@tangle-network/agent-eval 1.0.0 → 1.0.1

package/CHANGELOG.md

@@ -4,6 +4,125 @@ All notable changes to `@tangle-network/agent-eval` and its sibling `agent-eval-
 
 ---
 
+## [1.0.1] — 2026-10-28
+
+- Refresh the live benchmark dependency identity after interface updates
+- Preserve exact execution controls in evaluation-backed analysts
+- ci(release): admit the release version before publishing (#1154)
+- fix(adopt-stack): dedupe the lockfile after locking the planned versions (#1153)
+- wip
+- wip
+- wip
+- Lock per importer, keep npm ranges, deliver after 20 h
+- Adopt compatible releases through the lockfile, once a day
+- docs(readme): drop the unmeasured 30-second promise; link the no-change gate example (#1151)
+- fix(adopt-stack): restart an updater-only proposal that no longer merges (#1147)
+- fix(deps): adopt agent-interface 5.10.2 (#1150)
+- fix(deps): adopt agent-interface 5.10.1 for the Zod advisory (#1149)
+- feat(statistics): export meanConfidenceInterval and studentTCritical, a one-sample Student-t interval on the mean
+- factory: Student t mean interval in agent eval
+- feat(analyst): export the behavioral failure-class mapping, its load-time guard and pinBehavioralFailureClass
+- factory: Failure mapping in agent eval
+- fix(adopt-stack): an untracked file the gate wrote does not change the gated tree (#1145)
+- refactor(trace): OTLP ids, encoding and decoding from agent-trace-contract (#1144)
+- feat(redaction): publish shared detectors with safe substring ranges (#1139)
+- chore(release): 1.0.0 (#1142)
+- test(adopt-stack): explain a refusal by any peer whose latest is an excluded major; post-pin cases use a manifest with no 0.x peer
+- test(adopt-stack): the live-registry plans hold across a package's first 1.0 release
+- deps: repin the analyst benchmark's dependency-lock digest
+- deps: lock agent-core 1.0.0, harness-sessions 1.0.1, agent-interface 5.8.1 and agent-trace-contract 2.2.0
+- deps: floor agent-interface at 5.8.1 and agent-trace-contract at 2.2.0
+- deps: agent-core ^1.0.0 and harness-sessions ^1.0.1
+- chore(release): 0.223.0 (#1140)
+- feat(storyboard): one tool classifier, classifyTool, shared by scenes and extractVisual
+- refactor: use statistics mean helpers instead of private copies (#1137)
+- fix(adopt-stack): restart a branch that outlived another's squash merge (#1136)
+- fix(analyst): repin the benchmark dependency lock after the zod range (#1135)
+- chore(release): 0.222.1 (#1134)
+- fix(deps): depend on zod ^4.6.6 so consumers resolve one copy (#1133)
+- chore(test): retire incomplete Runtime adoption snapshot
+- fix(release): build public exports before integration tests
+- chore(release): 0.222.0
+- feat(campaign): share concurrency for measured scenario adapters
+- fix(supervisor-run): staged play-check reward is stages passed, read from the SCORE line only (#1126)
+- refactor(statistics): delete duplicate numeric summary helpers
+- fix(rollout): import survives a damaged deliverables archive; named ids are not keys (#1125)
+- feat(supervisor-run): rollouts importer for Discovery runs, harness sessions and gate receipts (#1124)
+- feat(statistics): extend series summaries with nearest-rank evidence
+- test(rollout): isolate bundle permission probe from runner identity
+- fix(rollout): license class is provenance and an opt-in filter, not a default exclusion (#1121)
+- fix(rollout): allow nonroot images to read the base bundle
+- feat(rollout): license class on rollout lines and whole-row privacy drop in rollout-release (#1119)
+- test(rollout): inherit configured identity for Git fixtures
+- test(rollout): capture empty tool definitions at the ledger boundary
+- feat(rollout): write verified episodes as Harbor tasks
+- feat(rollout): project captured policy steps into Tinker data
+- adopt-stack: delete the adoption branch when its squash merge lands (#1115)
+- chore(deps)!: require Python 3.11 for patched locks and refresh dependencies
+- feat(rollout): read retained Codex sessions through the shared reader
+- chore(release): 0.221.8
+- test(adopt): retain failed browser child outcomes
+- docs: leave changelog preparation to the release workflow
+- chore(release): 0.221.7
+- feat(multishot): conversation legs report the USD they spent
+- chore(release): 0.221.6
+- docs(traces): bound the incomplete usage coverage claim
+- feat: own judge-panel calibration, multishot conversation legs and served-panel identity
+- test(traces): typecheck OTLP usage integration proof
+- fix(traces): retain incomplete OTLP token usage
+- chore(release): 0.221.5
+- ci: retain private adopter failure evidence (#1106)
+- test(process): isolate process-tree fixture environment
+- fix(adopt): invoke each resolved Playwright CLI (#1096)
+- test(systemone): close disposable provider connections
+- chore(release): 0.221.4
+- test(adoption): inherit the maintained integration deadline
+- fix(adoption): scope guard allowance to disposable Git fixtures
+- fix(analyst): carry root-cause target into failure-mode rubric
+- ci: check locally and in publication paths
+- test(adopt): give real-git fixtures their own commit identity (#1101)
+- chore(release): 0.221.3 (#1099)
+- fix(analyst): repin the benchmark implementation digest after #1097 (#1098)
+- fix(trace-analyst): build OTLP stores from bytes, not one joined string
+- fix(adopt): prepare caller-locked Chromium before gates
+- test(adopt): retain admitted identity in real Git fixtures
+- fix(adopt): retain the sparse caller entrypoint
+- fix(adopt): settle pushed proposal reads against remote refs
+- chore(release): 0.221.2
+- Return plain analyst findings from decision replay
+- chore(release): 0.221.1 (#1091)
+- fix(deps): pin zod 4.6.6 (GHSA-mrhx-c7xc-m3w5) (#1088)
+- Give the disposable adoption fixture its own Git author
+- Fix real adoption drift fixture under the declared package manager
+- Capture supporting trace events in the original native decision
+- test: inherit the integration default for behavioral questions
+- test: set the integration default from the observed loaded tail
+- chore(release): 0.221.0
+- test: own the measured integration deadline in Vitest config
+- test: await observed integration completion across tool paths
+- feat: preserve finding composition and assigned arm populations
+- test: await decision-point HTTP and ledger completion
+- test: await durable readiness in filesystem integration cases
+- test(process): wait for argv child readiness before abort
+- feat(analyst): add bounded calibrated trace question observations
+- fix(analyst): reject surrounding whitespace in caller gold ids
+- test(analyst): reproduce whitespace source-id bypass
+- test(optimizer): retire fake CLI metering harness
+- test(analyst): isolate recorded gold admission scenarios
+- fix(systemone): validate decision boundaries before routing
+- ci: pin the eval publication dispatcher body
+- fix(analyst): reserve excluded source ids across caller gold
+- fix(analyst): validate root provenance before gold admission
+- fix(analyst): preserve actor boundaries in AgentRx calibration gold
+- Preserve manual drafts observed while recording failed adoption (#1076)
+- Recover updater-owned failure drafts while preserving manual holds (#1075)
+- feat(systemone): persist calibrated decision points and outcomes
+- ci: move eval verification and Python publishing to owned runners
+- chore(deps): bump opentelemetry-api in /clients/python (#1068)
+- fix(adopt): quote optional workflow flags
+- fix(adopt): preserve shallow history and caller-owned dependency references
+- fix(adopt): reconcile effective workspace pins and preserve proposal history
+
 ## [1.0.0] — 2026-10-26
 
 **1.0.0 ships the 0.223.0 API unchanged.** From here a breaking change is a major release, and a caret range (`^1.0.0`) takes every later compatible release, so a consumer no longer needs a release of its own for each minor of this package. The dependencies move onto the 1.0 releases of agent-core and harness-sessions. The commits below are everything since the v0.220.2 tag.

package/README.md

@@ -23,7 +23,7 @@ pnpm add @tangle-network/agent-eval
 
 ## Contents
 
-- [Quickstart](#quickstart) — score your first change in 30 seconds
+- [Quickstart](#quickstart) — score a change offline
 - [Choose a workflow](#choose-a-workflow) — which API for which job
 - [Automated improvement](#make-automated-improvement-accountable) — the held-out gate
 - [Configure model calls](#configure-model-calls) — transports and pricing
@@ -86,6 +86,7 @@ candidate: 1
 The baseline scores `0`; the candidate scores `1` on all three cases.
 These scores describe the three examples.
 They do not establish a release decision or performance on new tasks.
+For a release decision, including the cases where the evidence justifies no change, see the [held-out gate example](./examples/held-out-gate/).
 
 A **case** is one task.
 A **surface** is the prompt, skill, or configuration being changed.

package/dist/adapters/http.d.ts

@@ -1,6 +1,6 @@
 import { d as CostReceipt } from "../cost-ledger-Czro9wmo.js";
-import { R as Scenario, d as DispatchContext, f as DispatchFn } from "../types-C54Uk_a1.js";
-import "../index-aDPVYBIM.js";
+import { R as Scenario, d as DispatchContext, f as DispatchFn } from "../types-BF-xQGQn.js";
+import "../index-ByJ20lyu.js";
 export interface HttpDispatchOptions<TScenario extends Scenario, _TArtifact> {
   /** Static endpoint URL. Mutually exclusive with `resolveUrl`. */
   url?: string;

package/dist/{agreement-irr-BTjDN0i6.js → agreement-irr-w-t3XUTD.js}

@@ -1,5 +1,5 @@
 import { s as ValidationError } from "./errors-DeEEV9Vx.js";
-import { r as continuousAgreement } from "./judge-calibration-DGqlfWW8.js";
+import { r as continuousAgreement } from "./judge-calibration-D93UCm1k.js";
 // src/statistics/agreement-irr.ts
 /**
 * Inter-rater reliability — Krippendorff's α under the squared-difference
@@ -182,4 +182,4 @@ function nominalAgreement(pairs) {
 }
 export { nominalAgreement as i, corpusInterRaterAgreementFromJudgeScores as n, interRaterReliability as r, corpusInterRaterAgreement as t };
 
-//# sourceMappingURL=agreement-irr-BTjDN0i6.js.map
+//# sourceMappingURL=agreement-irr-w-t3XUTD.js.map

