@tangle-network/agent-eval 0.120.2 → 0.120.4

package/dist/campaign/index.js

@@ -0,0 +1,272 @@
+import {
+  DEFAULT_POLICY_EDIT_HISTORY_LIMITS,
+  FileSearchLedger,
+  FsLabeledScenarioStore,
+  LabeledScenarioStoreError,
+  Lineage,
+  ProfileMatrixError,
+  SEARCH_LEDGER_SCHEMA,
+  SkillPatchParseError,
+  WorktreeAdapterError,
+  aceProposer,
+  acquireSingleRunLock,
+  analyzeCrossSurfaceInteractions,
+  applySkillPatch,
+  assertPolicyEditAuthorContextBudget,
+  buildAnalystSurfaceDispatch,
+  callbackGovernor,
+  classifyUngroundedLiterals,
+  compareProposers,
+  compositeProposer,
+  discoverEvalFixtures,
+  extractFapoAttributionSignals,
+  failureModeRecallJudge,
+  fapoEscalationEntry,
+  fapoProposer,
+  fsLineageStore,
+  gepaParetoEntry,
+  gepaReflectionEntry,
+  gitWorktreeAdapter,
+  haloProposer,
+  heuristicGovernor,
+  isTransientTransportFailure,
+  lineageNodeId,
+  llmPolicyEditProposer,
+  loadEvalFixture,
+  loadEvalFixtureScenarios,
+  makePlaybackDispatch,
+  memLineageStore,
+  memoryCurationProposer,
+  neutralizationGate,
+  neutralizeText,
+  openSearchLedger,
+  parameterSweepProposer,
+  parseSkillPatchResponse,
+  patchEditCount,
+  planEvalFixtureRun,
+  policyEditProposer,
+  projectPolicyEditHistory,
+  renderScoreboardMarkdown,
+  resolveWorktreePath,
+  rolloutArgumentDiff,
+  runLineage,
+  runLineageLoop,
+  runProfileMatrix,
+  runSkillOpt,
+  scoreDiscrimination,
+  scoreUserStory,
+  scoreboardSummary,
+  selectDiscriminative,
+  selectPolicyEditAuthorRows,
+  sequentialDecide,
+  sequentialPairedGate,
+  skillOptEntry,
+  skillOptProposer,
+  traceAnalystProposer,
+  userStoryScoreboard,
+  validateSearchLedgerEvent,
+  verifyCodeSurface
+} from "../chunk-ZTJPIIVI.js";
+import {
+  assertCodeSurfaceIdentity,
+  buildEvidenceVector,
+  buildLoopProvenanceRecord,
+  campaignBreakdown,
+  campaignMeanComposite,
+  campaignMeasurementDigest,
+  codeSurfaceIdentityMaterial,
+  composeGate,
+  countSentenceEdits,
+  createReferenceEquivalenceJudge,
+  defaultProductionGate,
+  detectScale,
+  dimensionRegressions,
+  emitLoopProvenance,
+  evolutionaryProposer,
+  extractH2Sections,
+  gepaProposer,
+  heldOutGate,
+  heldoutSignificance,
+  isProposedCandidate,
+  labelTrustRank,
+  llmJudge,
+  loopProvenanceArgsFromResult,
+  loopProvenanceSpans,
+  openAutoPr,
+  pairHoldout,
+  paretoPolicy,
+  paretoSignificanceGate,
+  powerPreflight,
+  provenanceRecordPath,
+  provenanceSpansPath,
+  renderSurfaceDiff,
+  runEval,
+  runImprovementLoop,
+  runOptimization,
+  surfaceContentHash,
+  surfaceHash,
+  verifyLoopProvenanceRecord
+} from "../chunk-7QBFOJ3E.js";
+import {
+  SearchLedgerConflictError,
+  SearchLedgerError,
+  SearchLedgerIntegrityError,
+  assertCampaignDesign,
+  assertCampaignSplitIdentity,
+  campaignScenarioIdentity,
+  campaignSplitDigest,
+  campaignSplitDigestFromIdentities,
+  createRunCostLedger,
+  fsCampaignStorage,
+  inMemoryCampaignStorage,
+  planCampaignRun,
+  resolveRunDir,
+  runCampaign,
+  tangleTracesRoot
+} from "../chunk-SYV364BL.js";
+import "../chunk-3YYRZDON.js";
+import {
+  POLICY_EDIT_CANDIDATE_RECORD_SCHEMA,
+  validatePolicyEditCandidateRecord
+} from "../chunk-5CVUPHJ4.js";
+import "../chunk-ARU2PZFM.js";
