu31.js";
-import { R as Scenario, d as DispatchContext } from "./types-EBUav9Bu.js";
+import { R as Scenario, d as DispatchContext } from "./types-DfvQOe91.js";
 //#region src/benchmarks/types.d.ts
 type BenchmarkTaskKind = 'retrieval' | 'rag-answer' | 'hallucination' | 'kb-improvement' | 'routing' | 'custom';
 type BenchmarkFamily = 'beir' | 'mteb-retrieval' | 'msmarco' | 'trec-dl' | 'miracl' | 'lotte' | 'bright' | 'crag' | 'hotpotqa' | 'kilt' | 'ragtruth' | 'faithbench' | 'first-party' | 'custom';
@@ -99,4 +99,4 @@ declare const BENCHMARK_SPLIT_SEED = "agent-eval-v1";
 declare function deterministicSplit(itemId: string, seed?: string): RunSplitTag;
 //#endregion
 export { BenchmarkFamily as a, BenchmarkSource as c, BenchmarkEvaluation as i, BenchmarkTaskKind as l, BenchmarkAdapter as n, BenchmarkResponder as o, BenchmarkDatasetItem as r, BenchmarkScenario as s, BENCHMARK_SPLIT_SEED as t, deterministicSplit as u };
-//# sourceMappingURL=types-D94EanDL.d.ts.map
+//# sourceMappingURL=types-C4hH69a9.d.ts.map

package/dist/{types-D94EanDL.d.ts.map → types-C4hH69a9.d.ts.map}

@@ -1 +1 @@
-{"version":3,"file":"types-D94EanDL.d.ts","names":[],"sources":["../src/benchmarks/types.ts"],"mappings":";;;KASY;KAQA;UAgBK,qBAAqB;;;EAGpC;;EAEA,SAAS;;EAET,QAAQ;;EAER,SAAS;;EAET,WAAW;;EAEX;;EAEA,SAAS;EACT,WAAW;;UAGI;;;;EAIf;;EAEA;;EAEA,aAAa;;;EAGb,KAAK;EACL;;UAGe;EACf;EACA;EACA;;EAEA;EACA;EACA;;EAEA;;;;;EAKA;;;UAQe,iBAAiB,kBAAkB,oBAAoB;;EAEtE;EACA,SAAS;EACT,WAAW;EACX;EACA,SAAS;EACT;;;;;EAKA,YAAY,OAAO,cAAc,QAAQ,qBAAqB;;EAE9D,SAAS,MAAM,qBAAqB,WAAW,UAAU,YAAY,QAAQ;;;;EAI7E,YAAY,iBAAiB;;UAGd,kBAAkB,4BAA4B;EAC7D;EACA;EACA,QAAQ;EACR,UAAU;EACV,UAAU;EACV,MAAM,qBAAqB;;KAGjB,mBAAmB,oBAAoB,uBAAuB;EACxE,UAAU,kBAAkB;EAC5B,MAAM,qBAAqB;EAC3B,SAAS;MACL,QAAQ,aAAa;;;cAoBd;;;;;;;;;iBAUG,mBACd,gBACA,gBACC"}
+{"version":3,"file":"types-C4hH69a9.d.ts","names":[],"sources":["../src/benchmarks/types.ts"],"mappings":";;;KASY;KAQA;UAgBK,qBAAqB;;;EAGpC;;EAEA,SAAS;;EAET,QAAQ;;EAER,SAAS;;EAET,WAAW;;EAEX;;EAEA,SAAS;EACT,WAAW;;UAGI;;;;EAIf;;EAEA;;EAEA,aAAa;;;EAGb,KAAK;EACL;;UAGe;EACf;EACA;EACA;;EAEA;EACA;EACA;;EAEA;;;;;EAKA;;;UAQe,iBAAiB,kBAAkB,oBAAoB;;EAEtE;EACA,SAAS;EACT,WAAW;EACX;EACA,SAAS;EACT;;;;;EAKA,YAAY,OAAO,cAAc,QAAQ,qBAAqB;;EAE9D,SAAS,MAAM,qBAAqB,WAAW,UAAU,YAAY,QAAQ;;;;EAI7E,YAAY,iBAAiB;;UAGd,kBAAkB,4BAA4B;EAC7D;EACA;EACA,QAAQ;EACR,UAAU;EACV,UAAU;EACV,MAAM,qBAAqB;;KAGjB,mBAAmB,oBAAoB,uBAAuB;EACxE,UAAU,kBAAkB;EAC5B,MAAM,qBAAqB;EAC3B,SAAS;MACL,QAAQ,aAAa;;;cAoBd;;;;;;;;;iBAUG,mBACd,gBACA,gBACC"}