package/dist/{agreement-irr-BTjDN0i6.js.map → agreement-irr-w-t3XUTD.js.map}

@@ -1 +1 @@
-{"version":3,"file":"agreement-irr-BTjDN0i6.js","names":["prev"],"sources":["../src/statistics/agreement-irr.ts"],"sourcesContent":["import { ValidationError } from '../errors'\nimport {\n  type ContinuousAgreement,\n  type ContinuousAgreementOptions,\n  continuousAgreement,\n} from '../judge-calibration'\nimport type { DimensionJudgeScore } from '../types'\n\n/**\n * Inter-rater reliability — Krippendorff's α under the squared-difference\n * metric, pooled across dimensions.\n *\n * Each inner array is one judge's scores. Items are matched by position\n * WITHIN a dimension: the k-th score a judge supplies carrying dimension\n * `d` is item k of `d`, and the ratings compared against each other are\n * the ones different judges gave to the same item. Every judge that scores\n * a dimension at all must supply the same number of scores for it —\n * ragged input cannot be aligned into items and throws rather than\n * comparing mismatched items.\n *\n * α = 1 − D_observed / D_expected: D_observed averages the squared\n * difference over within-item judge pairs, D_expected over every pair of\n * ratings irrespective of item. α = 1 is perfect agreement, 0 is chance,\n * negative is systematic disagreement.\n */\nexport function interRaterReliability(judgeScores: DimensionJudgeScore[][]): number {\n  if (judgeScores.length < 2) return 1\n\n  // dimension → one score sequence per judge, in that judge's supplied order.\n  const perDimension = new Map<string, number[][]>()\n  for (let judgeIndex = 0; judgeIndex < judgeScores.length; judgeIndex++) {\n    for (const s of judgeScores[judgeIndex]!) {\n      let byJudge = perDimension.get(s.dimension)\n      if (byJudge === undefined) {\n        byJudge = Array.from({ length: judgeScores.length }, () => [] as number[])\n        perDimension.set(s.dimension, byJudge)\n      }\n      byJudge[judgeIndex]!.push(s.score)\n    }\n  }\n\n  const allValues: number[] = []\n  const pairDiffs: number[] = []\n\n  for (const [dimension, byJudge] of perDimension) {\n    const scoring = byJudge.filter((scores) => scores.length > 0)\n    if (scoring.length < 2) continue\n    const itemCount = scoring[0]!.length\n    if (scoring.some((scores) => scores.length !== itemCount)) {\n      throw new ValidationError(\n        `interRaterReliability: dimension '${dimension}' has judges supplying ` +\n          `${scoring.map((scores) => scores.length).join('/')} scores — items cannot be aligned`,\n      )\n    }\n    for (let item = 0; item < itemCount; item++) {\n      const ratings = scoring.map((scores) => scores[item]!)\n      for (const v of ratings) allValues.push(v)\n      for (let i = 0; i < ratings.length; i++) {\n        for (let j = i + 1; j < ratings.length; j++) {\n          pairDiffs.push((ratings[i]! - ratings[j]!) ** 2)\n        }\n      }\n    }\n  }\n\n  if (pairDiffs.length === 0 || allValues.length < 2) return 1\n\n  const observedDisagreement = pairDiffs.reduce((a, b) => a + b, 0) / pairDiffs.length\n\n  // Expected disagreement from all possible pairings of values\n  let expectedDisagreement = 0\n  let expectedCount = 0\n  for (let i = 0; i < allValues.length; i++) {\n    for (let j = i + 1; j < allValues.length; j++) {\n      expectedDisagreement += (allValues[i]! - allValues[j]!) ** 2\n      expectedCount++\n    }\n  }\n  expectedDisagreement = expectedCount > 0 ? expectedDisagreement / expectedCount : 0\n\n  if (expectedDisagreement === 0) return 1\n  return 1 - observedDisagreement / expectedDisagreement\n}\n\n// ── Corpus-wide inter-rater agreement ──────────────────────────────\n//\n// `interRaterReliability(judgeScores)` computes a within-item\n// Krippendorff α — multiple judges score *the same item* and we ask\n// \"how much do their scores agree on that item?\" Useful for a single\n// scenario, but it cannot answer \"how reliable are these judges across\n// the whole evaluation corpus?\"\n//\n// `corpusInterRaterAgreement` does the corpus-wide question properly.\n// Inputs are flat per-(item, judge, dimension) score records. For each\n// dimension we pivot to a complete [n_items × n_judges] matrix and feed\n// it to the ICC(2,1) + κ_w machinery already validated in\n// `judge-calibration.ts`. An overall pooled metric averages the\n// per-dimension ICC/κ across dimensions.\n\nexport interface CorpusScoreRecord {\n  /** Stable identifier for the rated item (scenario, span, turn, …). */\n  itemId: string\n  /** Identifier for the judge that produced this score. */\n  judgeName: string\n  /** Dimension name (matches `DimensionJudgeScore.dimension`). */\n  dimension: string\n  /** Numeric score; must be finite. */\n  score: number\n}\n\nexport interface CorpusAgreementPerDimension extends ContinuousAgreement {\n  dimension: string\n  /** Item IDs that contributed to this dimension's matrix (every judge scored them). */\n  itemIds: string[]\n  /** Judge IDs that contributed to this dimension's matrix. */\n  judgeIds: string[]\n}\n\nexport interface CorpusAgreementReport {\n  /** Per-dimension ICC(2,1) + κ_w + Pearson + Spearman + bootstrap CIs. */\n  perDimension: CorpusAgreementPerDimension[]\n  /** Mean ICC across dimensions (NaN if no dimension yielded a finite ICC). */\n  overallIcc: number\n  /** Mean weighted κ across dimensions (NaN if none finite). */\n  overallWeightedKappa: number\n  /** Dimensions evaluated (sorted). */\n  dimensions: string[]\n  /** Judges seen across the corpus (sorted). */\n  judgeIds: string[]\n}\n\nexport interface CorpusAgreementOptions extends ContinuousAgreementOptions {\n  /**\n   * Restrict the audit to these dimensions. Default = every dimension\n   * that appears in the input. A dimension named here but absent from\n   * the input throws — silent omission would corrupt the overall metric.\n   */\n  dimensions?: string[]\n  /**\n   * Restrict the audit to these judges. Default = every judge that\n   * appears in the input. A judge named here but absent from a\n   * dimension throws (see \"fail loud\" below).\n   */\n  judges?: string[]\n}\n\n/**\n * Corpus-wide inter-rater agreement across N items × M judges × D dimensions.\n *\n * For each dimension, builds the [n_items][n_judges] matrix of scores\n * (keeping only items every judge rated on that dimension), then runs\n * `continuousAgreement` to get ICC(2,1), κ_w, Pearson, Spearman, and\n * bootstrap CIs. Reports a pooled mean across dimensions as a single\n * \"is this judge panel reliable on this corpus?\" number.\n *\n * Fail-loud contract:\n *   - Empty input throws.\n *   - Fewer than 2 judges or fewer than 2 items per dimension throws.\n *   - A judge present in some dimensions but with zero scored items on\n *     another dimension throws (would silently shrink the matrix).\n *   - Duplicate (itemId, judgeName, dimension) records throw.\n */\nexport function corpusInterRaterAgreement(\n  records: CorpusScoreRecord[],\n  opts: CorpusAgreementOptions = {},\n): CorpusAgreementReport {\n  if (records.length === 0) {\n    throw new ValidationError('corpusInterRaterAgreement: no score records supplied')\n  }\n\n  const judgesSeen = new Set<string>()\n  const dimsSeen = new Set<string>()\n  // dimension → judge → itemId → score\n  const grid = new Map<string, Map<string, Map<string, number>>>()\n\n  for (const r of records) {\n    if (!Number.isFinite(r.score)) {\n      throw new ValidationError(\n        `corpusInterRaterAgreement: non-finite score for (item=${r.itemId}, judge=${r.judgeName}, dim=${r.dimension})`,\n      )\n    }\n    judgesSeen.add(r.judgeName)\n    dimsSeen.add(r.dimension)\n    const byJudge = grid.get(r.dimension) ?? new Map<string, Map<string, number>>()\n    const byItem = byJudge.get(r.judgeName) ?? new Map<string, number>()\n    if (byItem.has(r.itemId)) {\n      throw new ValidationError(\n        `corpusInterRaterAgreement: duplicate record for (item=${r.itemId}, judge=${r.judgeName}, dim=${r.dimension})`,\n      )\n    }\n    byItem.set(r.itemId, r.score)\n    byJudge.set(r.judgeName, byItem)\n    grid.set(r.dimension, byJudge)\n  }\n\n  const targetDims = opts.dimensions ?? [...dimsSeen].sort()\n  for (const d of targetDims) {\n    if (!dimsSeen.has(d)) {\n      throw new ValidationError(\n        `corpusInterRaterAgreement: dimension '${d}' was requested but no records carry it`,\n      )\n    }\n  }\n  const targetJudges = opts.judges ? [...opts.judges] : [...judgesSeen].sort()\n  for (const j of targetJudges) {\n    if (!judgesSeen.has(j)) {\n      throw new ValidationError(\n        `corpusInterRaterAgreement: judge '${j}' was requested but produced no records`,\n      )\n    }\n  }\n  if (targetJudges.length < 2) {\n    throw new ValidationError(\n      `corpusInterRaterAgreement: need ≥2 judges, got ${targetJudges.length}`,\n    )\n  }\n\n  const perDimension: CorpusAgreementPerDimension[] = []\n  const iccs: number[] = []\n  const kappas: number[] = []\n\n  for (const dim of targetDims) {\n    const byJudge = grid.get(dim)!\n    // Fail loud: every requested judge must have scored ≥1 item on this dim.\n    const judgeItemCounts: Record<string, number> = {}\n    for (const j of targetJudges) {\n      const m = byJudge.get(j)\n      judgeItemCounts[j] = m?.size ?? 0\n    }\n    const emptyJudges = targetJudges.filter((j) => judgeItemCounts[j] === 0)\n    if (emptyJudges.length > 0) {\n      throw new ValidationError(\n        `corpusInterRaterAgreement: dimension '${dim}' has no scores from judge(s) ${emptyJudges.join(', ')} (counts: ${JSON.stringify(judgeItemCounts)})`,\n      )\n    }\n\n    // Items rated by *every* requested judge on this dim.\n    let commonItems: Set<string> | null = null\n    for (const j of targetJudges) {\n      const ids = new Set(byJudge.get(j)!.keys())\n      if (commonItems === null) {\n        commonItems = ids\n      } else {\n        const prev: Set<string> = commonItems\n        commonItems = new Set([...prev].filter((x) => ids.has(x)))\n      }\n    }\n    const sortedItems = [...(commonItems ?? new Set<string>())].sort()\n    if (sortedItems.length < 2) {\n      throw new ValidationError(\n        `corpusInterRaterAgreement: dimension '${dim}' has ${sortedItems.length} item(s) rated by all ${targetJudges.length} judges (need ≥2)`,\n      )\n    }\n\n    const matrix: number[][] = sortedItems.map((itemId) =>\n      targetJudges.map((j) => byJudge.get(j)!.get(itemId)!),\n    )\n    const agreement = continuousAgreement(matrix, opts)\n    perDimension.push({\n      ...agreement,\n      dimension: dim,\n      itemIds: sortedItems,\n      judgeIds: [...targetJudges],\n    })\n    if (Number.isFinite(agreement.icc)) iccs.push(agreement.icc)\n    if (Number.isFinite(agreement.weightedKappa)) kappas.push(agreement.weightedKappa)\n  }\n\n  const mean = (xs: number[]) =>\n    xs.length === 0 ? Number.NaN : xs.reduce((a, b) => a + b, 0) / xs.length\n  return {\n    perDimension,\n    overallIcc: mean(iccs),\n    overallWeightedKappa: mean(kappas),\n    dimensions: targetDims,\n    judgeIds: targetJudges,\n  }\n}\n\n/**\n * Convenience adapter for `DimensionJudgeScore[]` data keyed externally by item.\n *\n * Use when you have per-item arrays of `DimensionJudgeScore[]` (e.g. one\n * `ScenarioResult.judgeScores` per scenario) and want corpus-wide\n * agreement without manually flattening. `itemId` must be unique per\n * row of `itemsScores`.\n */\nexport function corpusInterRaterAgreementFromJudgeScores(\n  itemsScores: Array<{ itemId: string; scores: DimensionJudgeScore[] }>,\n  opts: CorpusAgreementOptions = {},\n): CorpusAgreementReport {\n  const records: CorpusScoreRecord[] = []\n  const seen = new Set<string>()\n  for (const { itemId, scores } of itemsScores) {\n    if (seen.has(itemId)) {\n      throw new ValidationError(\n        `corpusInterRaterAgreementFromJudgeScores: duplicate itemId '${itemId}'`,\n      )\n    }\n    seen.add(itemId)\n    for (const s of scores) {\n      records.push({\n        itemId,\n        judgeName: s.judgeName,\n        dimension: s.dimension,\n        score: s.score,\n      })\n    }\n  }\n  return corpusInterRaterAgreement(records, opts)\n}\n\n/** Agreement between two raters who each assigned one nominal label per item. */\ninterface NominalAgreement {\n  /** Items both raters labeled. */\n  n: number\n  /** Share of items with identical labels. */\n  agreement: number\n  /**\n   * Cohen's κ = (pₒ − pₑ) / (1 − pₑ), with pₑ from each rater's own label\n   * marginals. NaN when n = 0, or when both raters used one identical label\n   * for every item (pₑ = 1): chance and observed agreement are then the same\n   * thing, and no number distinguishes skill from luck.