+import "../chunk-NJC7U437.js";
+import "../chunk-PJQFMIOX.js";
+import "../chunk-ZMXDQ4K7.js";
+import "../chunk-VI2UW6B6.js";
+import "../chunk-WW2A73HW.js";
+import "../chunk-GGE4NNQT.js";
+import "../chunk-PXD6ZFNY.js";
+import "../chunk-PC4UYEBM.js";
+import "../chunk-S3UZOQ5Y.js";
+import "../chunk-MA6HLL3S.js";
+import "../chunk-XJYR7XFV.js";
+import "../chunk-VSMTAMNK.js";
+import "../chunk-ONWEPEDO.js";
+import "../chunk-K4DBDHLK.js";
+import "../chunk-PZ5AY32C.js";
+export {
+  DEFAULT_POLICY_EDIT_HISTORY_LIMITS,
+  FileSearchLedger,
+  FsLabeledScenarioStore,
+  LabeledScenarioStoreError,
+  Lineage,
+  POLICY_EDIT_CANDIDATE_RECORD_SCHEMA,
+  ProfileMatrixError,
+  SEARCH_LEDGER_SCHEMA,
+  SearchLedgerConflictError,
+  SearchLedgerError,
+  SearchLedgerIntegrityError,
+  SkillPatchParseError,
+  WorktreeAdapterError,
+  aceProposer,
+  acquireSingleRunLock,
+  analyzeCrossSurfaceInteractions,
+  applySkillPatch,
+  assertCampaignDesign,
+  assertCampaignSplitIdentity,
+  assertCodeSurfaceIdentity,
+  assertPolicyEditAuthorContextBudget,
+  buildAnalystSurfaceDispatch,
+  buildEvidenceVector,
+  buildLoopProvenanceRecord,
+  callbackGovernor,
+  campaignBreakdown,
+  campaignMeanComposite,
+  campaignMeasurementDigest,
+  campaignScenarioIdentity,
+  campaignSplitDigest,
+  campaignSplitDigestFromIdentities,
+  classifyUngroundedLiterals,
+  codeSurfaceIdentityMaterial,
+  compareProposers,
+  composeGate,
+  compositeProposer,
+  countSentenceEdits,
+  createReferenceEquivalenceJudge,
+  createRunCostLedger,
+  defaultProductionGate,
+  detectScale,
+  dimensionRegressions,
+  discoverEvalFixtures,
+  emitLoopProvenance,
+  evolutionaryProposer,
+  extractFapoAttributionSignals,
+  extractH2Sections,
+  failureModeRecallJudge,
+  fapoEscalationEntry,
+  fapoProposer,
+  fsCampaignStorage,
+  fsLineageStore,
+  gepaParetoEntry,
+  gepaProposer,
+  gepaReflectionEntry,
+  gitWorktreeAdapter,
+  haloProposer,
+  heldOutGate,
+  heldoutSignificance,
+  heuristicGovernor,
+  inMemoryCampaignStorage,
+  isProposedCandidate,
+  isTransientTransportFailure,
+  labelTrustRank,
+  lineageNodeId,
+  llmJudge,
+  llmPolicyEditProposer,
+  loadEvalFixture,
+  loadEvalFixtureScenarios,
+  loopProvenanceArgsFromResult,
+  loopProvenanceSpans,
+  makePlaybackDispatch,
+  memLineageStore,
+  memoryCurationProposer,
+  neutralizationGate,
+  neutralizeText,
+  openAutoPr,
+  openSearchLedger,
+  pairHoldout,
+  parameterSweepProposer,
+  paretoPolicy,
+  paretoSignificanceGate,
+  parseSkillPatchResponse,
+  patchEditCount,
+  planCampaignRun,
+  planEvalFixtureRun,
+  policyEditProposer,
+  powerPreflight,
+  projectPolicyEditHistory,
+  provenanceRecordPath,
+  provenanceSpansPath,
+  renderScoreboardMarkdown,
+  renderSurfaceDiff,
+  resolveRunDir,
+  resolveWorktreePath,
+  rolloutArgumentDiff,
+  runCampaign,
+  runEval,
+  runImprovementLoop,
+  runLineage,
+  runLineageLoop,
+  runOptimization,
+  runProfileMatrix,
+  runSkillOpt,
+  scoreDiscrimination,
+  scoreUserStory,
+  scoreboardSummary,
+  selectDiscriminative,
+  selectPolicyEditAuthorRows,
+  sequentialDecide,
+  sequentialPairedGate,
+  skillOptEntry,
+  skillOptProposer,
+  surfaceContentHash,
+  surfaceHash,
+  tangleTracesRoot,
+  traceAnalystProposer,