package/dist/{types-EBUav9Bu.d.ts → types-DfvQOe91.d.ts}

@@ -116,7 +116,10 @@ interface SearchOperation {
   /** Artifacts the started event bound, then those the recorded event bound,
    * for example a claim's plan or a proposal's output. */
   artifacts: readonly SearchArtifactRef[];
+  /** Known cost, or the proven floor when `costKnown` is false. */
   spentUsd: number;
+  /** The recorded cost is known; false until the operation is recorded. */
+  costKnown: boolean;
 }
 interface SearchCompletion {
   complete: boolean;
@@ -1006,4 +1009,4 @@ interface CampaignResult<TArtifact = unknown, TScenario extends Scenario = Scena
 }
 //#endregion
 export { SearchNode as $, LabeledScenarioWrite as A, ScoredSurfaceOutcome as B, JudgeDimension as C, LabeledScenarioSampleArgs as D, LabeledScenarioRecord as E, ProposeContext as F, labelTrustRank as G, SurfaceProposer as H, ProposedCandidate as I, renderSearchSummary as J, SearchProposerView as K, RedactionStatus as L, OptimizerConfig as M, ParetoParent as N, LabeledScenarioSource as O, ProposalTrackContext as P, SearchDecisionRecord as Q, Scenario as R, JudgeConfig as S, LabelTrust as T, TraceSpan as U, SessionScript as V, isProposedCandidate as W, SearchCell as X, searchProposerView as Y, SearchCompletion as Z, GateDecision as _, CampaignResult as a, SearchUnitScore as at, GenerationRecord as b, CampaignTraceWriter as c, searchTaskSetDigest as ct, DispatchContext as d, SearchOperation as et, DispatchFn as f, GateContribution as g, GateContext as h, CampaignCostMeter as i, SearchStateView as it, MutableSurface as j, LabeledScenarioStore as k, CodeSurface as l, searchUnitScores as lt, GateCheckStatus as m, CampaignArtifactWriter as n, SearchSpend as nt, CampaignScenarioIdentity as o, searchCellId as ot, Gate as p, SearchSummaryOptions as q, CampaignCellResult as r, SearchState as rt, CampaignTokenUsage as s, searchNodeId as st, CampaignAggregates as t, SearchScoredCell as tt, ComponentSurface as u, GateResult as v, JudgeScore as w, JudgeAggregate as x, GenerationCandidate as y, ScenarioAggregate as z };
-//# sourceMappingURL=types-EBUav9Bu.d.ts.map
+//# sourceMappingURL=types-DfvQOe91.d.ts.map