\n   */\n  kappa: number\n}\n\n/**\n * Unweighted Cohen's κ for paired nominal labels. Labels carry no order, so a\n * near miss counts as a miss; use `calibrateJudge` for ordinal scores.\n */\nexport function nominalAgreement(\n  pairs: ReadonlyArray<readonly [string, string]>,\n): NominalAgreement {\n  const n = pairs.length\n  if (n === 0) return { n: 0, agreement: Number.NaN, kappa: Number.NaN }\n  const left = new Map<string, number>()\n  const right = new Map<string, number>()\n  let same = 0\n  for (const [a, b] of pairs) {\n    if (typeof a !== 'string' || typeof b !== 'string') {\n      throw new ValidationError('nominalAgreement: every label must be a string')\n    }\n    if (a === b) same += 1\n    left.set(a, (left.get(a) ?? 0) + 1)\n    right.set(b, (right.get(b) ?? 0) + 1)\n  }\n  const observed = same / n\n  let expected = 0\n  for (const [label, count] of left) expected += (count / n) * ((right.get(label) ?? 0) / n)\n  const kappa = expected === 1 ? Number.NaN : (observed - expected) / (1 - expected)\n  return { n, agreement: observed, kappa }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;AAyBA,SAAgB,sBAAsB,aAA8C;CAClF,IAAI,YAAY,SAAS,GAAG,OAAO;CAGnC,MAAM,+BAAe,IAAI,IAAwB;CACjD,KAAK,IAAI,aAAa,GAAG,aAAa,YAAY,QAAQ,cACxD,KAAK,MAAM,KAAK,YAAY,aAAc;EACxC,IAAI,UAAU,aAAa,IAAI,EAAE,SAAS;EAC1C,IAAI,YAAY,KAAA,GAAW;GACzB,UAAU,MAAM,KAAK,EAAE,QAAQ,YAAY,OAAO,SAAS,CAAC,CAAa;GACzE,aAAa,IAAI,EAAE,WAAW,OAAO;EACvC;EACA,QAAQ,WAAW,CAAE,KAAK,EAAE,KAAK;CACnC;CAGF,MAAM,YAAsB,CAAC;CAC7B,MAAM,YAAsB,CAAC;CAE7B,KAAK,MAAM,CAAC,WAAW,YAAY,cAAc;EAC/C,MAAM,UAAU,QAAQ,QAAQ,WAAW,OAAO,SAAS,CAAC;EAC5D,IAAI,QAAQ,SAAS,GAAG;EACxB,MAAM,YAAY,QAAQ,EAAE,CAAE;EAC9B,IAAI,QAAQ,MAAM,WAAW,OAAO,WAAW,SAAS,GACtD,MAAM,IAAI,gBACR,qCAAqC,UAAU,yBAC1C,QAAQ,KAAK,WAAW,OAAO,MAAM,CAAC,CAAC,KAAK,GAAG,EAAE,kCACxD;EAEF,KAAK,IAAI,OAAO,GAAG,OAAO,WAAW,QAAQ;GAC3C,MAAM,UAAU,QAAQ,KAAK,WAAW,OAAO,KAAM;GACrD,KAAK,MAAM,KAAK,SAAS,UAAU,KAAK,CAAC;GACzC,KAAK,IAAI,IAAI,GAAG,IAAI,QAAQ,QAAQ,KAClC,KAAK,IAAI,IAAI,IAAI,GAAG,IAAI,QAAQ,QAAQ,KACtC,UAAU,MAAM,QAAQ,KAAM,QAAQ,OAAQ,CAAC;EAGrD;CACF;CAEA,IAAI,UAAU,WAAW,KAAK,UAAU,SAAS,GAAG,OAAO;CAE3D,MAAM,uBAAuB,UAAU,QAAQ,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,UAAU;CAG9E,IAAI,uBAAuB;CAC3B,IAAI,gBAAgB;CACpB,KAAK,IAAI,IAAI,GAAG,IAAI,UAAU,QAAQ,KACpC,KAAK,IAAI,IAAI,IAAI,GAAG,IAAI,UAAU,QAAQ,KAAK;EAC7C,yBAAyB,UAAU,KAAM,UAAU,OAAQ;EAC3D;CACF;CAEF,uBAAuB,gBAAgB,IAAI,uBAAuB,gBAAgB;CAElF,IAAI,yBAAyB,GAAG,OAAO;CACvC,OAAO,IAAI,uBAAuB;AACpC;;;;;;;;;;;;;;;;;AAgFA,SAAgB,0BACd,SACA,OAA+B,CAAC,GACT;CACvB,IAAI,QAAQ,WAAW,GACrB,MAAM,IAAI,gBAAgB,sDAAsD;CAGlF,MAAM,6BAAa,IAAI,IAAY;CACnC,MAAM,2BAAW,IAAI,IAAY;CAEjC,MAAM,uBAAO,IAAI,IAA8C;CAE/D,KAAK,MAAM,KAAK,SAAS;EACvB,IAAI,CAAC,OAAO,SAAS,EAAE,KAAK,GAC1B,MAAM,IAAI,gBACR,yDAAyD,EAAE,OAAO,UAAU,EAAE,UAAU,QAAQ,EAAE,UAAU,EAC9G;EAEF,WAAW,IAAI,EAAE,SAAS;EAC1B,SAAS,IAAI,EAAE,SAAS;EACxB,MAAM,UAAU,KAAK,IAAI,EAAE,SAAS,qBAAK,IAAI,IAAiC;EAC9E,MAAM,SAAS,QAAQ,IAAI,EAAE,SAAS,qBAAK,IAAI,IAAoB;EACnE,IAAI,OAAO,IAAI,EAAE,MAAM,GACrB,MAAM,IAAI,gBACR,yDAAyD,EAAE,OAAO,UAAU,EAAE,UAAU,QAAQ,EAAE,UAAU,EAC9G;EAEF,OAAO,IAAI,EAAE,QAAQ,EAAE,KAAK;EAC5B,QAAQ,IAAI,EAAE,WAAW,MAAM;EAC/B,KAAK,IAAI,EAAE,WAAW,OAAO;CAC/B;CAEA,MAAM,aAAa,KAAK,cAAc,CAAC,GAAG,QAAQ,CAAC,CAAC,KAAK;CACzD,KAAK,MAAM,KAAK,YACd,IAAI,CAAC,SAAS,IAAI,CAAC,GACjB,MAAM,IAAI,gBACR,yCAAyC,EAAE,wCAC7C;CAGJ,MAAM,eAAe,KAAK,SAAS,CAAC,GAAG,KAAK,MAAM,IAAI,CAAC,GAAG,UAAU,CAAC,CAAC,KAAK;CAC3E,KAAK,MAAM,KAAK,cACd,IAAI,CAAC,WAAW,IAAI,CAAC,GACnB,MAAM,IAAI,gBACR,qCAAqC,EAAE,wCACzC;CAGJ,IAAI,aAAa,SAAS,GACxB,MAAM,IAAI,gBACR,kDAAkD,aAAa,QACjE;CAGF,MAAM,eAA8C,CAAC;CACrD,MAAM,OAAiB,CAAC;CACxB,MAAM,SAAmB,CAAC;CAE1B,KAAK,MAAM,OAAO,YAAY;EAC5B,MAAM,UAAU,KAAK,IAAI,GAAG;EAE5B,MAAM,kBAA0C,CAAC;EACjD,KAAK,MAAM,KAAK,cAEd,gBAAgB,KADN,QAAQ,IAAI,CACD,CAAC,EAAE,QAAQ;EAElC,MAAM,cAAc,aAAa,QAAQ,MAAM,gBAAgB,OAAO,CAAC;EACvE,IAAI,YAAY,SAAS,GACvB,MAAM,IAAI,gBACR,yCAAyC,IAAI,gCAAgC,YAAY,KAAK,IAAI,EAAE,YAAY,KAAK,UAAU,eAAe,EAAE,EAClJ;EAIF,IAAI,cAAkC;EACtC,KAAK,MAAM,KAAK,cAAc;GAC5B,MAAM,MAAM,IAAI,IAAI,QAAQ,IAAI,CAAC,CAAC,CAAE,KAAK,CAAC;GAC1C,IAAI,gBAAgB,MAClB,cAAc;QAGd,cAAc,IAAI,IAAI,CAAC,GAAGA,WAAI,CAAC,CAAC,QAAQ,MAAM,IAAI,IAAI,CAAC,CAAC,CAAC;EAE7D;EACA,MAAM,cAAc,CAAC,GAAI,+BAAe,IAAI,IAAY,CAAE,CAAC,CAAC,KAAK;EACjE,IAAI,YAAY,SAAS,GACvB,MAAM,IAAI,gBACR,yCAAyC,IAAI,QAAQ,YAAY,OAAO,wBAAwB,aAAa,OAAO,kBACtH;EAGF,MAAM,SAAqB,YAAY,KAAK,WAC1C,aAAa,KAAK,MAAM,QAAQ,IAAI,CAAC,CAAC,CAAE,IAAI,MAAM,CAAE,CACtD;EACA,MAAM,YAAY,oBAAoB,QAAQ,IAAI;EAClD,aAAa,KAAK;GAChB,GAAG;GACH,WAAW;GACX,SAAS;GACT,UAAU,CAAC,GAAG,YAAY;EAC5B,CAAC;EACD,IAAI,OAAO,SAAS,UAAU,GAAG,GAAG,KAAK,KAAK,UAAU,GAAG;EAC3D,IAAI,OAAO,SAAS,UAAU,aAAa,GAAG,OAAO,KAAK,UAAU,aAAa;CACnF;CAEA,MAAM,QAAQ,OACZ,GAAG,WAAW,IAAI,MAAa,GAAG,QAAQ,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,GAAG;CACpE,OAAO;EACL;EACA,YAAY,KAAK,IAAI;EACrB,sBAAsB,KAAK,MAAM;EACjC,YAAY;EACZ,UAAU;CACZ;AACF;;;;;;;;;AAUA,SAAgB,yCACd,aACA,OAA+B,CAAC,GACT;CACvB,MAAM,UAA+B,CAAC;CACtC,MAAM,uBAAO,IAAI,IAAY;CAC7B,KAAK,MAAM,EAAE,QAAQ,YAAY,aAAa;EAC5C,IAAI,KAAK,IAAI,MAAM,GACjB,MAAM,IAAI,gBACR,+DAA+D,OAAO,EACxE;EAEF,KAAK,IAAI,MAAM;EACf,KAAK,MAAM,KAAK,QACd,QAAQ,KAAK;GACX;GACA,WAAW,EAAE;GACb,WAAW,EAAE;GACb,OAAO,EAAE;EACX,CAAC;CAEL;CACA,OAAO,0BAA0B,SAAS,IAAI;AAChD;;;;;AAqBA,SAAgB,iBACd,OACkB;CAClB,MAAM,IAAI,MAAM;CAChB,IAAI,MAAM,GAAG,OAAO;EAAE,GAAG;EAAG,WAAW;EAAY,OAAO;CAAW;CACrE,MAAM,uBAAO,IAAI,IAAoB;CACrC,MAAM,wBAAQ,IAAI,IAAoB;CACtC,IAAI,OAAO;CACX,KAAK,MAAM,CAAC,GAAG,MAAM,OAAO;EAC1B,IAAI,OAAO,MAAM,YAAY,OAAO,MAAM,UACxC,MAAM,IAAI,gBAAgB,gDAAgD;EAE5E,IAAI,MAAM,GAAG,QAAQ;EACrB,KAAK,IAAI,IAAI,KAAK,IAAI,CAAC,KAAK,KAAK,CAAC;EAClC,MAAM,IAAI,IAAI,MAAM,IAAI,CAAC,KAAK,KAAK,CAAC;CACtC;CACA,MAAM,WAAW,OAAO;CACxB,IAAI,WAAW;CACf,KAAK,MAAM,CAAC,OAAO,UAAU,MAAM,YAAa,QAAQ,MAAO,MAAM,IAAI,KAAK,KAAK,KAAK;CAExF,OAAO;EAAE;EAAG,WAAW;EAAU,OADnB,aAAa,IAAI,OAAc,WAAW,aAAa,IAAI;CAClC;AACzC"}
+{"version":3,"file":"agreement-irr-w-t3XUTD.js","names":["prev"],"sources":["../src/statistics/agreement-irr.ts"],"sourcesContent":["import { ValidationError } from '../errors'\nimport {\n  type ContinuousAgreement,\n  type ContinuousAgreementOptions,\n  continuousAgreement,\n} from '../judge-calibration'\nimport type { DimensionJudgeScore } from '../types'\n\n/**\n * Inter-rater reliability — Krippendorff's α under the squared-difference\n * metric, pooled across dimensions.\n *\n * Each inner array is one judge's scores. Items are matched by position\n * WITHIN a dimension: the k-th score a judge supplies carrying dimension\n * `d` is item k of `d`, and the ratings compared against each other are\n * the ones different judges gave to the same item. Every judge that scores\n * a dimension at all must supply the same number of scores for it —\n * ragged input cannot be aligned into items and throws rather than\n * comparing mismatched items.\n *\n * α = 1 − D_observed / D_expected: D_observed averages the squared\n * difference over within-item judge pairs, D_expected over every pair of\n * ratings irrespective of item. α = 1 is perfect agreement, 0 is chance,\n * negative is systematic disagreement.\n */\nexport function interRaterReliability(judgeScores: DimensionJudgeScore[][]): number {\n  if (judgeScores.length < 2) return 1\n\n  // dimension → one score sequence per judge, in that judge's supplied order.\n  const perDimension = new Map<string, number[][]>()\n  for (let judgeIndex = 0; judgeIndex < judgeScores.length; judgeIndex++) {\n    for (const s of judgeScores[judgeIndex]!) {\n      let byJudge = perDimension.get(s.dimension)\n      if (byJudge === undefined) {\n        byJudge = Array.from({ length: judgeScores.length }, () => [] as number[])\n        perDimension.set(s.dimension, byJudge)\n      }\n      byJudge[judgeIndex]!.push(s.score)\n    }\n  }\n\n  const allValues: number[] = []\n  const pairDiffs: number[] = []\n\n  for (const [dimension, byJudge] of perDimension) {\n    const scoring = byJudge.filter((scores) => scores.length > 0)\n    if (scoring.length < 2) continue\n    const itemCount = scoring[0]!.length\n    if (scoring.some((scores) => scores.length !== itemCount)) {\n      throw new ValidationError(\n        `interRaterReliability: dimension '${dimension}' has judges supplying ` +\n          `${scoring.map((scores) => scores.length).join('/')} scores — items cannot be aligned`,\n      )\n    }\n    for (let item = 0; item < itemCount; item++) {\n      const ratings = scoring.map((scores) => scores[item]!)\n      for (const v of ratings) allValues.push(v)\n      for (let i = 0; i < ratings.length; i++) {\n        for (let j = i + 1; j < ratings.length; j++) {\n          pairDiffs.push((ratings[i]! - ratings[j]!) ** 2)\n        }\n      }\n    }\n  }\n\n  if (pairDiffs.length === 0 || allValues.length < 2) return 1\n\n  const observedDisagreement = pairDiffs.reduce((a, b) => a + b, 0) / pairDiffs.length\n\n  // Expected disagreement from all possible pairings of values\n  let expectedDisagreement = 0\n  let expectedCount = 0\n  for (let i = 0; i < allValues.length; i++) {\n    for (let j = i + 1; j < allValues.length; j++) {\n      expectedDisagreement += (allValues[i]! - allValues[j]!) ** 2\n      expectedCount++\n    }\n  }\n  expectedDisagreement = expectedCount > 0 ? expectedDisagreement / expectedCount : 0\n\n  if (expectedDisagreement === 0) return 1\n  return 1 - observedDisagreement / expectedDisagreement\n}\n\n// ── Corpus-wide inter-rater agreement ──────────────────────────────\n//\n// `interRaterReliability(judgeScores)` computes a within-item\n// Krippendorff α — multiple judges score *the same item* and we ask\n// \"how much do their scores agree on that item?\" Useful for a single\n// scenario, but it cannot answer \"how reliable are these judges across\n// the whole evaluation corpus?\"\n//\n// `corpusInterRaterAgreement` does the corpus-wide question properly.\n// Inputs are flat per-(item, judge, dimension) score records. For each\n// dimension we pivot to a complete [n_items × n_judges] matrix and feed\n// it to the ICC(2,1) + κ_w machinery already validated in\n// `judge-calibration.ts`. An overall pooled metric averages the\n// per-dimension ICC/κ across dimensions.\n\nexport interface CorpusScoreRecord {\n  /** Stable identifier for the rated item (scenario, span, turn, …). */\n  itemId: string\n  /** Identifier for the judge that produced this score. */\n  judgeName: string\n  /** Dimension name (matches `DimensionJudgeScore.dimension`). */\n  dimension: string\n  /** Numeric score; must be finite. */\n  score: number\n}\n\nexport interface CorpusAgreementPerDimension extends ContinuousAgreement {\n  dimension: string\n  /** Item IDs that contributed to this dimension's matrix (every judge scored them). */\n  itemIds: string[]\n  /** Judge IDs that contributed to this dimension's matrix. */\n  judgeIds: string[]\n}\n\nexport interface CorpusAgreementReport {\n  /** Per-dimension ICC(2,1) + κ_w + Pearson + Spearman + bootstrap CIs. */\n  perDimension: CorpusAgreementPerDimension[]\n  /** Mean ICC across dimensions (NaN if no dimension yielded a finite ICC). */\n  overallIcc: number\n  /** Mean weighted κ across dimensions (NaN if none finite). */\n  overallWeightedKappa: number\n  /** Dimensions evaluated (sorted). */\n  dimensions: string[]\n  /** Judges seen across the corpus (sorted). */\n  judgeIds: string[]\n}\n\nexport interface CorpusAgreementOptions extends ContinuousAgreementOptions {\n  /**\n   * Restrict the audit to these dimensions. Default = every dimension\n   * that appears in the input. A dimension named here but absent from\n   * the input throws — silent omission would corrupt the overall metric.\n   */\n  dimensions?: string[]\n  /**\n   * Restrict the audit to these judges. Default = every judge that\n   * appears in the input. A judge named here but absent from a\n   * dimension throws (see \"fail loud\" below).\n   */\n  judges?: string[]\n}\n\n/**\n * Corpus-wide inter-rater agreement across N items × M judges × D dimensions.