+  userStoryScoreboard,
+  validatePolicyEditCandidateRecord,
+  validateSearchLedgerEvent,
+  verifyCodeSurface,
+  verifyLoopProvenanceRecord
+};
+//# sourceMappingURL=index.js.map

package/dist/campaign/index.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":[],"sourcesContent":[],"mappings":"","names":[]}

package/dist/chunk-3RF76KTD.js

@@ -0,0 +1,84 @@
+// src/meta-eval/outcome-store.ts
+var InMemoryOutcomeStore = class {
+  items = [];
+  async append(outcome) {
+    this.items.push({ ...outcome });
+  }
+  async forRun(runId) {
+    return this.items.filter((o) => o.runId === runId).map((o) => ({ ...o }));
+  }
+  async list(filter = {}) {
+    return this.items.filter((o) => matches(o, filter)).map((o) => ({ ...o }));
+  }
+};
+var FileSystemOutcomeStore = class {
+  dir;
+  maxBytes;
+  memo;
+  loaded = false;
+  constructor(options) {
+    this.dir = options.dir;
+    this.maxBytes = options.maxBytes ?? 32 * 1024 * 1024;
+  }
+  async ensureDir() {
+    const fs = await import("fs/promises");
+    await fs.mkdir(this.dir, { recursive: true });
+  }
+  async append(outcome) {
+    await this.ensureDir();
+    const fs = await import("fs/promises");
+    const path = await import("path");
+    const active = path.join(this.dir, "outcomes.ndjson");
+    try {
+      const stat = await fs.stat(active);
+      if (stat.size >= this.maxBytes) {
+        await fs.rename(active, path.join(this.dir, `outcomes.${Date.now()}.ndjson`));
+      }
+    } catch {
+    }
+    await fs.appendFile(active, `${JSON.stringify(outcome)}
+`, "utf8");
+    if (this.memo) await this.memo.append(outcome);
+  }
+  async load() {
+    if (this.loaded && this.memo) return this.memo;
+    const fs = await import("fs/promises");
+    const path = await import("path");
+    const memo = new InMemoryOutcomeStore();
+    try {
+      const entries = await fs.readdir(this.dir);
+      for (const file of entries) {
+        if (!file.endsWith(".ndjson")) continue;
+        const content = await fs.readFile(path.join(this.dir, file), "utf8");
+        for (const line of content.split("\n")) {
+          if (!line.trim()) continue;
+          await memo.append(JSON.parse(line));
+        }
+      }
+    } catch {
+    }
+    this.memo = memo;
+    this.loaded = true;
+    return memo;
+  }
+  async forRun(runId) {
+    return (await this.load()).forRun(runId);
+  }
+  async list(filter) {
+    return (await this.load()).list(filter);
+  }
+};
+function matches(o, f) {
+  if (f.runIds && !f.runIds.includes(o.runId)) return false;
+  if (f.since !== void 0 && o.capturedAt < f.since) return false;
+  if (f.until !== void 0 && o.capturedAt > f.until) return false;
+  if (f.source && o.source !== f.source) return false;
+  if (f.label && o.labels?.[f.label.key] !== f.label.value) return false;
+  return true;
+}
+
+export {
+  InMemoryOutcomeStore,
+  FileSystemOutcomeStore
+};
+//# sourceMappingURL=chunk-3RF76KTD.js.map