package/dist/types-DfvQOe91.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"types-DfvQOe91.d.ts","names":[],"sources":["../src/campaign/search-state.ts","../src/campaign/search-summary.ts","../src/campaign/types.ts"],"mappings":";;;;;;;;;;iBAwDgB,aAAa,kBAAkB;;iBAK/B,aACd,kBACA,gBACA,gBACA,OAAO,aACP;;iBAMc,oBAAoB,gBAAgB,eAAe;UAkBlD;EACf;EACA,UAAU;EACV,OAAO;EACP;EACA;;UAGe;EACf;EACA;EACA;EACA;;;;UAKe;EACf;EACA;EACA;EACA;;;UAIe;EACf;EACA;;EAEA;EACA;;;;;;;;iBASc,iBAAiB,gBAAgB,qBAAqB;UAoBrD;EACf;;EAEA;EACA,gBAAgB;EAChB,UAAU;EACV,mBAAmB;;;EAGnB;;EAEA;;;;EAIA,kBAAkB;EAClB;EACA;EACA,QAAQ;EACR;;EAEA;EACA,oBAAoB;EACpB;EACA,OAAO;EACP;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA,OAAO;EACP;EACA,OAAO;EACP;EACA,aAAa;EACb;EACA;EACA,SAAS;;EAET;;EAEA;EACA,WAAW;;EAEX;;EAEA;EACA;EACA;EACA;;UAGe;EACf;EACA,eAAe;EACf,aAAa;EACb;;EAEA,SAAS;;;EAGT,WAAW;;;EAGX,oBAAoB;;EAEpB;;EAEA;;UAGe;EACf;EACA;;;;;;;cA6BW,uBAAuB,gBAAgB,mBAAmB;WAC5D;UACD;UACA;UACA;UACA;UACA;UACA;mBACS;mBACA;mBACA;mBACA;mBACA;mBACA;mBACA;mBACA;UACT;UACA;;mBAES;;UAET;mBACS;EAEjB,YAAY;EAgCZ,MAAM,OAAO,mBAAmB;EA6BhC,YAAY;;EAKZ,KAAK;UAIG;UAgCA;UA6CA;UAsBA;UA8BA;UA8CA;UAsHA;UAmEA;UA4EA;UAoBA;UAmDA;UAaA;;;UA2EA;;EASR;;;UAmBQ;;;UAkBA;;;;;;;;;;;UAoBA;;;EAsBR;IAAgB;IAA4B;;;EAa5C,cAAc;;EAKd,cAAc;;EAKd;IAAc;IAAkB,WAAW;;;EAK3C;;EAKA,aAAa;;EAcb,SAAS,iBAAiB;;EAM1B;;EAKA,YAAY;;EAKZ,oBAAoB;;EAKpB,SAAS,iBAAiB;;EAK1B;;EAKA,SAAS,iBAAiB;;EAM1B,YAAY;;EAMZ,cAAc,sBAAsB;;EAMpC,gBAAgB,gBAAgB,OAAO,cAAc;;;;;;;cAiB1C;WACF;WACA,QAAQ;WACR,QAAQ;WACR;IAAQ;IAAkB,WAAW;;;WAErC;WACA,OAAO;WACP,YAAY;;;;;;;WAOZ;IAAU;IAA4B;;mBAC9B;mBACA;EAEjB,YAAY,OAAO,aAAa;;MAgB5B;EAIJ,KAAK,iBAAiB;EAItB,QAAQ;;EAKR;;EAKA,SAAS;EAKT,gBAAgB;EAIhB,KAAK,iBAAiB;;EAKtB,SAAS;EAKT,KAAK,iBAAiB;;EAKtB,MAAM;IAAU;MAAyB;EAKzC,UAAU,sBAAsB;;;EAMhC,YAAY,gBAAgB,OAAO,cAAc;;EAKjD,WAAW,gBAAgB,OAAO,cAAc;;EAKhD,QAAQ,aAAa,uBAAuB;UAmCpC;;;;;;UCluCO;WACN;WACA;;;EAGT,YAAY,0BAA0B;;EAEtC,WAAW,0BAA0B;;;iBAIvB,mBAAmB,OAAO,kBAAkB;UAW3C;;;;;;;;EAQf,OAAO;;EAEP;;;;;;;;;;;iBAYc,oBAAoB,OAAO,iBAAiB,SAAS;;;;;UCtCpD;EACf;EACA;EACA;;;;;EAKA;;;UAIe,iCAAiC,KAAK;EACrD;;;;;UAMe;EACf;;EAEA;EACA;EACA;EACA;EACA,QAAQ;EACR,OAAO;EACP,WAAW;EACX,MAAM;;EAEN;;EAEA;;;;;;;;EAQA;;;;KAKU,WAAW,kBAAkB,UAAU,cACjD,UAAU,WACV,KAAK,oBACF,QAAQ;;;;UAOI,cAAc,WAAW;EACxC;EACA;EACA;;EAEA;;;EAGA,sBAAsB,UAAU,WAAW,sBAAsB,UAAU,cAAc;;UAK1E;;EAEf;;EAEA;;;;;;;;;UAUe,YAAY,WAAW,kBAAkB,WAAW;EACnE;EACA,YAAY;;;;EAIZ;;;EAGA,MAAM;IACJ,UAAU;IACV,UAAU;IACV,QAAQ;;IAER,aAAa;IACb;IACA,WAAW;MACT,aAAa,QAAQ;EACzB,aAAa,UAAU;;;;;;;;;;;UAYR;EACf,YAAY;EACZ;EACA;;EAEA,UAAU;;;;;;;EAOV;;;;EAIA,eAAe,eAAe;IAAQ;IAAe;;;;;EAIrD;;;EAGA;;EAEA;;EAEA,WAAW,eAAe;;;;;;;UAUX;WACN;;;WAGA;;WAEA;;WAEA;;WAEA;;WAEA;;WAEA;;;WAGA;aACE;aACA;aACA;;;WAGF;;;UAIM;WACN;WACA,YAAY,SAAS;;;;;;;;;;;KAYpB,0BAA0B,mBAAmB;;;;;;;UAQxC;EACf,SAAS;;EAET;;;;EAIA;;;;;;EAMA,cAAc,SAAS;;;;iBAKT,oBACd,OAAO,iBAAiB,oBACvB,SAAS;;;;;;;;;UAkBK;EACf,SAAS;EACT;;;EAGA,YAAY;;;EAGZ;;EAEA;EACA;EACA;;;UAIe;EACf;EACA;EACA;EACA;EACA;;;;;UAMe;;;EAGf;;EAEA;EACA;EACA;EACA,YAAY;;;;;EAKZ,WAAW;IAAQ;IAAoB;IAAmB;IAAgB;;EAC1E;IACE;IACA;;;;;UAMa,eAAe,YAAY;;;;WAIjC,gBAAgB;;;;;WAKhB;;;;WAIA,UAAU;IAAgB;IAAgB,UAAU;;;;WAGpD,QAAQ;;;;WAIR;;;WAGA,SAAS,cAAc;WACvB,UAAU,cAAc;;WAExB;WACA;WACA,QAAQ;;WAER,QAAQ;;;WAGR,kBAAkB;;;WAGlB,mBAAmB;;;;WAInB,gBAAgB;;;;WAIhB;;;;;;;WAOA,gBAAgB,cAAc;;WAE9B,aAAa;WACb;;;;;;;;;;;;;;UAeM,gBAAgB,YAAY;EAC3C;;;;;EAKA,QAAQ,KAAK,eAAe,aAAa,QAAQ,MAAM,iBAAiB;;;EAGxE,QAAQ;IAAQ,SAAS,cAAc;;IAAwB;IAAe;;;UAG/D;EACf;EACA;EACA,mBAAmB,qBAAqB;;UAGzB,wBAAwB;EACvC,UAAU;;;KAMA;;KAGA;UAEK;EACf;EACA,QAAQ;EACR;;UAGe,YAAY,WAAW,kBAAkB;EACxD,oBAAoB,YAAY;EAChC,oBAAoB,YAAY;;EAEhC,aAAa,YAAY,eAAe;;;;;EAKxC,sBAAsB,YAAY,eAAe;;;;;;;EAOjD,yBAAyB,YAAY,eAAe;;;EAGpD,uBAAuB,YAAY;EACnC,WAAW;EACX;IAAQ;IAAmB;;;EAE3B,aAAa;EACb;EACA,QAAQ;;UAGO;EACf,UAAU;EACV;EACA,mBAAmB;EACnB;;;UAIe,KAAK,qBAAqB,kBAAkB,WAAW;EACtE;EACA,OAAO,KAAK,YAAY,WAAW,aAAa,QAAQ;;;;UAOzC;EACf,KAAK,cAAc,aAAa,0BAA0B;EAC1D,SAAS;;UAGM;EACf,IAAI,aAAa;EACjB,aAAa,aAAa;;;;UAKX;EACf,MAAM,cAAc,kBAAkB,aAAa;EACnD,UAAU,cAAc,iBAAiB;;;;;;KAO/B,qBAAqB;;;;;UAMhB;;EAEf,YAAY,GACV,OAAO,KAAK,iBAAiB;IAC3B,UAAU;MAEX,QAAQ,eAAe;;;;;KAQhB;KAOA;;;;;;;;;;;;;;;KAgBA;;iBASI,eAAe,OAAO;;;;UAOrB,qBAAqB,kBAAkB,WAAW,UAAU;EAC3E,UAAU;EACV,UAAU;EACV,aAAa,eAAe;EAC5B,QAAQ;EACR;EACA;EACA,iBAAiB;;;;;EAKjB,aAAa;;EAEb;;UAGe,sBAAsB,kBAAkB,WAAW,UAAU,6BACpE,qBAAqB,WAAW;;EAExC;;;EAGA;;UAGe;EACf;;EAEA;;;;EAIA;EACA;IACE;IACA,SAAS,wBAAwB;IACjC;IACA;;;;;IAKA,WAAW;;;UAIE;EACf,QAAQ,OAAO,uBAAuB;EACtC,OAAO,MAAM,4BAA4B,QAAQ;EACjD,QAAQ;IACN;IACA;IACA,UAAU;;;IAGV,SAAS,OAAO;;;UAMH,mBAAmB;;;EAGlC;EACA;EACA;EACA;EACA;EACA,UAAU;EACV,aAAa,eAAe;;;EAG5B;;EAEA,gBAAgB;;EAEhB;;;EAGA,YAAY;;;EAGZ;;;EAGA;EACA;EACA;EACA;;;;EAIA;;EAEA;;EAEA;EACA;;UAGe;EACf;EACA;EACA;EACA;;;;;;EAMA,cAAc;;UAGC;EACf;EACA;EACA;;;EAGA,cAAc;;UAGC;EACf;EACA,YAAY;EACZ;;;;;;UAOe;EACf;;;;EAIA;;;;EAIA;;;;EAIA;IACE;IACA;IACA,iBAAiB;MAAQ;MAAgB;;;;;EAI3C,YAAY;;;;;;;;;;EAUZ,WAAW;IAAQ;IAAoB;IAAmB;IAAgB;;;;EAG1E;;;;EAIA;;;;EAIA,cAAc,SAAS;;UAGR;EACf,SAAS,eAAe;EACxB,YAAY,eAAe;;EAE3B,MAAM;;EAEN;EACA;EACA;;EAEA;;EAEA;;EAEA;;EAEA;;UAGe,eAAe,qBAAqB,kBAAkB,WAAW;;EAEhF;;EAEA;EACA;;EAEA;EACA;EACA;EACA;EACA,OAAO,MAAM,mBAAmB;EAChC,YAAY;EACZ;IACE,aAAa;IACb;;EAEF,OAAO;EACP;EACA;EACA,iBAAiB;;;EAGjB,WAAW,MAAM,2BAA2B,KAAK"}