\n *\n * For each dimension, builds the [n_items][n_judges] matrix of scores\n * (keeping only items every judge rated on that dimension), then runs\n * `continuousAgreement` to get ICC(2,1), κ_w, Pearson, Spearman, and\n * bootstrap CIs. Reports a pooled mean across dimensions as a single\n * \"is this judge panel reliable on this corpus?\" number.\n *\n * Fail-loud contract:\n *   - Empty input throws.\n *   - Fewer than 2 judges or fewer than 2 items per dimension throws.\n *   - A judge present in some dimensions but with zero scored items on\n *     another dimension throws (would silently shrink the matrix).\n *   - Duplicate (itemId, judgeName, dimension) records throw.\n */\nexport function corpusInterRaterAgreement(\n  records: CorpusScoreRecord[],\n  opts: CorpusAgreementOptions = {},\n): CorpusAgreementReport {\n  if (records.length === 0) {\n    throw new ValidationError('corpusInterRaterAgreement: no score records supplied')\n  }\n\n  const judgesSeen = new Set<string>()\n  const dimsSeen = new Set<string>()\n  // dimension → judge → itemId → score\n  const grid = new Map<string, Map<string, Map<string, number>>>()\n\n  for (const r of records) {\n    if (!Number.isFinite(r.score)) {\n      throw new ValidationError(\n        `corpusInterRaterAgreement: non-finite score for (item=${r.itemId}, judge=${r.judgeName}, dim=${r.dimension})`,\n      )\n    }\n    judgesSeen.add(r.judgeName)\n    dimsSeen.add(r.dimension)\n    const byJudge = grid.get(r.dimension) ?? new Map<string, Map<string, number>>()\n    const byItem = byJudge.get(r.judgeName) ?? new Map<string, number>()\n    if (byItem.has(r.itemId)) {\n      throw new ValidationError(\n        `corpusInterRaterAgreement: duplicate record for (item=${r.itemId}, judge=${r.judgeName}, dim=${r.dimension})`,\n      )\n    }\n    byItem.set(r.itemId, r.score)\n    byJudge.set(r.judgeName, byItem)\n    grid.set(r.dimension, byJudge)\n  }\n\n  const targetDims = opts.dimensions ?? [...dimsSeen].sort()\n  for (const d of targetDims) {\n    if (!dimsSeen.has(d)) {\n      throw new ValidationError(\n        `corpusInterRaterAgreement: dimension '${d}' was requested but no records carry it`,\n      )\n    }\n  }\n  const targetJudges = opts.judges ? [...opts.judges] : [...judgesSeen].sort()\n  for (const j of targetJudges) {\n    if (!judgesSeen.has(j)) {\n      throw new ValidationError(\n        `corpusInterRaterAgreement: judge '${j}' was requested but produced no records`,\n      )\n    }\n  }\n  if (targetJudges.length < 2) {\n    throw new ValidationError(\n      `corpusInterRaterAgreement: need ≥2 judges, got ${targetJudges.length}`,\n    )\n  }\n\n  const perDimension: CorpusAgreementPerDimension[] = []\n  const iccs: number[] = []\n  const kappas: number[] = []\n\n  for (const dim of targetDims) {\n    const byJudge = grid.get(dim)!\n    // Fail loud: every requested judge must have scored ≥1 item on this dim.\n    const judgeItemCounts: Record<string, number> = {}\n    for (const j of targetJudges) {\n      const m = byJudge.get(j)\n      judgeItemCounts[j] = m?.size ?? 0\n    }\n    const emptyJudges = targetJudges.filter((j) => judgeItemCounts[j] === 0)\n    if (emptyJudges.length > 0) {\n      throw new ValidationError(\n        `corpusInterRaterAgreement: dimension '${dim}' has no scores from judge(s) ${emptyJudges.join(', ')} (counts: ${JSON.stringify(judgeItemCounts)})`,\n      )\n    }\n\n    // Items rated by *every* requested judge on this dim.\n    let commonItems: Set<string> | null = null\n    for (const j of targetJudges) {\n      const ids = new Set(byJudge.get(j)!.keys())\n      if (commonItems === null) {\n        commonItems = ids\n      } else {\n        const prev: Set<string> = commonItems\n        commonItems = new Set([...prev].filter((x) => ids.has(x)))\n      }\n    }\n    const sortedItems = [...(commonItems ?? new Set<string>())].sort()\n    if (sortedItems.length < 2) {\n      throw new ValidationError(\n        `corpusInterRaterAgreement: dimension '${dim}' has ${sortedItems.length} item(s) rated by all ${targetJudges.length} judges (need ≥2)`,\n      )\n    }\n\n    const matrix: number[][] = sortedItems.map((itemId) =>\n      targetJudges.map((j) => byJudge.get(j)!.get(itemId)!),\n    )\n    const agreement = continuousAgreement(matrix, opts)\n    perDimension.push({\n      ...agreement,\n      dimension: dim,\n      itemIds: sortedItems,\n      judgeIds: [...targetJudges],\n    })\n    if (Number.isFinite(agreement.icc)) iccs.push(agreement.icc)\n    if (Number.isFinite(agreement.weightedKappa)) kappas.push(agreement.weightedKappa)\n  }\n\n  const mean = (xs: number[]) =>\n    xs.length === 0 ? Number.NaN : xs.reduce((a, b) => a + b, 0) / xs.length\n  return {\n    perDimension,\n    overallIcc: mean(iccs),\n    overallWeightedKappa: mean(kappas),\n    dimensions: targetDims,\n    judgeIds: targetJudges,\n  }\n}\n\n/**\n * Convenience adapter for `DimensionJudgeScore[]` data keyed externally by item.\n *\n * Use when you have per-item arrays of `DimensionJudgeScore[]` (e.g. one\n * `ScenarioResult.judgeScores` per scenario) and want corpus-wide\n * agreement without manually flattening. `itemId` must be unique per\n * row of `itemsScores`.\n */\nexport function corpusInterRaterAgreementFromJudgeScores(\n  itemsScores: Array<{ itemId: string; scores: DimensionJudgeScore[] }>,\n  opts: CorpusAgreementOptions = {},\n): CorpusAgreementReport {\n  const records: CorpusScoreRecord[] = []\n  const seen = new Set<string>()\n  for (const { itemId, scores } of itemsScores) {\n    if (seen.has(itemId)) {\n      throw new ValidationError(\n        `corpusInterRaterAgreementFromJudgeScores: duplicate itemId '${itemId}'`,\n      )\n    }\n    seen.add(itemId)\n    for (const s of scores) {\n      records.push({\n        itemId,\n        judgeName: s.judgeName,\n        dimension: s.dimension,\n        score: s.score,\n      })\n    }\n  }\n  return corpusInterRaterAgreement(records, opts)\n}\n\n/** Agreement between two raters who each assigned one nominal label per item. */\ninterface NominalAgreement {\n  /** Items both raters labeled. */\n  n: number\n  /** Share of items with identical labels. */\n  agreement: number\n  /**\n   * Cohen's κ = (pₒ − pₑ) / (1 − pₑ), with pₑ from each rater's own label\n   * marginals. NaN when n = 0, or when both raters used one identical label\n   * for every item (pₑ = 1): chance and observed agreement are then the same\n   * thing, and no number distinguishes skill from luck.\n   */\n  kappa: number\n}\n\n/**\n * Unweighted Cohen's κ for paired nominal labels. Labels carry no order, so a\n * near miss counts as a miss; use `calibrateJudge` for ordinal scores.\n */\nexport function nominalAgreement(\n  pairs: ReadonlyArray<readonly [string, string]>,\n): NominalAgreement {\n  const n = pairs.length\n  if (n === 0) return { n: 0, agreement: Number.NaN, kappa: Number.NaN }\n  const left = new Map<string, number>()\n  const right = new Map<string, number>()\n  let same = 0\n  for (const [a, b] of pairs) {\n    if (typeof a !== 'string' || typeof b !== 'string') {\n      throw new ValidationError('nominalAgreement: every label must be a string')\n    }\n    if (a === b) same += 1\n    left.set(a, (left.get(a) ?? 0) + 1)\n    right.set(b, (right.get(b) ?? 0) + 1)\n  }\n  const observed = same / n\n  let expected = 0\n  for (const [label, count] of left) expected += (count / n) * ((right.get(label) ?? 0) / n)\n  const kappa = expected === 1 ? Number.NaN : (observed - expected) / (1 - expected)\n  return { n, agreement: observed, kappa }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;AAyBA,SAAgB,sBAAsB,aAA8C;CAClF,IAAI,YAAY,SAAS,GAAG,OAAO;CAGnC,MAAM,+BAAe,IAAI,IAAwB;CACjD,KAAK,IAAI,aAAa,GAAG,aAAa,YAAY,QAAQ,cACxD,KAAK,MAAM,KAAK,YAAY,aAAc;EACxC,IAAI,UAAU,aAAa,IAAI,EAAE,SAAS;EAC1C,IAAI,YAAY,KAAA,GAAW;GACzB,UAAU,MAAM,KAAK,EAAE,QAAQ,YAAY,OAAO,SAAS,CAAC,CAAa;GACzE,aAAa,IAAI,EAAE,WAAW,OAAO;EACvC;EACA,QAAQ,WAAW,CAAE,KAAK,EAAE,KAAK;CACnC;CAGF,MAAM,YAAsB,CAAC;CAC7B,MAAM,YAAsB,CAAC;CAE7B,KAAK,MAAM,CAAC,WAAW,YAAY,cAAc;EAC/C,MAAM,UAAU,QAAQ,QAAQ,WAAW,OAAO,SAAS,CAAC;EAC5D,IAAI,QAAQ,SAAS,GAAG;EACxB,MAAM,YAAY,QAAQ,EAAE,CAAE;EAC9B,IAAI,QAAQ,MAAM,WAAW,OAAO,WAAW,SAAS,GACtD,MAAM,IAAI,gBACR,qCAAqC,UAAU,yBAC1C,QAAQ,KAAK,WAAW,OAAO,MAAM,CAAC,CAAC,KAAK,GAAG,EAAE,kCACxD;EAEF,KAAK,IAAI,OAAO,GAAG,OAAO,WAAW,QAAQ;GAC3C,MAAM,UAAU,QAAQ,KAAK,WAAW,OAAO,KAAM;GACrD,KAAK,MAAM,KAAK,SAAS,UAAU,KAAK,CAAC;GACzC,KAAK,IAAI,IAAI,GAAG,IAAI,QAAQ,QAAQ,KAClC,KAAK,IAAI,IAAI,IAAI,GAAG,IAAI,QAAQ,QAAQ,KACtC,UAAU,MAAM,QAAQ,KAAM,QAAQ,OAAQ,CAAC;EAGrD;CACF;CAEA,IAAI,UAAU,WAAW,KAAK,UAAU,SAAS,GAAG,OAAO;CAE3D,MAAM,uBAAuB,UAAU,QAAQ,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,UAAU;CAG9E,IAAI,uBAAuB;CAC3B,IAAI,gBAAgB;CACpB,KAAK,IAAI,IAAI,GAAG,IAAI,UAAU,QAAQ,KACpC,KAAK,IAAI,IAAI,IAAI,GAAG,IAAI,UAAU,QAAQ,KAAK;EAC7C,yBAAyB,UAAU,KAAM,UAAU,OAAQ;EAC3D;CACF;CAEF,uBAAuB,gBAAgB,IAAI,uBAAuB,gBAAgB;CAElF,IAAI,yBAAyB,GAAG,OAAO;CACvC,OAAO,IAAI,uBAAuB;AACpC;;;;;;;;;;;;;;;;;AAgFA,SAAgB,0BACd,SACA,OAA+B,CAAC,GACT;CACvB,IAAI,QAAQ,WAAW,GACrB,MAAM,IAAI,gBAAgB,sDAAsD;CAGlF,MAAM,6BAAa,IAAI,IAAY;CACnC,MAAM,2BAAW,IAAI,IAAY;CAEjC,MAAM,uBAAO,IAAI,IAA8C;CAE/D,KAAK,MAAM,KAAK,SAAS;EACvB,IAAI,CAAC,OAAO,SAAS,EAAE,KAAK,GAC1B,MAAM,IAAI,gBACR,yDAAyD,EAAE,OAAO,UAAU,EAAE,UAAU,QAAQ,EAAE,UAAU,EAC9G;EAEF,WAAW,IAAI,EAAE,SAAS;EAC1B,SAAS,IAAI,EAAE,SAAS;EACxB,MAAM,UAAU,KAAK,IAAI,EAAE,SAAS,qBAAK,IAAI,IAAiC;EAC9E,MAAM,SAAS,QAAQ,IAAI,EAAE,SAAS,qBAAK,IAAI,IAAoB;EACnE,IAAI,OAAO,IAAI,EAAE,MAAM,GACrB,MAAM,IAAI,gBACR,yDAAyD,EAAE,OAAO,UAAU,EAAE,UAAU,QAAQ,EAAE,UAAU,EAC9G;EAEF,OAAO,IAAI,EAAE,QAAQ,EAAE,KAAK;EAC5B,QAAQ,IAAI,EAAE,WAAW,MAAM;EAC/B,KAAK,IAAI,EAAE,WAAW,OAAO;CAC/B;CAEA,MAAM,aAAa,KAAK,cAAc,CAAC,GAAG,QAAQ,CAAC,CAAC,KAAK;CACzD,KAAK,MAAM,KAAK,YACd,IAAI,CAAC,SAAS,IAAI,CAAC,GACjB,MAAM,IAAI,gBACR,yCAAyC,EAAE,wCAC7C;CAGJ,MAAM,eAAe,KAAK,SAAS,CAAC,GAAG,KAAK,MAAM,IAAI,CAAC,GAAG,UAAU,CAAC,CAAC,KAAK;CAC3E,KAAK,MAAM,KAAK,cACd,IAAI,CAAC,WAAW,IAAI,CAAC,GACnB,MAAM,IAAI,gBACR,qCAAqC,EAAE,wCACzC;CAGJ,IAAI,aAAa,SAAS,GACxB,MAAM,IAAI,gBACR,kDAAkD,aAAa,QACjE;CAGF,MAAM,eAA8C,CAAC;CACrD,MAAM,OAAiB,CAAC;CACxB,MAAM,SAAmB,CAAC;CAE1B,KAAK,MAAM,OAAO,YAAY;EAC5B,MAAM,UAAU,KAAK,IAAI,GAAG;EAE5B,MAAM,kBAA0C,CAAC;EACjD,KAAK,MAAM,KAAK,cAEd,gBAAgB,KADN,QAAQ,IAAI,CACD,CAAC,EAAE,QAAQ;EAElC,MAAM,cAAc,aAAa,QAAQ,MAAM,gBAAgB,OAAO,CAAC;EACvE,IAAI,YAAY,SAAS,GACvB,MAAM,IAAI,gBACR,yCAAyC,IAAI,gCAAgC,YAAY,KAAK,IAAI,EAAE,YAAY,KAAK,UAAU,eAAe,EAAE,EAClJ;EAIF,IAAI,cAAkC;EACtC,KAAK,MAAM,KAAK,cAAc;GAC5B,MAAM,MAAM,IAAI,IAAI,QAAQ,IAAI,CAAC,CAAC,CAAE,KAAK,CAAC;GAC1C,IAAI,gBAAgB,MAClB,cAAc;QAGd,cAAc,IAAI,IAAI,CAAC,GAAGA,WAAI,CAAC,CAAC,QAAQ,MAAM,IAAI,IAAI,CAAC,CAAC,CAAC;EAE7D;EACA,MAAM,cAAc,CAAC,GAAI,+BAAe,IAAI,IAAY,CAAE,CAAC,CAAC,KAAK;EACjE,IAAI,YAAY,SAAS,GACvB,MAAM,IAAI,gBACR,yCAAyC,IAAI,QAAQ,YAAY,OAAO,wBAAwB,aAAa,OAAO,kBACtH;EAGF,MAAM,SAAqB,YAAY,KAAK,WAC1C,aAAa,KAAK,MAAM,QAAQ,IAAI,CAAC,CAAC,CAAE,IAAI,MAAM,CAAE,CACtD;EACA,MAAM,YAAY,oBAAoB,QAAQ,IAAI;EAClD,aAAa,KAAK;GAChB,GAAG;GACH,WAAW;GACX,SAAS;GACT,UAAU,CAAC,GAAG,YAAY;EAC5B,CAAC;EACD,IAAI,OAAO,SAAS,UAAU,GAAG,GAAG,KAAK,KAAK,UAAU,GAAG;EAC3D,IAAI,OAAO,SAAS,UAAU,aAAa,GAAG,OAAO,KAAK,UAAU,aAAa;CACnF;CAEA,MAAM,QAAQ,OACZ,GAAG,WAAW,IAAI,MAAa,GAAG,QAAQ,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,GAAG;CACpE,OAAO;EACL;EACA,YAAY,KAAK,IAAI;EACrB,sBAAsB,KAAK,MAAM;EACjC,YAAY;EACZ,UAAU;CACZ;AACF;;;;;;;;;AAUA,SAAgB,yCACd,aACA,OAA+B,CAAC,GACT;CACvB,MAAM,UAA+B,CAAC;CACtC,MAAM,uBAAO,IAAI,IAAY;CAC7B,KAAK,MAAM,EAAE,QAAQ,YAAY,aAAa;EAC5C,IAAI,KAAK,IAAI,MAAM,GACjB,MAAM,IAAI,gBACR,+DAA+D,OAAO,EACxE;EAEF,KAAK,IAAI,MAAM;EACf,KAAK,MAAM,KAAK,QACd,QAAQ,KAAK;GACX;GACA,WAAW,EAAE;GACb,WAAW,EAAE;GACb,OAAO,EAAE;EACX,CAAC;CAEL;CACA,OAAO,0BAA0B,SAAS,IAAI;AAChD;;;;;AAqBA,SAAgB,iBACd,OACkB;CAClB,MAAM,IAAI,MAAM;CAChB,IAAI,MAAM,GAAG,OAAO;EAAE,GAAG;EAAG,WAAW;EAAY,OAAO;CAAW;CACrE,MAAM,uBAAO,IAAI,IAAoB;CACrC,MAAM,wBAAQ,IAAI,IAAoB;CACtC,IAAI,OAAO;CACX,KAAK,MAAM,CAAC,GAAG,MAAM,OAAO;EAC1B,IAAI,OAAO,MAAM,YAAY,OAAO,MAAM,UACxC,MAAM,IAAI,gBAAgB,gDAAgD;EAE5E,IAAI,MAAM,GAAG,QAAQ;EACrB,KAAK,IAAI,IAAI,KAAK,IAAI,CAAC,KAAK,KAAK,CAAC;EAClC,MAAM,IAAI,IAAI,MAAM,IAAI,CAAC,KAAK,KAAK,CAAC;CACtC;CACA,MAAM,WAAW,OAAO;CACxB,IAAI,WAAW;CACf,KAAK,MAAM,CAAC,OAAO,UAAU,MAAM,YAAa,QAAQ,MAAO,MAAM,IAAI,KAAK,KAAK,KAAK;CAExF,OAAO;EAAE;EAAG,WAAW;EAAU,OADnB,aAAa,IAAI,OAAc,WAAW,aAAa,IAAI;CAClC;AACzC"}