package/dist/chunk-3RF76KTD.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/meta-eval/outcome-store.ts"],"sourcesContent":["/**\n * OutcomeStore — deployment outcomes attached to Run IDs.\n *\n * Outcomes arrive asynchronously from production telemetry after the\n * eval run completed: user ratings, retention flags, conversion events,\n * revenue, support-ticket rate, anything a product team can measure.\n * The store is a peer to TraceStore — separate lifecycle, same runId\n * foreign key.\n *\n * The whole point of this module is to make the meta-eval correlation\n * question computable: `correlate(evalMetric, outcomeMetric) → r, ρ, n, CI`.\n */\n\nexport interface DeploymentOutcome {\n  runId: string\n  capturedAt: number\n  /** Numeric outcomes keyed by name — retention_7d, csat, revenue_usd, etc. */\n  metrics: Record<string, number>\n  /** Dimensions for stratified analysis — cohort, region, user_segment. */\n  labels?: Record<string, string>\n  /** Free-form provenance (source system, pipeline version). */\n  source?: string\n}\n\nexport interface OutcomeFilter {\n  runIds?: string[]\n  since?: number\n  until?: number\n  label?: { key: string; value: string }\n  source?: string\n}\n\nexport interface OutcomeStore {\n  append(outcome: DeploymentOutcome): Promise<void>\n  /** All outcomes attached to this run (a single run can have many — multiple\n   *  capture windows over deployment time). */\n  forRun(runId: string): Promise<DeploymentOutcome[]>\n  list(filter?: OutcomeFilter): Promise<DeploymentOutcome[]>\n}\n\nexport class InMemoryOutcomeStore implements OutcomeStore {\n  private items: DeploymentOutcome[] = []\n\n  async append(outcome: DeploymentOutcome): Promise<void> {\n    this.items.push({ ...outcome })\n  }\n\n  async forRun(runId: string): Promise<DeploymentOutcome[]> {\n    return this.items.filter((o) => o.runId === runId).map((o) => ({ ...o }))\n  }\n\n  async list(filter: OutcomeFilter = {}): Promise<DeploymentOutcome[]> {\n    return this.items.filter((o) => matches(o, filter)).map((o) => ({ ...o }))\n  }\n}\n\nexport interface FileSystemOutcomeStoreOptions {\n  dir: string\n  maxBytes?: number\n}\n\nexport class FileSystemOutcomeStore implements OutcomeStore {\n  private dir: string\n  private maxBytes: number\n  private memo?: InMemoryOutcomeStore\n  private loaded = false\n\n  constructor(options: FileSystemOutcomeStoreOptions) {\n    this.dir = options.dir\n    this.maxBytes = options.maxBytes ?? 32 * 1024 * 1024\n  }\n\n  private async ensureDir(): Promise<void> {\n    const fs = await import('node:fs/promises')\n    await fs.mkdir(this.dir, { recursive: true })\n  }\n\n  async append(outcome: DeploymentOutcome): Promise<void> {\n    await this.ensureDir()\n    const fs = await import('node:fs/promises')\n    const path = await import('node:path')\n    const active = path.join(this.dir, 'outcomes.ndjson')\n    try {\n      const stat = await fs.stat(active)\n      if (stat.size >= this.maxBytes) {\n        await fs.rename(active, path.join(this.dir, `outcomes.${Date.now()}.ndjson`))\n      }\n    } catch {\n      /* first write */\n    }\n    await fs.appendFile(active, `${JSON.stringify(outcome)}\\n`, 'utf8')\n    if (this.memo) await this.memo.append(outcome)\n  }\n\n  private async load(): Promise<InMemoryOutcomeStore> {\n    if (this.loaded && this.memo) return this.memo\n    const fs = await import('node:fs/promises')\n    const path = await import('node:path')\n    const memo = new InMemoryOutcomeStore()\n    try {\n      const entries = await fs.readdir(this.dir)\n      