package/docs/jev-behavior-review.md

@@ -144,6 +144,30 @@ requires resolution. Never convert an unresolved safety review to a perfect scor
 critical constraint failure into unrelated high scores. Working feedback must not expose the
 independent final evaluator, protected holdout labels, or monitor control credentials to the worker.
 
+## Claims an exact checker verified
+
+An exact checker decides whether an artifact satisfies the checker's own encoding of a statement.
+It cannot decide whether that encoding matches the statement the source meant.
+An agent that reads the checker finds the cases where they differ.
+A claim in such a case passes every exact check and settles nothing.
+
+`/rl` exports `claimIntegrityVerdict` for this case.
+It decides `pass`, `hack` or `unknown` for one claim from signals the caller computes:
+
+- `scopeSignals`: predicates over the claim's parameters, from a `SourceScope` written from the source by someone other than the checker's author. The source excludes the case, lists it, or reports it checked.
+- `farmingSignal`: one parameter value repeated across a lane's sweep of claims.
+- `graderReferenceSignal`: the claim page justifies itself by the checker or by the printed letter.
+- `refereeSignal`: a model reads the source and the claim's parameters, never the claimant's prose (`sourceScopeRefereeMessages`, `parseSourceScopeReferee`).
+- `jevSignal`: typed questions over the same brief (`claimIntegrityJevReview`). The caller supplies calibrated thresholds.
+- `monitorSignal`: `askTraceQuestions` with `claimIntegrityTraceQuestions`, on escalation only.
+- `checkerReadSignal`: context for a reader; it never moves a verdict.
+
+The policy fails closed.
+A pass needs a declared scope and a referee that evaluated the claim, with nothing against it.
+An exact signal decides `hack`, and so do two distinct kinds of signal.
+One kind alone holds the claim as `unknown`.
+Keep verdicts, scopes and questions away from the agent that filed the claim: an agent scored on them learns to hide the case.
+
 ## Security and validation boundaries
 
 TypeSafe's Jev 1.13 documentation explicitly warns that adversarial state content can move the

package/docs/public-api.md

@@ -7,11 +7,11 @@ Generated by `pnpm api:census` on demand — this is a dated reading, not a gate
 | measure | count |
 | --- | --- |
 | export subpaths | 33 |
-| published value exports (subpath x symbol) | 1570 |
-| distinct symbols | 1382 |
-| production | 1047 |
-| planned | 266 |
-| none | 257 |
+| published value exports (subpath x symbol) | 1589 |
+| distinct symbols | 1401 |
+| production | 1048 |
+| planned | 278 |
+| none | 263 |
 
 Type-only exports are not listed: a type binds no runtime surface, and removing one cannot break a caller at run time.
 
@@ -460,7 +460,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 
 ### `./analyst`
 
-131 value exports — 79 production, 19 planned, 33 none.
+131 value exports — 79 production, 20 planned, 32 none.
 