package/dist/analyst/index.d.ts

@@ -2,17 +2,17 @@ import { m as CustomTokenPricing, o as CostLedgerHandle } from "../cost-ledger-C
 import { A as FindingReaderLabel, B as TREATMENT_LAYERS, C as FINDING_LAYERS, D as FindingMeasure, E as FindingLayer, F as lineageRoots, I as parseAnalystFinding, J as TraceAnalysisStore, L as readFindingScope, M as FindingScopeReading, N as ProposalFindingSchema, O as FindingMethod, P as independentUnitCount, R as FindingConditions, S as makeProposalFinding, T as FindingEvidenceUnit, V as TreatmentLayer, _ as ExecutionProbeRequest, a as AnalystFindingSchemaVersion, at as TraceAnalystSpan, b as computeFindingId, c as AnalystRunEvent, d as AnalystRunSummary, f as AnalystSeverity, g as ExecutionProbeOutcome, h as ExecutionProbe, i as AnalystFinding, j as FindingScope, k as FindingParseOutcome, l as AnalystRunInputs, m as EvidenceRef, n as AnalystContext, o as AnalystInputKind, p as AnalystUsageReceipt, r as AnalystCost, s as AnalystRequirements, t as Analyst, u as AnalystRunResult, v as ProposalFinding, w as FindingCalibration, x as makeFinding, y as ProposalFindingOrigin, z as FindingTreatment } from "../types-DzCO-S1m.js";
 import { A as createChatClient, C as ChatResponse, D as MockTransportOpts, E as CustomTransportOpts, O as OpenAiCompatibleTransportOpts, S as ChatRequest, T as CreateChatClientOpts, b as ChatCallOpts, k as SandboxSdkTransportOpts, w as ChatTransport, x as ChatClient } from "../types-0KglnQ9o.js";
 import { A as traceStoreEvidenceResolver, C as AnalystEvidenceResolutionError, D as AnalystLatencyDistribution, E as AnalystIssueExpectation, G as ExternalOptimizerRunnerCommand, H as ExternalOptimizerModelExecutionObservation, O as RunAnalystBenchmarkOptions, S as AnalystEvidenceResolution, T as AnalystFindingScore, _ as AnalystBenchmarkProvenance, b as AnalystBenchmarkSummary, d as AnalystBenchmarkDatasetRef, f as AnalystBenchmarkDescriptor, g as AnalystBenchmarkOutput, h as AnalystBenchmarkObservation, k as runAnalystBenchmark, m as AnalystBenchmarkLabelState, p as AnalystBenchmarkError, t as AgentProfile, u as AnalystBenchmarkCase, v as AnalystBenchmarkResult, w as AnalystEvidenceResolver, x as AnalystEvidenceExpectation, y as AnalystBenchmarkRunner, z as ExternalOptimizerModelCall } from "../agent-profile-CoD1Dotf.js";
+import { C as BehavioralWindowPolicy, S as BehavioralTraceWindow, _ as BehavioralTraceEventMap, a as createTraceAnalyst, b as BehavioralTraceQuestionsOptions, c as BehavioralAnalystOptions, d as pinBehavioralFailureClass, f as BEHAVIORAL_TRACE_QUESTIONS, g as BehavioralTraceEvent, h as BehavioralTraceAnswer, i as TraceAnalystDefinition, l as assertFailureClasses, m as BehavioralQuestionCalibration, n as buildDefaultAnalystRegistry, o as runTraceAnalyst, p as BEHAVIORAL_TRACE_QUESTIONS_VERSION, r as CreateTraceAnalystOptions, s as BEHAVIORAL_FAILURE_CLASS, t as DefaultAnalystRegistryOptions, u as behavioralAnalyst, v as BehavioralTraceLayer, w as behavioralTraceQuestionDigest, x as BehavioralTraceSupport, y as BehavioralTraceQuestionId } from "../default-registry-DKlKVXMx.js";
+import { a as ExactAnalystRunPolicySnapshot, c as ExactAnalystSnapshot, d as ExactExecutionComponentSnapshot, i as ExactAnalystRunEvent, l as ExactCapableAnalyst, n as ExactAnalystExecutionPlanSnapshot, o as ExactAnalystRunResult, r as ExactAnalystRunCompletion, s as ExactAnalystRunSummary, t as ExactAnalystBudgetSnapshot, u as ExactExecutionComponentIdentity } from "../exact-types-CIViGj98.js";
 import { g as SystemOneState } from "../systemone-protocol-C4t7JH4Y.js";
-import { A as FindingsLedger, B as FindingsDiff, F as LedgerFinding, H as PersistedFinding, I as diffFindingsOverTime, L as openFindingsLedger, M as FindingsLedgerRead, N as FindingsQuery, P as FindingsTimeStep, R as queryFindings, U as diffFindings, V as FindingsStore, j as FindingsLedgerOutcome, r as SystemOneEvaluatorOptions, z as DiffPolicy } from "../systemone-Cw2txK5m.js";
-import { a as ExactAnalystRunPolicySnapshot, c as ExactAnalystSnapshot, d as ExactExecutionComponentSnapshot, i as ExactAnalystRunEvent, l as ExactCapableAnalyst, n as ExactAnalystExecutionPlanSnapshot, o as ExactAnalystRunResult, r as ExactAnalystRunCompletion, s as ExactAnalystRunSummary, t as ExactAnalystBudgetSnapshot, u as ExactExecutionComponentIdentity } from "../exact-types-D98BKpxo.js";
-import { _ as BehavioralTraceQuestionsOptions, a as createTraceAnalyst, b as BehavioralWindowPolicy, c as behavioralAnalyst, d as BehavioralQuestionCalibration, f as BehavioralTraceAnswer, g as BehavioralTraceQuestionId, h as BehavioralTraceLayer, i as TraceAnalystDefinition, l as BEHAVIORAL_TRACE_QUESTIONS, m as BehavioralTraceEventMap, n as buildDefaultAnalystRegistry, o as runTraceAnalyst, p as BehavioralTraceEvent, r as CreateTraceAnalystOptions, s as BehavioralAnalystOptions, t as DefaultAnalystRegistryOptions, u as BEHAVIORAL_TRACE_QUESTIONS_VERSION, v as BehavioralTraceSupport, x as behavioralTraceQuestionDigest, y as BehavioralTraceWindow } from "../default-registry-B1DstPYg.js";
+import { A as FindingsLedger, B as FindingsDiff, F as LedgerFinding, H as PersistedFinding, I as diffFindingsOverTime, L as openFindingsLedger, M as FindingsLedgerRead, N as FindingsQuery, P as FindingsTimeStep, R as queryFindings, U as diffFindings, V as FindingsStore, j as FindingsLedgerOutcome, r as SystemOneEvaluatorOptions, z as DiffPolicy } from "../systemone-BDVgb6XV.js";
 import { a as RAW_FINDING_SCHEMA_PROMPT, c as RawAnalystFindingSchema, i as TraceAnalystLimits, l as evidenceRefsFromRawFinding, n as TraceAnalysisEngineRequest, o as RawAnalystEvidence, r as TraceAnalysisEngineResult, s as RawAnalystFinding, t as TraceAnalysisEngine, u as parseRawFinding } from "../engine-CaHaCc3W.js";
 import { n as buildTraceToolsForGroup, t as TraceToolGroupName } from "../tool-groups-C3K-d3gd.js";
-import { a as BudgetPolicy, c as RegistryRunOpts, i as AnalystRegistryOptions, n as ExactAnalystRunExecutionError, o as ExactAnalystBudgetPolicy, r as AnalystHooks, s as ExactRegistryRunOpts, t as AnalystRegistry } from "../registry-dhdDoAV8.js";
-import { T as SemanticConceptJudgeOptions, _ as DspyRlmTraceEngineOptions, a as FAILURE_MODE_KIND_SPEC, b as createChatTraceEngine, c as emitControlIntegrityFindings, d as FindingSubject, f as FindingSubjectKind, g as renderFindingSubject, h as parseFindingSubject, i as IMPROVEMENT_KIND_SPEC, l as FINDING_SUBJECT_KINDS, m as findingSubjectGrammarPromptFor, n as KNOWLEDGE_POISONING_KIND_SPEC, o as CONTROL_INTEGRITY_ANALYST, p as KIND_EXPECTED_SUBJECTS, r as KNOWLEDGE_GAP_KIND_SPEC, s as ControlIntegrityAnalyst, t as DEFAULT_TRACE_ANALYST_KINDS, u as FINDING_SUBJECT_SYNTAX, v as createDspyRlmTraceEngine, y as ChatTraceEngineOptions } from "../index-CUjeTDPv.js";
+import { a as BudgetPolicy, c as RegistryRunOpts, i as AnalystRegistryOptions, n as ExactAnalystRunExecutionError, o as ExactAnalystBudgetPolicy, r as AnalystHooks, s as ExactRegistryRunOpts, t as AnalystRegistry } from "../registry-CMtGAxOw.js";
+import { T as SemanticConceptJudgeOptions, _ as DspyRlmTraceEngineOptions, a as FAILURE_MODE_KIND_SPEC, b as createChatTraceEngine, c as emitControlIntegrityFindings, d as FindingSubject, f as FindingSubjectKind, g as renderFindingSubject, h as parseFindingSubject, i as IMPROVEMENT_KIND_SPEC, l as FINDING_SUBJECT_KINDS, m as findingSubjectGrammarPromptFor, n as KNOWLEDGE_POISONING_KIND_SPEC, o as CONTROL_INTEGRITY_ANALYST, p as KIND_EXPECTED_SUBJECTS, r as KNOWLEDGE_GAP_KIND_SPEC, s as ControlIntegrityAnalyst, t as DEFAULT_TRACE_ANALYST_KINDS, u as FINDING_SUBJECT_SYNTAX, v as createDspyRlmTraceEngine, y as ChatTraceEngineOptions } from "../index-Bn-P_XAl.js";
 import { C as runPrimeExchange, D as ReplyRowDecoded, E as ReplyEnvelope, O as ReplyRowRejection, S as projectPrimeTrajectory, T as ReplyContract, _ as buildPrimePrompt, a as PrimeProjectionOutcome, b as normalizePrimeUsage, c as PrimeProtocolIdentity, d as PrimeRepairPromptSpec, f as PrimeRepairState, g as analystUsageReceiptFromPrimeUsage, h as PrimeTurnRecord, i as PrimeProjectionDelivery, l as PrimeRawUsage, m as PrimeRowDecoded, n as PrimeExchangeOutcome, o as PrimeProjectionSource, p as PrimeReplyContract, r as PrimeFailure, s as PrimePromptSpec, t as PrimeExchangeOptions, u as PrimeRejectedRow, v as buildPrimeRepairPrompt, w as DecodedReply, x as primeProtocolSha256, y as mergePrimeRawUsage } from "../prime-protocol-COFkxas6.js";
 import { i as nodeHttpPrimeBridgeTransport, n as PrimeBridgeTransportRequest, r as PrimeBridgeTransportResult, t as PrimeBridgeTransport } from "../prime-bridge-transport-Cfqir_Im.js";
-import { b as SupervisorRunSources } from "../types-C-aOqgt2.js";
+import { b as SupervisorRunSources } from "../types-DNB9hIXa.js";
 import { a as askTraceQuestions, i as VerifiedTraceFinding, n as TraceQuestionOutcome, r as TraceQuestionSpec, t as AskTraceQuestionsOptions } from "../trace-questions-ej5AEJ5X.js";
 import { z } from "zod";
 interface SemanticConceptJudgeAdapterOpts {
@@ -1633,5 +1633,5 @@ export declare function classifyTeamShape(input: TeamShapeInput): TeamShapeOutco
  * timestamps fails the analyst instead of emitting a shape it did not measure.
  */
 export declare function teamShapeAnalyst(): Analyst<TeamShapeInput>;
-export { type AgentRxBenchmarkCaseOptions, type AgentRxCalibrationRunnerSummary, type AgentRxCalibrationSummary, type AgentRxFailure, type AgentRxPrediction, type AgentRxPredictionReport, type AgentRxRow, type Analyst, type AnalystBenchmarkArtifact, type AnalystBenchmarkCase, type AnalystBenchmarkCommandConfig, type AnalystBenchmarkCommandDependencies, type AnalystBenchmarkDatasetRef, type AnalystBenchmarkDescriptor, type AnalystBenchmarkError, type AnalystBenchmarkLabelState, type AnalystBenchmarkLocalRunReceipt, type AnalystBenchmarkObservation, type AnalystBenchmarkOutput, type AnalystBenchmarkProgressRow, type AnalystBenchmarkProvenance, type AnalystBenchmarkResult, type AnalystBenchmarkRunIdentity, type AnalystBenchmarkRunManifest, type AnalystBenchmarkRunner, type AnalystBenchmarkSummary, type AnalystBudgetDeclaration, type AnalystComparisonMetric, type AnalystContext, type AnalystCost, type AnalystDefinition, type AnalystDefinitionAsymmetry, type AnalystDefinitionAsymmetryReport, type AnalystEvidenceBinding, type AnalystEvidenceExpectation, type AnalystEvidenceResolution, type AnalystEvidenceResolutionError, type AnalystEvidenceResolver, type AnalystFinding, type AnalystFindingSchemaVersion, type AnalystFindingScore, type AnalystHooks, type AnalystInputKind, type AnalystInstructionsOverride, type AnalystIssueExpectation, type AnalystLatencyDistribution, type AnalystMetricComparison, type AnalystProfileFragment, AnalystRegistry, type AnalystRegistryOptions, type AnalystRepairDeclaration, type AnalystRequirements, type AnalystRowExpansion, type AnalystRunEvent, type AnalystRunInputs, type AnalystRunResult, type AnalystRunSummary, type AnalystRunnerComparison, type AnalystSeverity, type AnalystTransportBinding, type AnalystUsageReceipt, type AskTraceQuestionsOptions, BEHAVIORAL_TRACE_QUESTIONS, BEHAVIORAL_TRACE_QUESTIONS_VERSION, type BehavioralAnalystOptions, type BehavioralQuestionCalibration, type BehavioralTraceAnswer, type BehavioralTraceCalibrationRow, type BehavioralTraceEvent, type BehavioralTraceEventMap, type BehavioralTraceGold, type BehavioralTraceLayer, type BehavioralTraceQuestionId, type BehavioralTraceQuestionsOptions, type BehavioralTraceSupport, type BehavioralTraceWindow, type BehavioralWindowPolicy, type BudgetPolicy, CONTROL_INTEGRITY_ANALYST, type ChatCallOpts, type ChatClient, type ChatRequest, type ChatResponse, type ChatTraceEngineOptions, type ChatTransport, type ChunkedEvidenceBinding, type CodeTraceBenchCaseOptions, type CodeTraceBenchLabelOptions, type CodeTraceBenchLabelSet, type CodeTraceBenchRow, type CodeTraceBlockDiagnostics, type CodeTraceCalibrationRunnerSummary, type CodeTraceCalibrationSummary, type CodeTraceFailureBlock, type CodeTraceStageAnnotation, type CodeTracerLabelGroup, type CodeTracerPredictionAdapterOptions, type CodeTracerPredictions, type CodeTracerStepLabel, ControlIntegrityAnalyst, type CreateChatClientOpts, type CreateTraceAnalystOptions, type CustomTransportOpts, DEFAULT_TRACE_ANALYST_KINDS, type DecodedReply, type DefaultAnalystRegistryOptions, type DefineCustomAnalystOptions, type DefineExactCustomAnalystOptions, type DiffPolicy, type DomainTaxonomy, type DspyRlmTraceEngineOptions, type EvidenceProjection, type EvidenceRef, type ExactAnalystBudgetPolicy, type ExactAnalystBudgetSnapshot, type ExactAnalystExecutionPlanSnapshot, type ExactAnalystRunCompletion, type ExactAnalystRunEvent, ExactAnalystRunExecutionError, type ExactAnalystRunPolicySnapshot, type ExactAnalystRunResult, type ExactAnalystRunSummary, type