for (const file of entries) {\n        if (!file.endsWith('.ndjson')) continue\n        const content = await fs.readFile(path.join(this.dir, file), 'utf8')\n        for (const line of content.split('\\n')) {\n          if (!line.trim()) continue\n          await memo.append(JSON.parse(line))\n        }\n      }\n    } catch {\n      /* empty */\n    }\n    this.memo = memo\n    this.loaded = true\n    return memo\n  }\n\n  async forRun(runId: string): Promise<DeploymentOutcome[]> {\n    return (await this.load()).forRun(runId)\n  }\n\n  async list(filter?: OutcomeFilter): Promise<DeploymentOutcome[]> {\n    return (await this.load()).list(filter)\n  }\n}\n\nfunction matches(o: DeploymentOutcome, f: OutcomeFilter): boolean {\n  if (f.runIds && !f.runIds.includes(o.runId)) return false\n  if (f.since !== undefined && o.capturedAt < f.since) return false\n  if (f.until !== undefined && o.capturedAt > f.until) return false\n  if (f.source && o.source !== f.source) return false\n  if (f.label && o.labels?.[f.label.key] !== f.label.value) return false\n  return true\n}\n"],"mappings":";AAwCO,IAAM,uBAAN,MAAmD;AAAA,EAChD,QAA6B,CAAC;AAAA,EAEtC,MAAM,OAAO,SAA2C;AACtD,SAAK,MAAM,KAAK,EAAE,GAAG,QAAQ,CAAC;AAAA,EAChC;AAAA,EAEA,MAAM,OAAO,OAA6C;AACxD,WAAO,KAAK,MAAM,OAAO,CAAC,MAAM,EAAE,UAAU,KAAK,EAAE,IAAI,CAAC,OAAO,EAAE,GAAG,EAAE,EAAE;AAAA,EAC1E;AAAA,EAEA,MAAM,KAAK,SAAwB,CAAC,GAAiC;AACnE,WAAO,KAAK,MAAM,OAAO,CAAC,MAAM,QAAQ,GAAG,MAAM,CAAC,EAAE,IAAI,CAAC,OAAO,EAAE,GAAG,EAAE,EAAE;AAAA,EAC3E;AACF;AAOO,IAAM,yBAAN,MAAqD;AAAA,EAClD;AAAA,EACA;AAAA,EACA;AAAA,EACA,SAAS;AAAA,EAEjB,YAAY,SAAwC;AAClD,SAAK,MAAM,QAAQ;AACnB,SAAK,WAAW,QAAQ,YAAY,KAAK,OAAO;AAAA,EAClD;AAAA,EAEA,MAAc,YAA2B;AACvC,UAAM,KAAK,MAAM,OAAO,aAAkB;AAC1C,UAAM,GAAG,MAAM,KAAK,KAAK,EAAE,WAAW,KAAK,CAAC;AAAA,EAC9C;AAAA,EAEA,MAAM,OAAO,SAA2C;AACtD,UAAM,KAAK,UAAU;AACrB,UAAM,KAAK,MAAM,OAAO,aAAkB;AAC1C,UAAM,OAAO,MAAM,OAAO,MAAW;AACrC,UAAM,SAAS,KAAK,KAAK,KAAK,KAAK,iBAAiB;AACpD,QAAI;AACF,YAAM,OAAO,MAAM,GAAG,KAAK,MAAM;AACjC,UAAI,KAAK,QAAQ,KAAK,UAAU;AAC9B,cAAM,GAAG,OAAO,QAAQ,KAAK,KAAK,KAAK,KAAK,YAAY,KAAK,IAAI,CAAC,SAAS,CAAC;AAAA,MAC9E;AAAA,IACF,QAAQ;AAAA,IAER;AACA,UAAM,GAAG,WAAW,QAAQ,GAAG,KAAK,UAAU,OAAO,CAAC;AAAA,GAAM,MAAM;AAClE,QAAI,KAAK,KAAM,OAAM,KAAK,KAAK,OAAO,OAAO;AAAA,EAC/C;AAAA,EAEA,MAAc,OAAsC;AAClD,QAAI,KAAK,UAAU,KAAK,KAAM,QAAO,KAAK;AAC1C,UAAM,KAAK,MAAM,OAAO,aAAkB;AAC1C,UAAM,OAAO,MAAM,OAAO,MAAW;AACrC,UAAM,OAAO,IAAI,qBAAqB;AACtC,QAAI;AACF,YAAM,UAAU,MAAM,GAAG,QAAQ,KAAK,GAAG;AACzC,iBAAW,QAAQ,SAAS;AAC1B,YAAI,CAAC,KAAK,SAAS,SAAS,EAAG;AAC/B,cAAM,UAAU,MAAM,GAAG,SAAS,KAAK,KAAK,KAAK,KAAK,IAAI,GAAG,MAAM;AACnE,mBAAW,QAAQ,QAAQ,MAAM,IAAI,GAAG;AACtC,cAAI,CAAC,KAAK,KAAK,EAAG;AAClB,gBAAM,KAAK,OAAO,KAAK,MAAM,IAAI,CAAC;AAAA,QACpC;AAAA,MACF;AAAA,IACF,QAAQ;AAAA,IAER;AACA,SAAK,OAAO;AACZ,SAAK,SAAS;AACd,WAAO;AAAA,EACT;AAAA,EAEA,MAAM,OAAO,OAA6C;AACxD,YAAQ,MAAM,KAAK,KAAK,GAAG,OAAO,KAAK;AAAA,EACzC;AAAA,EAEA,MAAM,KAAK,QAAsD;AAC/D,YAAQ,MAAM,KAAK,KAAK,GAAG,KAAK,MAAM;AAAA,EACxC;AACF;AAEA,SAAS,QAAQ,GAAsB,GAA2B;AAChE,MAAI,EAAE,UAAU,CAAC,EAAE,OAAO,SAAS,EAAE,KAAK,EAAG,QAAO;AACpD,MAAI,EAAE,UAAU,UAAa,EAAE,aAAa,EAAE,MAAO,QAAO;AAC5D,MAAI,EAAE,UAAU,UAAa,EAAE,aAAa,EAAE,MAAO,QAAO;AAC5D,MAAI,EAAE,UAAU,EAAE,WAAW,EAAE,OAAQ,QAAO;AAC9C,MAAI,EAAE,SAAS,EAAE,SAAS,EAAE,MAAM,GAAG,MAAM,EAAE,MAAM,MAAO,QAAO;AACjE,SAAO;AACT;","names":[]}