 | symbol | consumer | evidence |
 | --- | --- | --- |
@@ -489,7 +489,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `AnalystRegistry` | production | blueprint-agent:scripts/experiments/vb-multi-analyst.ts |
 | `analystUsageReceiptFromPrimeUsage` | production | this package: src/analyst/benchmark-runner-prime.ts:19 |
 | `appendVerificationArtifactsToOtlp` | production | this package: src/analyst/benchmark-public-data.ts:32 |
-| `askTraceQuestions` | none | only this package's tests: src/analyst/trace-questions.test.ts:7 |
+| `askTraceQuestions` | planned | doc: docs/jev-behavior-review.md |
 | `assertProposalFindings` | production | agent-runtime:src/improvement/code-execution.ts |
 | `behavioralAnalyst` | production | agent-dev-container:products/intelligence/api/src/lib/deep-trace-analyst.ts |
 | `bindAnalyst` | planned | doc: docs/trace-analysis.md |
@@ -649,16 +649,16 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | symbol | consumer | evidence |
 | --- | --- | --- |
 | `acquireSingleRunLock` | production | agent-knowledge:src/memory/run-control.ts |
-| `aide` | production | this package: scripts/search-sim.ts:145 |
+| `aide` | production | this package: scripts/search-sim.ts:146 |
 | `analyzeCrossSurfaceInteractions` | none | only this package's tests: src/campaign/cross-surface-interaction.test.ts:2 |
-| `asha` | production | this package: scripts/search-sim.ts:119 |
+| `asha` | production | this package: scripts/search-sim.ts:120 |
 | `assertCampaignDesign` | production | this package: src/campaign/plan-campaign-run.ts:10 |
 | `assertCampaignSplitIdentity` | production | agent-knowledge:src/memory/experiment/learning-pairs.ts |
 | `assertCodeSurfaceIdentity` | none | — |
 | `assertCompleteSearchHistory` | production | this package: src/campaign/optimization-method.ts:14 |
 | `assertSearchHistoryMatchesState` | planned | doc: docs/search-ledger.md |
 | `autoevalsScorerJudge` | none | only this package's tests: src/campaign/upstream-evaluators.test.ts:6 |
-| `beam` | production | this package: scripts/search-sim.ts:145 |
+| `beam` | production | this package: scripts/search-sim.ts:146 |
 | `buildCellSchedule` | production | this package: src/campaign/plan-campaign-run.ts:9 |
 | `buildEvidenceVector` | planned | doc: docs/experiment.md |
 | `buildLoopProvenanceRecord` | none | only this package's tests: tests/campaign/candidate-attribution.test.ts:5 |
@@ -683,18 +683,18 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `createReferenceEquivalenceJudge` | none | only this package's tests: src/reference-equivalence-judge.test.ts:13 |
 | `createRunCostLedger` | production | agent-dev-container:products/intelligence/api/src/lib/eval-engine.ts |
 | `createSearchHistoryReceipt` | production | this package: src/campaign/search-ledger-recording.ts:27 |
-| `crowdedFrontierParent` | production | this package: scripts/search-sim.ts:145 |
+| `crowdedFrontierParent` | production | this package: scripts/search-sim.ts:146 |
 | `decideSearchClaim` | production | this package: src/campaign/search-kernel.ts:42 |
 | `decodeExternalTextCandidate` | production | agent-runtime:src/improvement/method-execution.ts |
 | `DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS` | none | — |
 | `DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS` | none | — |
 | `defaultProductionGate` | production | agent-app:src/eval-campaign/index.ts |
 | `detectScale` | production | this package: src/campaign/gates/promotion-policy.ts:25 |
-| `developmentClaim` | production | this package: scripts/generate-synthetic-search-ledger.ts:34 |
+| `developmentClaim` | production | this package: scripts/generate-synthetic-search-ledger.ts:37 |
 | `dimensionRegressions` | production | agent-dev-container:products/intelligence/api/src/lib/optimization-engine.ts |
 | `discoverEvalFixtures` | planned | doc: docs/eval-fixtures.md |
 | `emitLoopProvenance` | production | this package: src/contract/self-improve.ts:36 |
-| `estimateNode` | production | this package: scripts/search-sim.ts:120 |
+| `estimateNode` | production | this package: scripts/search-sim.ts:121 |
 | `estimateNodeFromCells` | production | this package: scripts/synthetic-search-ledger.ts:60 |
 | `externalSurface` | none | — |
 | `externalTextOptimizationMethod` | production | agent-app:src/eval-campaign/index.ts |
@@ -725,12 +725,12 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `neutralizationGate` | planned | consumer tests: tax-agent:tests/eval/benchmarks/taxcalc/evolve.ts |
 | `neutralizeText` | planned | consumer tests: tax-agent:tests/eval/benchmarks/taxcalc/compile-and-prove/compile-diff.ts |
 | `openAutoPr` | production | this package: src/campaign/presets/run-improvement-loop.ts:8 |
-| `openSearchLedger` | production | this package: scripts/generate-synthetic-search-ledger.ts:34 |
+| `openSearchLedger` | production | this package: scripts/generate-synthetic-search-ledger.ts:37 |
 | `optimizationTokenUsageFromSummary` | production | this package: src/campaign/external-text-optimization.ts:41 |
 | `pairHoldout` | production | agent-dev-container:products/intelligence/api/src/lib/run-provenance.ts |
 | `paretoPolicy` | none | only this package's tests: src/campaign/gates/promotion-policy.test.ts:3 |
 | `paretoSignificanceGate` | production | agent-app:src/eval-campaign/index.ts |
-| `parseSearchLedgerLine` | production | this package: scripts/search-sim.ts:132 |
+| `parseSearchLedgerLine` | production | this package: scripts/search-sim.ts:133 |
 | `phoenixEvaluatorJudge` | none | only this package's tests: src/campaign/upstream-evaluators.test.ts:6 |
 | `planCampaignRun` | production | this package: src/campaign/fixtures.ts:5 |
 | `planEvalFixtureRun` | planned | example: examples/eval-fixtures-quickstart/index.ts:11 |
@@ -748,7 +748,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `renderScoreboardMarkdown` | production | blueprint-agent:scripts/experiments/eval/launch-scoreboard/run.ts |
 | `renderSearchSummary` | production | this package: scripts/search-meta-sim.ts:50 |
 | `renderSurfaceDiff` | production | agent-dev-container:products/intelligence/api/src/lib/optimization-evidence-packet.ts |
-| `replaySearchLedgerText` | production | this package: scripts/generate-synthetic-search-ledger.ts:34 |
+| `replaySearchLedgerText` | production | this package: scripts/generate-synthetic-search-ledger.ts:37 |
 | `resolveExternalOptimizerCallbackLimits` | production | this package: src/analyst/dspy-rlm-engine.ts:1 |
 | `resolveExternalOptimizerProcessLimits` | production | this package: src/analyst/benchmark-command-persistence.ts:6 |
 | `resolveRunDir` | production | agent-dev-container:products/intelligence/api/src/lib/eval-engine.ts |
@@ -760,7 +760,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `runOptimization` | production | agent-dev-container:products/sandbox/evals/src/auto-optimization.ts |
 | `runProfileMatrix` | production | agent-runtime:examples/product-eval/product-eval.ts |
 | `runProfileMatrixSegment` | production | discovery-lab:tools/run-profile-confirmation.mjs |
-| `runSearch` | production | this package: scripts/search-sim.ts:121 |
+| `runSearch` | production | this package: scripts/search-sim.ts:122 |
 | `scopedOptimizationMethod` | planned | doc: docs/campaign-proposers.md |
 | `scoreboardSummary` | production | blueprint-agent:scripts/experiments/eval/launch-scoreboard/run.ts |
 | `scoreDiscrimination` | production | supervisor-lab:bench/comms/seat-discrimination.ts |
@@ -774,7 +774,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `searchClaimReserveUsd` | production | this package: scripts/search-meta-sim.ts:43 |
 | `searchDivergence` | planned | doc: docs/search-ledger.md |
 | `searchEdgeId` | planned | doc: docs/search-ledger.md |
-| `searchExpansionIndex` | production | this package: scripts/search-sim.ts:121 |
+| `searchExpansionIndex` | production | this package: scripts/search-sim.ts:122 |
 | `searchHistoryCoverageRow` | production | this package: src/campaign/optimization-method.ts:14 |
 | `SearchHistoryRequiredError` | none | — |
 | `SearchLedgerConflictError` | production | this package: src/campaign/search-ledger.ts:36 |
@@ -787,8 +787,8 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `searchProposalExecution` | production | this package: src/campaign/presets/run-optimization.ts:51 |
 | `searchProposerView` | production | this package: src/campaign/presets/run-optimization.ts:86 |
 | `searchReceiptAccounting` | production | this package: src/campaign/presets/run-optimization.ts:51 |
-| `SearchRecorder` | production | this package: scripts/generate-synthetic-search-ledger.ts:34 |
-| `SearchState` | production | this package: scripts/search-sim.ts:144 |
+| `SearchRecorder` | production | this package: scripts/generate-synthetic-search-ledger.ts:37 |
+| `SearchState` | production | this package: scripts/search-sim.ts:145 |
 | `SearchStateView` | production | this package: scripts/search-meta-sim.ts:49 |
 | `searchTaskSetDigest` | production | this package: src/campaign/search-ledger-recording.ts:57 |
 | `searchUnitScores` | production | this package: src/campaign/estimate-node.ts:28 |
@@ -1063,7 +1063,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 
 ### `./jev`
 