ExactAnalystSnapshot, type ExactCapableAnalyst, type ExactExecutionComponentIdentity, type ExactExecutionComponentSnapshot, type ExactRegistryRunOpts, type ExecutionProbe, type ExecutionProbeOutcome, type ExecutionProbeRequest, type ExpandRowsArgs, FAILURE_MODE_KIND_SPEC, FINDING_LAYERS, FINDING_SUBJECT_KINDS, FINDING_SUBJECT_SYNTAX, type FailureModeGold, type FailureModeGoldSources, type FailureModeLabel, type FailureModeLabelItem, type FailureModesAnalystOptions, type FailureModesCalibration, type FailureModesOutcome, type FindingCalibration, type FindingConditions, type FindingEvidenceUnit, type FindingLayer, type FindingMeasure, type FindingMethod, type FindingParseOutcome, type FindingReaderLabel, type FindingScope, type FindingScopeReading, type FindingSubject, type FindingSubjectKind, type FindingTreatment, type FindingsDiff, type FindingsLedger, type FindingsLedgerOutcome, type FindingsLedgerRead, type FindingsQuery, FindingsStore, type FindingsTimeStep, IMPROVEMENT_KIND_SPEC, type InlineEvidenceBinding, KIND_EXPECTED_SUBJECTS, KNOWLEDGE_GAP_KIND_SPEC, KNOWLEDGE_POISONING_KIND_SPEC, type LedgerFinding, type LoadedVerificationArtifacts, type MockTransportOpts, type NotObservable, type OpenAiCompatibleTransportOpts, type PersistedFinding, type PreparedPublicAnalystBenchmark, type PrimeBenchmarkRunnerOptions, type PrimeBridgeTransport, type PrimeBridgeTransportRequest, type PrimeBridgeTransportResult, type PrimeCodeTraceDefinitionArgs, type PrimeExchangeOptions, type PrimeExchangeOutcome, type PrimeFailure, type PrimeProjectionDelivery, type PrimeProjectionOutcome, type PrimeProjectionSource, type PrimePromptSpec, type PrimeProtocolIdentity, type PrimeRawUsage, type PrimeRejectedRow, type PrimeRepairPromptSpec, type PrimeRepairState, type PrimeReplyContract, type PrimeRowDecoded, type PrimeTurnRecord, type PromoteOptions, type PromotionHold, type PromotionResult, type ProposalFinding, type ProposalFindingOrigin, ProposalFindingSchema, type PublicAnalystBenchmarkDataset, type PublicAnalystBenchmarkModelConfig, type PublicAnalystBenchmarkModelOwner, type PublicAnalystBenchmarkModelSettings, type PublicBenchmarkDistributions, type PublicBenchmarkModelPrediction, type PublicBenchmarkSelectionReport, type PublicBenchmarkValueDistribution, type PublicDirectDefinitionArgs, type PublicRlmDefinitionArgs, RAW_FINDING_SCHEMA_PROMPT, type RawAnalystEvidence, type RawAnalystFinding, RawAnalystFindingSchema, type RegistryRunOpts, type ReplVariableConsensusPort, type ReplVariableEvidenceBinding, type ReplyContract, type ReplyEnvelope, type ReplyRowDecoded, type ReplyRowRejection, type RollupInput, type RollupOptions, type RollupResult, type RunAnalystBenchmarkOptions, type SandboxSdkTransportOpts, type SemanticConceptJudgeAdapterOpts, type SkillUsageRecord, type SkillUsageReport, type SkillUsageScanConfig, type StepLabelAdapterOptions, TREATMENT_LAYERS, type TaxonomyAttachment, type TaxonomyNode, type TaxonomyNodeLayer, type TaxonomyOutcome, type TeamIntegration, type TeamShape, type TeamShapeCheck, type TeamShapeClass, type TeamShapeEdge, type TeamShapeFeatures, type TeamShapeInput, type TeamShapeObserved, type TeamShapeOutcome, type TraceAnalysisEngine, type TraceAnalysisEngineRequest, type TraceAnalysisEngineResult, type TraceAnalystDefinition, type TraceAnalystLimits, type TraceQuestionOutcome, type TraceQuestionSpec, type TraceToolGroupName, type TreatmentLayer, type UpstreamPredictionAdapterOptions, type VerificationArtifactFile, type VerificationArtifactManifest, type VerificationArtifactRole, type VerificationAvailabilitySummary, type VerificationOutcome, type VerificationOutcomeSource, type VerificationOutcomeStatus, type VerificationResultFile, type VerifiedTraceFinding, analystUsageReceiptFromPrimeUsage, askTraceQuestions, behavioralAnalyst, behavioralTraceQuestionDigest, buildDefaultAnalystRegistry, buildPrimePrompt, buildPrimeRepairPrompt, buildTraceToolsForGroup, computeFindingId, createChatClient, createChatTraceEngine, createDspyRlmTraceEngine, createTraceAnalyst, diffFindings, diffFindingsOverTime, emitControlIntegrityFindings, evidenceRefsFromRawFinding, findingSubjectGrammarPromptFor, independentUnitCount, lineageRoots, makeFinding, makeProposalFinding, mergePrimeRawUsage, nodeHttpPrimeBridgeTransport, normalizePrimeUsage, openFindingsLedger, parseAnalystFinding, parseFindingSubject, parseRawFinding, primeProtocolSha256, projectPrimeTrajectory, queryFindings, readFindingScope, renderFindingSubject, runAnalystBenchmark, runPrimeExchange, runTraceAnalyst, traceStoreEvidenceResolver };
+export { type AgentRxBenchmarkCaseOptions, type AgentRxCalibrationRunnerSummary, type AgentRxCalibrationSummary, type AgentRxFailure, type AgentRxPrediction, type AgentRxPredictionReport, type AgentRxRow, type Analyst, type AnalystBenchmarkArtifact, type AnalystBenchmarkCase, type AnalystBenchmarkCommandConfig, type AnalystBenchmarkCommandDependencies, type AnalystBenchmarkDatasetRef, type AnalystBenchmarkDescriptor, type AnalystBenchmarkError, type AnalystBenchmarkLabelState, type AnalystBenchmarkLocalRunReceipt, type AnalystBenchmarkObservation, type AnalystBenchmarkOutput, type AnalystBenchmarkProgressRow, type AnalystBenchmarkProvenance, type AnalystBenchmarkResult, type AnalystBenchmarkRunIdentity, type AnalystBenchmarkRunManifest, type AnalystBenchmarkRunner, type AnalystBenchmarkSummary, type AnalystBudgetDeclaration, type AnalystComparisonMetric, type AnalystContext, type AnalystCost, type AnalystDefinition, type AnalystDefinitionAsymmetry, type AnalystDefinitionAsymmetryReport, type AnalystEvidenceBinding, type AnalystEvidenceExpectation, type AnalystEvidenceResolution, type AnalystEvidenceResolutionError, type AnalystEvidenceResolver, type AnalystFinding, type AnalystFindingSchemaVersion, type AnalystFindingScore, type AnalystHooks, type AnalystInputKind, type AnalystInstructionsOverride, type AnalystIssueExpectation, type AnalystLatencyDistribution, type AnalystMetricComparison, type AnalystProfileFragment, AnalystRegistry, type AnalystRegistryOptions, type AnalystRepairDeclaration, type AnalystRequirements, type AnalystRowExpansion, type AnalystRunEvent, type AnalystRunInputs, type AnalystRunResult, type AnalystRunSummary, type AnalystRunnerComparison, type AnalystSeverity, type AnalystTransportBinding, type AnalystUsageReceipt, type AskTraceQuestionsOptions, BEHAVIORAL_FAILURE_CLASS, BEHAVIORAL_TRACE_QUESTIONS, BEHAVIORAL_TRACE_QUESTIONS_VERSION, type BehavioralAnalystOptions, type BehavioralQuestionCalibration, type BehavioralTraceAnswer, type BehavioralTraceCalibrationRow, type BehavioralTraceEvent, type BehavioralTraceEventMap, type BehavioralTraceGold, type BehavioralTraceLayer, type BehavioralTraceQuestionId, type BehavioralTraceQuestionsOptions, type BehavioralTraceSupport, type BehavioralTraceWindow, type BehavioralWindowPolicy, type BudgetPolicy, CONTROL_INTEGRITY_ANALYST, type ChatCallOpts, type ChatClient, type ChatRequest, type ChatResponse, type ChatTraceEngineOptions, type ChatTransport, type ChunkedEvidenceBinding, type CodeTraceBenchCaseOptions, type CodeTraceBenchLabelOptions, type CodeTraceBenchLabelSet, type CodeTraceBenchRow, type CodeTraceBlockDiagnostics, type CodeTraceCalibrationRunnerSummary, type CodeTraceCalibrationSummary, type CodeTraceFailureBlock, type CodeTraceStageAnnotation, type CodeTracerLabelGroup, type CodeTracerPredictionAdapterOptions, type CodeTracerPredictions, type CodeTracerStepLabel, ControlIntegrityAnalyst, type CreateChatClientOpts, type CreateTraceAnalystOptions, type CustomTransportOpts, DEFAULT_TRACE_ANALYST_KINDS, type DecodedReply, type DefaultAnalystRegistryOptions, type DefineCustomAnalystOptions, type DefineExactCustomAnalystOptions, type DiffPolicy, type DomainTaxonomy, type DspyRlmTraceEngineOptions, type EvidenceProjection, type EvidenceRef, type ExactAnalystBudgetPolicy, type ExactAnalystBudgetSnapshot, type ExactAnalystExecutionPlanSnapshot, type ExactAnalystRunCompletion, type ExactAnalystRunEvent, ExactAnalystRunExecutionError, type ExactAnalystRunPolicySnapshot, type ExactAnalystRunResult, type ExactAnalystRunSummary, type ExactAnalystSnapshot, type ExactCapableAnalyst, type ExactExecutionComponentIdentity, type ExactExecutionComponentSnapshot, type ExactRegistryRunOpts, type ExecutionProbe, type ExecutionProbeOutcome, type ExecutionProbeRequest, type ExpandRowsArgs, FAILURE_MODE_KIND_SPEC, FINDING_LAYERS, FINDING_SUBJECT_KINDS, FINDING_SUBJECT_SYNTAX, type FailureModeGold, type FailureModeGoldSources, type FailureModeLabel, type FailureModeLabelItem, type FailureModesAnalystOptions, type FailureModesCalibration, type FailureModesOutcome, type FindingCalibration, type FindingConditions, type FindingEvidenceUnit, type FindingLayer, type FindingMeasure, type FindingMethod, type FindingParseOutcome, type FindingReaderLabel, type FindingScope, type FindingScopeReading, type FindingSubject, type FindingSubjectKind, type FindingTreatment, type FindingsDiff, type FindingsLedger, type FindingsLedgerOutcome, type FindingsLedgerRead, type FindingsQuery, FindingsStore, type FindingsTimeStep, IMPROVEMENT_KIND_SPEC, type InlineEvidenceBinding, KIND_EXPECTED_SUBJECTS, KNOWLEDGE_GAP_KIND_SPEC, KNOWLEDGE_POISONING_KIND_SPEC, type LedgerFinding, type LoadedVerificationArtifacts, type MockTransportOpts, type NotObservable, type OpenAiCompatibleTransportOpts, type PersistedFinding, type PreparedPublicAnalystBenchmark, type PrimeBenchmarkRunnerOptions, type PrimeBridgeTransport, type PrimeBridgeTransportRequest, type PrimeBridgeTransportResult, type PrimeCodeTraceDefinitionArgs, type PrimeExchangeOptions, type PrimeExchangeOutcome, type PrimeFailure, type PrimeProjectionDelivery, type PrimeProjectionOutcome, type PrimeProjectionSource, type PrimePromptSpec, type PrimeProtocolIdentity, type PrimeRawUsage, type PrimeRejectedRow, type PrimeRepairPromptSpec, type PrimeRepairState, type PrimeReplyContract, type PrimeRowDecoded, type PrimeTurnRecord, type PromoteOptions, type PromotionHold, type PromotionResult, type ProposalFinding, type ProposalFindingOrigin, ProposalFindingSchema, type PublicAnalystBenchmarkDataset, type PublicAnalystBenchmarkModelConfig, type PublicAnalystBenchmarkModelOwner, type PublicAnalystBenchmarkModelSettings, type PublicBenchmarkDistributions, type PublicBenchmarkModelPrediction, type PublicBenchmarkSelectionReport, type PublicBenchmarkValueDistribution, type PublicDirectDefinitionArgs, type PublicRlmDefinitionArgs, RAW_FINDING_SCHEMA_PROMPT, type RawAnalystEvidence, type RawAnalystFinding, RawAnalystFindingSchema, type RegistryRunOpts, type ReplVariableConsensusPort, type ReplVariableEvidenceBinding, type ReplyContract, type ReplyEnvelope, type ReplyRowDecoded, type ReplyRowRejection, type RollupInput, type RollupOptions, type RollupResult, type RunAnalystBenchmarkOptions, type SandboxSdkTransportOpts, type SemanticConceptJudgeAdapterOpts, type SkillUsageRecord, type SkillUsageReport, type SkillUsageScanConfig, type StepLabelAdapterOptions, TREATMENT_LAYERS, type TaxonomyAttachment, type TaxonomyNode, type TaxonomyNodeLayer, type TaxonomyOutcome, type TeamIntegration, type TeamShape, type TeamShapeCheck, type TeamShapeClass, type TeamShapeEdge, type TeamShapeFeatures, type TeamShapeInput, type TeamShapeObserved, type TeamShapeOutcome, type TraceAnalysisEngine, type TraceAnalysisEngineRequest, type TraceAnalysisEngineResult, type TraceAnalystDefinition, type TraceAnalystLimits, type TraceQuestionOutcome, type TraceQuestionSpec, type TraceToolGroupName, type TreatmentLayer, type UpstreamPredictionAdapterOptions, type VerificationArtifactFile, type VerificationArtifactManifest, type VerificationArtifactRole, type VerificationAvailabilitySummary, type VerificationOutcome, type VerificationOutcomeSource, type VerificationOutcomeStatus, type VerificationResultFile, type VerifiedTraceFinding, analystUsageReceiptFromPrimeUsage, askTraceQuestions, assertFailureClasses, behavioralAnalyst, behavioralTraceQuestionDigest, buildDefaultAnalystRegistry, buildPrimePrompt, buildPrimeRepairPrompt, buildTraceToolsForGroup, computeFindingId, createChatClient, createChatTraceEngine, createDspyRlmTraceEngine, createTraceAnalyst, diffFindings, diffFindingsOverTime, emitControlIntegrityFindings, evidenceRefsFromRawFinding, findingSubjectGrammarPromptFor, independentUnitCount, lineageRoots, makeFinding, makeProposalFinding, mergePrimeRawUsage, nodeHttpPrimeBridgeTransport, normalizePrimeUsage, openFindingsLedger, parseAnalystFinding, parseFindingSubject, parseRawFinding, pinBehavioralFailureClass, primeProtocolSha256, projectPrimeTrajectory, queryFindings, readFindingScope, renderFindingSubject, runAnalystBenchmark, runPrimeExchange, runTraceAnalyst, traceStoreEvidenceResolver };
 //# sourceMappingURL=index.d.ts.map