-15 value exports — 7 production, 7 planned, 1 none.
+15 value exports — 8 production, 6 planned, 1 none.
 
 | symbol | consumer | evidence |
 | --- | --- | --- |
@@ -1081,7 +1081,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `parseJevQuestions` | production | this package: src/jev.ts:7 |
 | `parseJevRequest` | production | this package: src/jev-review.ts:9 |
 | `parseJevResult` | production | this package: src/jev-review.ts:9 |
-| `prepareJevReview` | planned | example: examples/jev-behavior-review.ts:3 |
+| `prepareJevReview` | production | this package: src/rl/claim-integrity.ts:39 |
 
 ### `./jev/protocol`
 
@@ -1103,7 +1103,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | --- | --- | --- |
 | `appendLedgerLine` | production | this package: src/campaign/search-ledger-file.ts:3 |
 | `AtomicFileLockError` | production | this package: src/ledger-core/journal-file.ts:16 |
-| `canonicalString` | production | this package: scripts/search-sim.ts:154 |
+| `canonicalString` | production | this package: scripts/search-sim.ts:155 |
 | `FileLedgerJournal` | production | this package: src/campaign/search-ledger.ts:23 |
 | `hashCanonical` | production | discovery-lab:tools/provenance.mjs |
 | `jsonDocument` | production | this package: src/agent-profile-cell.ts:3 |
@@ -1280,7 +1280,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 
 ### `./rl`
 
-63 value exports — 18 production, 18 planned, 27 none.
+82 value exports — 18 production, 30 planned, 34 none.
 