package/dist/analyst/index.js

@@ -1,20 +1,20 @@
 import { n as assertConsistentFindingConditions, o as mergeFindingComposition, r as findingCompositionKey, t as TREATMENT_LAYERS } from "../finding-composition-DlvE1Qul.js";
-import { i as nominalAgreement } from "../agreement-irr-BTjDN0i6.js";
+import { i as nominalAgreement } from "../agreement-irr-w-t3XUTD.js";
 import { t as FileLedgerJournal } from "../journal-C1uLJi0w.js";
 import { r as merkleRoot } from "../ledger-core-CfT-FzPf.js";
-import { A as spanningWindow, C as UNKNOWN_LINEAGE, D as mostSevere, E as mergeEvidenceUnits, O as parseAnalystFinding, S as ProposalFindingSchema, T as lineageRoots, b as AnalystFindingSchema, k as readFindingScope, w as independentUnitCount, x as FINDING_LAYERS } from "../external-optimizer-subprocess-U616hV_l.js";
+import { A as spanningWindow, C as UNKNOWN_LINEAGE, D as mostSevere, E as mergeEvidenceUnits, O as parseAnalystFinding, S as ProposalFindingSchema, T as lineageRoots, b as AnalystFindingSchema, k as readFindingScope, w as independentUnitCount, x as FINDING_LAYERS } from "../external-optimizer-subprocess-wzMaA0oN.js";
 import { a as makeProposalFinding, i as makeFinding, r as computeFindingId } from "../types-BK0blWhe.js";
-import { t as assertProposalFindings } from "../proposal-findings-WFQrxnfe.js";
+import { t as assertProposalFindings } from "../proposal-findings-duV5flPv.js";
 import { A as RawAnalystFindingSchema, B as parseFindingSubject, F as stripCodeFences, I as FINDING_SUBJECT_KINDS, L as FINDING_SUBJECT_SYNTAX, M as parseRawFinding, N as parseTraceSpanEvidenceUri, R as KIND_EXPECTED_SUBJECTS, V as renderFindingSubject, j as evidenceRefsFromRawFinding, k as RAW_FINDING_SCHEMA_PROMPT, n as runTraceAnalyst, r as buildTraceToolsForGroup, t as createTraceAnalyst, z as findingSubjectGrammarPromptFor } from "../kind-factory-VKysp1ds.js";
-import { _ as behavioralTraceQuestionDigest, a as KNOWLEDGE_POISONING_KIND_SPEC, c as FAILURE_MODE_KIND_SPEC, d as emitControlIntegrityFindings, f as behavioralAnalyst, g as BEHAVIORAL_TRACE_QUESTIONS_VERSION, h as BEHAVIORAL_TRACE_QUESTIONS, i as DEFAULT_TRACE_ANALYST_KINDS, l as CONTROL_INTEGRITY_ANALYST, m as behavioralTraceSupportSchema, n as AnalystRegistry, o as KNOWLEDGE_GAP_KIND_SPEC, p as validBehavioralCalibration, r as ExactAnalystRunExecutionError, s as IMPROVEMENT_KIND_SPEC, t as buildDefaultAnalystRegistry, u as ControlIntegrityAnalyst } from "../default-registry-C1tPEjZ4.js";
-import { t as createDspyRlmTraceEngine } from "../dspy-rlm-engine-jDiWkul6.js";
-import { n as diffFindings, r as createChatTraceEngine, t as FindingsStore } from "../findings-store-B75Dspaq.js";
-import { a as systemOneAnalyst, p as validDecisionIdentity } from "../decision-point-lhMgODOR.js";
-import { t as createChatClient } from "../chat-client-CO3uB33i.js";
-import { A as agentRxBenchmarkCase, C as ANALYST_BENCHMARK_COST_LEDGER_FILE, D as AGENT_RX_UPSTREAM_REVISION, E as ANALYST_BENCHMARK_OBSERVATIONS_FILE, F as runAnalystBenchmark, I as traceStoreEvidenceResolver, M as normalizeAgentRxCategory, N as externalId, O as codeTraceBenchCase, P as positiveStep, S as ANALYST_BENCHMARK_IMPLEMENTATION_FILES, T as ANALYST_BENCHMARK_MANIFEST_FILE, _ as readAnalystBenchmarkArtifact, a as runInlineAnalystDefinition, b as ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256, c as publicRlmAnalystDefinition, d as publicDirectAnalystDefinition, f as runChunkedAnalystDefinition, g as expandCodeTraceFailureBlocks, h as analystDefinitionProtocolSha256, i as primeCodeTraceAnalystDefinition, j as agentRxPredictionsToFindings, k as codeTracerPredictionsToFindings, l as runReplVariableAnalystDefinition, m as analystDefinitionAsymmetries, n as createPrimeBenchmarkRunner, o as loadPublicBenchmarkRows, p as AnalystExpressivenessError, r as primeAnalystProtocolSha256, s as createPublicBenchmarkRlmRunner, t as runAnalystBenchmarkCommand, u as createPublicBenchmarkDirectRunner, v as CODE_TRACE_BENCH_ANALYST_PROMPT, w as ANALYST_BENCHMARK_LOCAL_RECEIPT_FILE, x as ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256, y as publicBenchmarkProtocolSha256 } from "../benchmark-command-DWkc5mzq.js";
+import { _ as behavioralTraceSupportSchema, a as KNOWLEDGE_POISONING_KIND_SPEC, b as behavioralTraceQuestionDigest, c as FAILURE_MODE_KIND_SPEC, d as emitControlIntegrityFindings, f as BEHAVIORAL_FAILURE_CLASS, g as validBehavioralCalibration, h as pinBehavioralFailureClass, i as DEFAULT_TRACE_ANALYST_KINDS, l as CONTROL_INTEGRITY_ANALYST, m as behavioralAnalyst, n as AnalystRegistry, o as KNOWLEDGE_GAP_KIND_SPEC, p as assertFailureClasses, r as ExactAnalystRunExecutionError, s as IMPROVEMENT_KIND_SPEC, t as buildDefaultAnalystRegistry, u as ControlIntegrityAnalyst, v as BEHAVIORAL_TRACE_QUESTIONS, y as BEHAVIORAL_TRACE_QUESTIONS_VERSION } from "../default-registry-gmDeN0KT.js";
+import { t as createDspyRlmTraceEngine } from "../dspy-rlm-engine-C0zR2xiM.js";
+import { n as diffFindings, r as createChatTraceEngine, t as FindingsStore } from "../findings-store-DNSvmy8U.js";
+import { a as systemOneAnalyst, p as validDecisionIdentity } from "../decision-point-CErE7GnP.js";
+import { t as createChatClient } from "../chat-client-D0me8iwq.js";
+import { A as agentRxBenchmarkCase, C as ANALYST_BENCHMARK_COST_LEDGER_FILE, D as AGENT_RX_UPSTREAM_REVISION, E as ANALYST_BENCHMARK_OBSERVATIONS_FILE, F as runAnalystBenchmark, I as traceStoreEvidenceResolver, M as normalizeAgentRxCategory, N as externalId, O as codeTraceBenchCase, P as positiveStep, S as ANALYST_BENCHMARK_IMPLEMENTATION_FILES, T as ANALYST_BENCHMARK_MANIFEST_FILE, _ as readAnalystBenchmarkArtifact, a as runInlineAnalystDefinition, b as ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256, c as publicRlmAnalystDefinition, d as publicDirectAnalystDefinition, f as runChunkedAnalystDefinition, g as expandCodeTraceFailureBlocks, h as analystDefinitionProtocolSha256, i as primeCodeTraceAnalystDefinition, j as agentRxPredictionsToFindings, k as codeTracerPredictionsToFindings, l as runReplVariableAnalystDefinition, m as analystDefinitionAsymmetries, n as createPrimeBenchmarkRunner, o as loadPublicBenchmarkRows, p as AnalystExpressivenessError, r as primeAnalystProtocolSha256, s as createPublicBenchmarkRlmRunner, t as runAnalystBenchmarkCommand, u as createPublicBenchmarkDirectRunner, v as CODE_TRACE_BENCH_ANALYST_PROMPT, w as ANALYST_BENCHMARK_LOCAL_RECEIPT_FILE, x as ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256, y as publicBenchmarkProtocolSha256 } from "../benchmark-command-BeDZB3BQ.js";
 import { a as normalizePrimeUsage, c as runPrimeExchange, i as mergePrimeRawUsage, n as buildPrimePrompt, o as primeProtocolSha256, r as buildPrimeRepairPrompt, s as projectPrimeTrajectory, t as analystUsageReceiptFromPrimeUsage } from "../prime-protocol-CrhpXRz9.js";
 import { t as nodeHttpPrimeBridgeTransport } from "../prime-bridge-transport-Csd5SQg7.js";
-import { t as clusterBootstrapMean } from "../cluster-bootstrap-SbaeM44s.js";
+import { t as clusterBootstrapMean } from "../cluster-bootstrap-hNXESoc1.js";
 import { z } from "zod";
 import { existsSync, readFileSync, readdirSync, statSync } from "node:fs";
 import { join } from "node:path";
@@ -2110,6 +2110,6 @@ function failed(question, failure) {
 function errorText(error) {
 	return error instanceof Error ? `${error.name}: ${error.message}` : String(error);
 }
-export { AGENT_RX_UPSTREAM_REVISION, ANALYST_BENCHMARK_COST_LEDGER_FILE, ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256, ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256, ANALYST_BENCHMARK_IMPLEMENTATION_FILES, ANALYST_BENCHMARK_LOCAL_RECEIPT_FILE, ANALYST_BENCHMARK_MANIFEST_FILE, ANALYST_BENCHMARK_OBSERVATIONS_FILE, AnalystExpressivenessError, AnalystRegistry, BEHAVIORAL_TRACE_QUESTIONS, BEHAVIORAL_TRACE_QUESTIONS_VERSION, CODE_TRACE_BENCH_ANALYST_PROMPT, CONTROL_INTEGRITY_ANALYST, ControlIntegrityAnalyst, DEFAULT_TRACE_ANALYST_KINDS, ExactAnalystRunExecutionError, FAILURE_MODES_ANALYST_ID, FAILURE_MODES_MIN_KAPPA, FAILURE_MODES_QUESTIONS, FAILURE_MODES_VERSION, FAILURE_MODE_KIND_SPEC, FAILURE_MODE_LABELS, FINDING_LAYERS, FINDING_SUBJECT_KINDS, FINDING_SUBJECT_SYNTAX, FindingsStore, IMPROVEMENT_KIND_SPEC, KIND_EXPECTED_SUBJECTS, KNOWLEDGE_GAP_KIND_SPEC, KNOWLEDGE_POISONING_KIND_SPEC, LAB_NODE_ID, NOT_OBSERVABLE, ProposalFindingSchema, RAW_FINDING_SCHEMA_PROMPT, RawAnalystFindingSchema, TEAM_SHAPE_ANALYST_ID, TEAM_SHAPE_CLASSES, TEAM_SHAPE_VERSION, TREATMENT_LAYERS, agentRxBenchmarkCase, agentRxPredictionsToFindings, analystDefinitionAsymmetries, analystDefinitionProtocolSha256, analystUsageReceiptFromPrimeUsage, askTraceQuestions, assertProposalFindings, attachToTaxonomy, behavioralAnalyst, behavioralTraceQuestionDigest, bindAnalyst, buildDefaultAnalystRegistry, buildPrimePrompt, buildPrimeRepairPrompt, buildSkillUsageReport, buildTraceToolsForGroup, calibrateBehavioralTraceQuestions, calibrateFailureModes, classifyTeamShape, codeTraceBenchCase, codeTracerPredictionsToFindings, computeFindingId, createChatClient, createChatTraceEngine, createDspyRlmTraceEngine, createFailureModesAnalyst, createPrimeBenchmarkRunner, createPublicBenchmarkDirectRunner, createPublicBenchmarkRlmRunner, createTraceAnalyst, defineCustomAnalyst, defineDomainTaxonomy, defineTraceAnalyst, diffFindings, diffFindingsOverTime, domainPathOf, emitControlIntegrityFindings, evidenceRefsFromRawFinding, expandCodeTraceFailureBlocks, findingSubjectGrammarPromptFor, independentUnitCount, lineageRoots, loadFailureModeGold, loadPublicBenchmarkRows, lowestCommonAncestor, makeFinding, makeProposalFinding, mergePrimeRawUsage, nodeHttpPrimeBridgeTransport, normalizePrimeUsage, openFindingsLedger, parseAnalystFinding, parseFindingSubject, parseRawFinding, primeAnalystProtocolSha256, primeCodeTraceAnalystDefinition, primeProtocolSha256, projectPrimeTrajectory, promote, publicBenchmarkProtocolSha256, publicDirectAnalystDefinition, publicRlmAnalystDefinition, queryFindings, readAnalystBenchmarkArtifact, readFindingScope, renderFindingSubject, resolveDomainPath, rollupFindings, runAnalystBenchmark, runAnalystBenchmarkCommand, runPrimeExchange, runTraceAnalyst, stripCodeFences, taxonomyAncestry, taxonomyLayerOf, teamShapeAnalyst, teamShapeInputFromSources, traceStoreEvidenceResolver };
+export { AGENT_RX_UPSTREAM_REVISION, ANALYST_BENCHMARK_COST_LEDGER_FILE, ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256, ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256, ANALYST_BENCHMARK_IMPLEMENTATION_FILES, ANALYST_BENCHMARK_LOCAL_RECEIPT_FILE, ANALYST_BENCHMARK_MANIFEST_FILE, ANALYST_BENCHMARK_OBSERVATIONS_FILE, AnalystExpressivenessError, AnalystRegistry, BEHAVIORAL_FAILURE_CLASS, BEHAVIORAL_TRACE_QUESTIONS, BEHAVIORAL_TRACE_QUESTIONS_VERSION, CODE_TRACE_BENCH_ANALYST_PROMPT, CONTROL_INTEGRITY_ANALYST, ControlIntegrityAnalyst, DEFAULT_TRACE_ANALYST_KINDS, ExactAnalystRunExecutionError, FAILURE_MODES_ANALYST_ID, FAILURE_MODES_MIN_KAPPA, FAILURE_MODES_QUESTIONS, FAILURE_MODES_VERSION, FAILURE_MODE_KIND_SPEC, FAILURE_MODE_LABELS, FINDING_LAYERS, FINDING_SUBJECT_KINDS, FINDING_SUBJECT_SYNTAX, FindingsStore, IMPROVEMENT_KIND_SPEC, KIND_EXPECTED_SUBJECTS, KNOWLEDGE_GAP_KIND_SPEC, KNOWLEDGE_POISONING_KIND_SPEC, LAB_NODE_ID, NOT_OBSERVABLE, ProposalFindingSchema, RAW_FINDING_SCHEMA_PROMPT, RawAnalystFindingSchema, TEAM_SHAPE_ANALYST_ID, TEAM_SHAPE_CLASSES, TEAM_SHAPE_VERSION, TREATMENT_LAYERS, agentRxBenchmarkCase, agentRxPredictionsToFindings, analystDefinitionAsymmetries, analystDefinitionProtocolSha256, analystUsageReceiptFromPrimeUsage, askTraceQuestions, assertFailureClasses, assertProposalFindings, attachToTaxonomy, behavioralAnalyst, behavioralTraceQuestionDigest, bindAnalyst, buildDefaultAnalystRegistry, buildPrimePrompt, buildPrimeRepairPrompt, buildSkillUsageReport, buildTraceToolsForGroup, calibrateBehavioralTraceQuestions, calibrateFailureModes, classifyTeamShape, codeTraceBenchCase, codeTracerPredictionsToFindings, computeFindingId, createChatClient, createChatTraceEngine, createDspyRlmTraceEngine, createFailureModesAnalyst, createPrimeBenchmarkRunner, createPublicBenchmarkDirectRunner, createPublicBenchmarkRlmRunner, createTraceAnalyst, defineCustomAnalyst, defineDomainTaxonomy, defineTraceAnalyst, diffFindings, diffFindingsOverTime, domainPathOf, emitControlIntegrityFindings, evidenceRefsFromRawFinding, expandCodeTraceFailureBlocks, findingSubjectGrammarPromptFor, independentUnitCount, lineageRoots, loadFailureModeGold, loadPublicBenchmarkRows, lowestCommonAncestor, makeFinding, makeProposalFinding, mergePrimeRawUsage, nodeHttpPrimeBridgeTransport, normalizePrimeUsage, openFindingsLedger, parseAnalystFinding, parseFindingSubject, parseRawFinding, pinBehavioralFailureClass, primeAnalystProtocolSha256, primeCodeTraceAnalystDefinition, primeProtocolSha256, projectPrimeTrajectory, promote, publicBenchmarkProtocolSha256, publicDirectAnalystDefinition, publicRlmAnalystDefinition, queryFindings, readAnalystBenchmarkArtifact, readFindingScope, renderFindingSubject, resolveDomainPath, rollupFindings, runAnalystBenchmark, runAnalystBenchmarkCommand, runPrimeExchange, runTraceAnalyst, stripCodeFences, taxonomyAncestry, taxonomyLayerOf, teamShapeAnalyst, teamShapeInputFromSources, traceStoreEvidenceResolver };
 