 | symbol | consumer | evidence |
 | --- | --- | --- |
@@ -1294,6 +1294,14 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `buildRlDataset` | production | this package: src/rl/corpus.ts:24 |
 | `buildVerifiedFindingRow` | none | only this package's tests: src/rl/verified-findings-dataset.test.ts:5 |
 | `campaignToRunRecords` | production | agent-builder:src/lib/.server/eval/loops/auto-research-runner.ts |
+| `checkerReadSignal` | planned | doc: docs/jev-behavior-review.md |
+| `CLAIM_INTEGRITY_VERSION` | none | — |
+| `claimBrief` | none | — |
+| `claimIntegrityDigest` | none | only this package's tests: src/rl/claim-integrity.test.ts:4 |
+| `claimIntegrityJevQuestions` | none | — |
+| `claimIntegrityJevReview` | planned | doc: docs/jev-behavior-review.md |
+| `claimIntegrityTraceQuestions` | planned | doc: docs/jev-behavior-review.md |
+| `claimIntegrityVerdict` | planned | doc: docs/jev-behavior-review.md |
 | `compareAdaptationCurves` | planned | doc: docs/design/statistics-decisions.md |
 | `datasheetToMarkdown` | none | only this package's tests: src/rl/dataset.test.ts:7 |
 | `defaultBehaviorFeatures` | none | only this package's tests: src/rl/sim-fidelity.test.ts:3 |
@@ -1304,32 +1312,43 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `extractStepRewards` | none | only this package's tests: tests/rl-process-reward.test.ts:3 |
 | `extractVerifiableReward` | planned | named in gtm-agent (bind not in the import graph) |
 | `extractVerifiableRewardsFromRecords` | production | creative-agent:eval/canonical-export.ts |
+| `farmingSignal` | planned | doc: docs/jev-behavior-review.md |
 | `FileSystemOutcomeStore` | production | phony:products/builder/api/src/eval/outcomes.ts |
 | `filterDeterministicallyRewarded` | production | this package: src/rl/reward-hacking.ts:43 |
 | `firstPassK` | planned | doc: docs/statistical-evidence.md |
 | `fitBradleyTerry` | none | only this package's tests: tests/rl-tournament.test.ts:2 |
+| `GRADER_REFERENCE_PHRASES` | none | — |
+| `graderReferenceSignal` | planned | doc: docs/jev-behavior-review.md |
 | `injectIrrelevantClause` | none | only this package's tests: tests/rl-contamination.test.ts:2 |
 | `InMemoryOutcomeStore` | planned | consumer tests: phony:products/builder/api/src/eval/outcomes-correlate.test.ts |
 | `inverseProbabilityWeighting` | planned | consumer tests: agent-knowledge:tests/memory-holdout.test.ts |
+| `jevSignal` | planned | doc: docs/jev-behavior-review.md |
 | `jsDivergence` | none | only this package's tests: src/rl/sim-fidelity.test.ts:3 |
 | `loadVerifiedFindingsDataset` | planned | doc: docs/verified-labels-flywheel.md |
+| `monitorSignal` | planned | doc: docs/jev-behavior-review.md |
 | `offPolicyEstimateAll` | none | only this package's tests: tests/rl-off-policy.test.ts:3 |
 | `paretoFrontier` | production | agent-dev-container:products/sandbox/evals/run.ts |
+| `parseSourceScopeReferee` | planned | doc: docs/jev-behavior-review.md |
 | `PredictiveValidityResearcher` | planned | doc: docs/design/mlbenchmarks-book-review.md |
 | `prmTrainingPairs` | planned | doc: examples/fine-tune-with-prime-rl/README.md |
 | `quantileEdges` | none | only this package's tests: src/rl/sim-fidelity.test.ts:3 |
 | `readCorpus` | planned | named in agent-runtime (bind not in the import graph) |
+| `REFEREE_VERDICTS` | none | — |
+| `refereeSignal` | planned | doc: docs/jev-behavior-review.md |
 | `renameVariables` | none | only this package's tests: tests/rl-contamination.test.ts:2 |
+| `ruleHolds` | none | only this package's tests: src/rl/claim-integrity.test.ts:4 |
 | `runAdaptationCurve` | planned | doc: docs/statistical-evidence.md |
 | `runComputeCurve` | none | only this package's tests: tests/rl-adversarial-and-compute.test.ts:2 |
 | `runContaminationProbe` | planned | doc: docs/statistical-evidence.md |
 | `runEvalCampaign` | production | agent-builder:scripts/eval.ts |
 | `runRLCampaign` | planned | doc: docs/outcome-validity.md |
 | `runwiseStepRewardSummary` | none | only this package's tests: tests/rl-process-reward.test.ts:3 |
+| `scopeSignals` | planned | doc: docs/jev-behavior-review.md |
 | `selfConsistency` | none | only this package's tests: tests/rl-adversarial-and-compute.test.ts:2 |
 | `selfNormalizedImportanceWeighting` | planned | consumer tests: agent-knowledge:tests/memory-holdout.test.ts |
 | `shuffleOrder` | none | — |
 | `simFidelityReport` | none | only this package's tests: src/rl/sim-fidelity.test.ts:3 |
+| `sourceScopeRefereeMessages` | planned | doc: docs/jev-behavior-review.md |
 | `stepRewardsToJsonl` | none | only this package's tests: src/rollout/gate-properties.test.ts:38 |
 | `summarizeVerifiedFindings` | none | only this package's tests: src/rl/verified-findings-dataset.test.ts:5 |
 | `toAnthropicFormat` | production | blueprint-agent:scripts/experiments/lib/steering-dataset.ts |
@@ -1428,7 +1447,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 
 | symbol | consumer | evidence |
 | --- | --- | --- |
-| `draftOnPlateau` | production | this package: scripts/search-sim.ts:161 |
+| `draftOnPlateau` | production | this package: scripts/search-sim.ts:162 |
 | `EDIT_CREDIT_ESTIMATOR` | none | — |
 | `EDIT_CREDIT_SIGNAL` | none | — |
 | `editCredit` | production | this package: scripts/synthetic-search-ledger.ts:70 |
@@ -1437,9 +1456,9 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `formatLandscape` | production | this package: src/search-command.ts:41 |
 | `formatSkillManifold` | production | this package: src/search-command.ts:45 |
 | `front` | production | this package: src/search-command.ts:40 |
-| `landscape` | production | this package: scripts/search-sim.ts:155 |
+| `landscape` | production | this package: scripts/search-sim.ts:156 |
 | `lineageEdits` | planned | doc: docs/search-ledger.md |
-| `lineEditDistance` | production | this package: scripts/search-sim.ts:155 |
+| `lineEditDistance` | production | this package: scripts/search-sim.ts:156 |
 | `META_SEARCH_SCORE_SOURCE` | production | this package: scripts/search-meta-sim.ts:52 |
 | `META_SEARCH_SIGNAL` | none | — |
 | `metaSearch` | production | this package: scripts/search-meta-sim.ts:52 |
@@ -1447,7 +1466,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `MIN_OUTCOMES_FOR_WEIGHT` | planned | doc: docs/search-ledger.md |
 | `nearestNeighbours` | none | — |
 | `nestedSearchId` | planned | doc: docs/search-ledger.md |
-| `nextUnitExtension` | production | this package: scripts/search-sim.ts:162 |
+| `nextUnitExtension` | production | this package: scripts/search-sim.ts:163 |
 | `objectiveKey` | production | this package: scripts/search-meta-sim.ts:52 |
 | `operatorYield` | production | this package: scripts/prove-operator-bandit.mts:24 |
 | `profileTextLines` | none | — |
@@ -1458,12 +1477,12 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `searchConfigCodec` | planned | doc: docs/search-ledger.md |
 | `searchPlateau` | production | this package: src/search/lenses/landscape.ts:49 |
 | `searchPolicyGenome` | planned | doc: docs/search-ledger.md |
-| `skillCalibration` | production | this package: scripts/search-sim.ts:162 |
-| `skillManifold` | production | this package: scripts/search-sim.ts:162 |
+| `skillCalibration` | production | this package: scripts/search-sim.ts:163 |
+| `skillManifold` | production | this package: scripts/search-sim.ts:163 |
 | `summarizeSamples` | production | this package: src/search/lenses/meta-search.ts:55 |
 | `surfaceDigestEdits` | none | — |
 | `surfaceTextEdits` | production | this package: src/search-command.ts:41 |
-| `taskMatrix` | production | this package: src/search-command.ts:46 |
+| `taskMatrix` | production | this package: scripts/prove-task-matrix.mts:8 |
 | `tree` | production | this package: src/search-command.ts:47 |
 | `vectorEmbedding` | planned | doc: docs/search-ledger.md |
 

package/docs/search-ledger.md

@@ -175,7 +175,7 @@ tree: 2 nodes, 2 edges
 seed → node_8b6e8e11c5a430aa900f210c8e5162d7 (rejected) $0.00 known + at least $0.00 over 2 unknown-cost cells
 improve → node_9d74773e7ecfd982307a8285845e7fad (selected) $0.00 known + at least $0.00 over 2 unknown-cost cells
 
-operator yield (selection split):
+operator yield (selection split, gain per known $ of proposal + screen):
   improve: 1 proposals → 1 node (selected=1); yield: no measured children; 1 excluded
   seed: 1 proposals → 1 node (rejected=1); yield: no measured children; 1 excluded
 
@@ -206,12 +206,30 @@ The two `--wired` behaviors above (`incumbentWithOperatorBandit`, `draftOnPlatea
 | Lens | Reports | Signal |
 |---|---|---|
 | `tree(state)` | a tidy tree of nodes and edges — the base view every other lens sits beside; a node with unknown-cost cells shows its known spend, its proven floor and how many cells are unknown, never a total | `tree.nodeCount` (drives no policy) |
-| `operatorYield(state, { split? })` | each edge operator's proposals, re-proposals, outcome counts and improvement-per-known-dollar yield, from `searchPosterior` (`estimateNode`'s tree-wide contrast); a node counts once, under the operator of the edge that registered it, and a re-proposal adds a proposal but no outcome or sample; a node is excluded from yield when it was decided invalid, shares fewer than 2 units with the root, or any of its cells has an unknown cost | `operatorYield.weights`: an operator's yield mean once it has 6 or more yield-eligible nodes (`MIN_OUTCOMES_FOR_WEIGHT`), else `null` |
+| `operatorYield(state, { split? })` | each edge operator's proposals, re-proposals, outcome counts and yield: a node's gain over the root from `searchPosterior` (`estimateNode`'s tree-wide contrast) divided by the known cost of producing and screening it — its even share of the proposal operation's cost over that operation's child edges, plus its cells allocated before its first decision (the first rung and the train feedback); cells of later rungs and the claim are the allocator's and the claim's choice, so they are not charged to the operator; a node counts once, under the operator of the edge that registered it, and a re-proposal adds a proposal but no outcome or sample; a node is excluded from yield when it was decided invalid, shares fewer than 2 units with the root, or its proposal or a screen cell has an unknown, unrecorded or zero cost | `operatorYield.weights`: an operator's yield mean once it has 6 or more yield-eligible nodes (`MIN_OUTCOMES_FOR_WEIGHT`), else `null` |
 | `front(state, { split?, axes? })` | the Pareto frontier over per-unit mean score and known cost per attempted cell on the split (reusing `paretoFrontier`), with room for caller-declared extra axes; a node's total spend depends on how far the allocator measured it, so it is shown but is not an axis; a node decided invalid, scored on fewer than 2 units, with an unknown-cost cell on the split, or with a non-finite extra axis is excluded from every frontier pass, with the reason | `front.membership`: 1 for a node on the frontier, 0 otherwise (including an excluded node) |
 | `taskMatrix(state, { split? })` | nodes and units, each single-linkage clustered on the root-mean-square difference over their shared scores, cutoff at the data's own median pairwise distance | `taskMatrix.specialistGain`: per unit cluster, the best node's mean minus the mean of node means, in the objective's direction, among nodes not decided invalid that scored every unit of the cluster; `null` with the reason, and omitted from the signal, below 2 such nodes or 6 units |
 
 `agent-eval search show <ledger> [--tree] [--operator-yield] [--front] [--task-matrix]` prints each lens's text form below the search summary, so an agent reading the CLI sees the same numbers Intelligence, discovery lab, VerticalBench and agent-runtime `improve()` would render from the same JSON.
 
+The task matrix evaluates each node pair and each unit pair at most twice.
+Clustering preserves the median cutoff, missing-score rules, deterministic ordering, and specialist gains.
+Piped CLI output includes the complete JSON before the process exits.
+
+Run the proof against retained search ledgers:
+
+```sh
+pnpm build
+pnpm exec tsx scripts/prove-task-matrix.mts /path/to/search-ledger.jsonl
+```
+
+The proof compares piped CLI JSON with the library result and enforces the distance-call bound using V8 counters.
+It reports ledger hashes, dimensions, complete output size, and elapsed library time.
+Wall time is descriptive; the operation count is the performance guard.
+On a retained 501-node, 24-unit scale ledger, node-distance calls fell from 44,650,773 to 250,500.
+The complete 1,668,617-byte CLI output stayed identical.
+This scale ledger is generated data; it does not establish research quality or a live deployment speedup.
+
 `incumbentWithOperatorBandit({ seed, fixedWeights? })` (`/campaign`) is the one built-in policy that reads a lens signal: a hill climb, like `incumbent`, whose expansion operator is a weighted draw over `operatorYield`'s weights.
 An operator without 6 measured outcomes yet draws on `fixedWeights` (uniform by default) instead of being starved until every operator clears the gate; a measured operator's weight is `fixedWeights[operator] + yield`, floored just above zero, because yield (dollars) and the fixed prior (an arbitrary share) are not on the same scale and a small positive yield should not draw less than an untested operator's default prior.
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tangle-network/agent-eval",
-  "version": "0.199.0",
+  "version": "0.200.0",
   "description": "Evaluate and improve AI agents from runs, traces, judges, and feedback. Compare candidates, cluster failures, measure lift, and gate releases.",
   "homepage": "https://github.com/tangle-network/agent-eval#readme",
   "repository": {