 //# sourceMappingURL=index.js.map

package/dist/{analyze-runs-C-pP1sK9.d.ts → analyze-runs-BYyMShsj.d.ts}

@@ -3,7 +3,7 @@ import { a as RunRecord, u as RunTerminalOutcome } from "./run-record-D2BFMPGl.j
 import { r as ContinuousAgreement } from "./judge-calibration-DnrbQaXI.js";
 import { r as DatasetScenario } from "./dataset-DThC1S6G.js";
 import { f as ParetoFigureSpec, l as GainDistributionBin } from "./summary-report-0La5vejY.js";
-import { t as AnalystRegistry } from "./registry-dhdDoAV8.js";
+import { t as AnalystRegistry } from "./registry-CMtGAxOw.js";
 interface InsightReport {
   /** Number of runs analyzed. */
   n: number;
@@ -480,4 +480,4 @@ interface AnalyzeRunsOptions {
 }
 declare function analyzeRuns(opts: AnalyzeRunsOptions): Promise<InsightReport>;
 export { ReleaseSummary as _, summarizeExecution as a, ExecutionInsight as c, InsightReport as d, InterRaterInsight as f, Recommendation as g, OutcomeCorrelationInsight as h, SummarizeExecutionOptions as i, FailureClassTally as l, LiftInsight as m, analyzeRuns as n, CostProvenanceSummary as o, JudgeInsight as p, ExecutionReport as r, ExecutionErrorOutcomeCell as s, AnalyzeRunsOptions as t, FailureClusterInsight as u, ScalarDistribution as v, TokenUsageInsight as y };
-//# sourceMappingURL=analyze-runs-C-pP1sK9.d.ts.map
+//# sourceMappingURL=analyze-runs-BYyMShsj.d.ts.map

package/dist/{analyze-runs-C-pP1sK9.d.ts.map → analyze-runs-BYyMShsj.d.ts.map}

@@ -1 +1 @@
-{"version":3,"file":"analyze-runs-C-pP1sK9.d.ts","names":[],"sources":["../src/contract/insight-report.ts","../src/contract/execution-insight.ts","../src/contract/analyze-runs.ts"],"mappings":";;;;;;UAqCiB;;EAEf;;;;EAKA,WAAW;;EAGX,WAAW;;;EAIX,cAAc,eAAe;;EAG7B;IACE,MAAM;IACN,QAAQ;;;;;IAKR,aAAa;;;;;IAKb;MAAa;MAAe;;;;;;EAM9B,QAAQ,eAAe;;;;EAKvB,aAAa;;;;EAKb,OAAO;;;EAIP,kBAAkB;;;EAIlB,gBAAgB;;;;;EAMhB,qBAAqB;;;;;EAMrB,SAAS;;;;;;EAOT,wBAAwB;;;;EAKxB,iBAAiB;;;EAIjB,iBAAiB;;UAGF;EACf;IAAY;IAAW;;EACvB;IAAa;IAAW;;;;EAGxB;IAAc;IAAW;;EACzB;IAAc;;EACd;;UAGe;;EAEf,YAAY;;EAEZ,SAAS;;;EAGT,YAAY;;;EAGZ;IACE;IACA,YAAY;IACZ,SAAS;IACT;;;EAGF,QAAQ;IAAQ;IAAe;;;;;EAI/B;IACE;IACA;IACA;;;;;EAKF;IACE;;;IAGA;;IAEA;;IAEA;;IAEA;;IAEA;;;;;IAKA,mBAAmB,OAAO,oBAAoB;;;;EAIhD;IACE;IACA;IACA;IACA;IACA;;;UAIa;;EAEf;;EAEA;;EAEA;;UAGe;EACf,OAAO;EACP,QAAQ;EACR,WAAW;EACX,QAAQ;EACR,YAAY;EACZ;IACE;IACA;IACA;IACA;IACA;;;;UAOa;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,WAAW;;;;;EAKX,WAAW;IAAQ;IAAe;;;UAGnB;;EAEf;;EAEA;;;EAGA,cAAc;;;EAGd;;;EAGA;;;EAGA;;UAGe;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,SAAS;;EAET,mBAAmB;IACjB;IACA,SAAS;MAAQ;MAAe;;IAChC;;;UAIa;EACf;EACA;;EAEA;;EAEA;;;EAGA;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;;;EAWA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;UAGe;;EAEf,UAAU;IACR;IACA;;IAEA;;IAEA;;IAEA;;EAEF;;;;;UAMe;;EAEf,cAAc;;EAEd;;EAEA;;UAGe;;EAEf;;;EAGA;EACA,UAAU;IAAQ;IAAe;IAAgB;;;UAGlC;;;EAGf;;EAEA;;EAEA;;EAEA;;EAEA;IACE;IACA;IACA;;;UAIa;;;EAGf;EACA,MAAM;IACJ;IACA;IACA;;;;EAIF;;UAGQ;;EAER;;EAEA;;;;EAIA;;EAEA;EACA;;KAGU,eACP;EACC;;EAEA;;EAEA;;EAEA;;EAEA;MAED;EACC;;EAEA;EACA;EACA;EACA;;UAGW;;EAEf;EACA;;EAEA;;;;EAIA,SAAS,eAAe;;;EAGxB;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;EACA;;EAEA;;UCzbe;EACf,MAAM;EACN;;UAGe;EACf,WAAW;EACX,gBAAgB;;;iBAIF,mBAAmB,MAAM,4BAA4B;UC4BpD;;EAEf,MAAM;;;EAGN;;;;;EAKA;EACA;;;EAGA,8BAA8B;;;EAG9B,kBAAkB;;;EAGlB,UAAU;;;;EAIV;IACE;IACA,cAAc;;;;;EAKhB,cAAc;IAAQ;IAAe;IAAe;;;EAEpD;;;;EAIA;;;;;;;;;EASA,eAAe;;;EAGf;;iBAGoB,YAAY,MAAM,qBAAqB,QAAQ"}
+{"version":3,"file":"analyze-runs-BYyMShsj.d.ts","names":[],"sources":["../src/contract/insight-report.ts","../src/contract/execution-insight.ts","../src/contract/analyze-runs.ts"],"mappings":";;;;;;UAqCiB;;EAEf;;;;EAKA,WAAW;;EAGX,WAAW;;;EAIX,cAAc,eAAe;;EAG7B;IACE,MAAM;IACN,QAAQ;;;;;IAKR,aAAa;;;;;IAKb;MAAa;MAAe;;;;;;EAM9B,QAAQ,eAAe;;;;EAKvB,aAAa;;;;EAKb,OAAO;;;EAIP,kBAAkB;;;EAIlB,gBAAgB;;;;;EAMhB,qBAAqB;;;;;EAMrB,SAAS;;;;;;EAOT,wBAAwB;;;;EAKxB,iBAAiB;;;EAIjB,iBAAiB;;UAGF;EACf;IAAY;IAAW;;EACvB;IAAa;IAAW;;;;EAGxB;IAAc;IAAW;;EACzB;IAAc;;EACd;;UAGe;;EAEf,YAAY;;EAEZ,SAAS;;;EAGT,YAAY;;;EAGZ;IACE;IACA,YAAY;IACZ,SAAS;IACT;;;EAGF,QAAQ;IAAQ;IAAe;;;;;EAI/B;IACE;IACA;IACA;;;;;EAKF;IACE;;;IAGA;;IAEA;;IAEA;;IAEA;;IAEA;;;;;IAKA,mBAAmB,OAAO,oBAAoB;;;;EAIhD;IACE;IACA;IACA;IACA;IACA;;;UAIa;;EAEf;;EAEA;;EAEA;;UAGe;EACf,OAAO;EACP,QAAQ;EACR,WAAW;EACX,QAAQ;EACR,YAAY;EACZ;IACE;IACA;IACA;IACA;IACA;;;;UAOa;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,WAAW;;;;;EAKX,WAAW;IAAQ;IAAe;;;UAGnB;;EAEf;;EAEA;;;EAGA,cAAc;;;EAGd;;;EAGA;;;EAGA;;UAGe;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,SAAS;;EAET,mBAAmB;IACjB;IACA,SAAS;MAAQ;MAAe;;IAChC;;;UAIa;EACf;EACA;;EAEA;;EAEA;;;EAGA;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;;;EAWA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;UAGe;;EAEf,UAAU;IACR;IACA;;IAEA;;IAEA;;IAEA;;EAEF;;;;;UAMe;;EAEf,cAAc;;EAEd;;EAEA;;UAGe;;EAEf;;;EAGA;EACA,UAAU;IAAQ;IAAe;IAAgB;;;UAGlC;;;EAGf;;EAEA;;EAEA;;EAEA;;EAEA;IACE;IACA;IACA;;;UAIa;;;EAGf;EACA,MAAM;IACJ;IACA;IACA;;;;EAIF;;UAGQ;;EAER;;EAEA;;;;EAIA;;EAEA;EACA;;KAGU,eACP;EACC;;EAEA;;EAEA;;EAEA;;EAEA;MAED;EACC;;EAEA;EACA;EACA;EACA;;UAGW;;EAEf;EACA;;EAEA;;;;EAIA,SAAS,eAAe;;;EAGxB;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;EACA;;EAEA;;UCzbe;EACf,MAAM;EACN;;UAGe;EACf,WAAW;EACX,gBAAgB;;;iBAIF,mBAAmB,MAAM,4BAA4B;UC4BpD;;EAEf,MAAM;;;EAGN;;;;;EAKA;EACA;;;EAGA,8BAA8B;;;EAG9B,kBAAkB;;;EAGlB,UAAU;;;;EAIV;IACE;IACA,cAAc;;;;;EAKhB,cAAc;IAAQ;IAAe;IAAe;;;EAEpD;;;;EAIA;;;;;;;;;EASA,eAAe;;;EAGf;;iBAGoB,YAAY,MAAM,qBAAqB,QAAQ"}

package/dist/{analyze-runs-Ck68mbgj.js → analyze-runs-Da7P_Mc5.js}

@@ -1,16 +1,15 @@
 import { s as ValidationError } from "./errors-DeEEV9Vx.js";
 import { r as observedSplitScore } from "./reward-P0VD_Aft.js";
 import { c as validateRunRecord } from "./run-record-DEjx43cI.js";
-import { c as pairedCohensDz, o as cohensD, r as paretoChart } from "./summary-report-BLnTyQKH.js";
+import { c as pairedCohensDz, o as cohensD, r as paretoChart } from "./summary-report-BstMpOfy.js";
 import { t as interpolatedQuantile } from "./quantile-CncXP-Ss.js";
-import { o as normalCdf, r as pairRunRecords } from "./paired-arms-D5H_58e8.js";
-import { c as lnGamma, l as regularizedIncompleteBeta, t as assertFiniteSample$1 } from "./internal-v6XfOeJW.js";
-import { a as spearmanR, r as pearsonR } from "./descriptive-cPwITVfy.js";
-import { r as continuousAgreement } from "./judge-calibration-DGqlfWW8.js";
-import { c as studentTQuantile, o as pairedTTest, r as pairedBootstrap } from "./paired-tests-C2zUqgUW.js";
-import { i as requiredPairedSampleSize, r as pairedMde } from "./power-and-mde-CHvyDO_H.js";
-import { r as meanOrZero, t as mean$1 } from "./mean-BPuUNznQ.js";
-import { n as aggregatePairedHoldout } from "./statistical-heldout-C9hz_AF9.js";
+import { o as normalCdf, r as pairRunRecords } from "./paired-arms-C0Vd9bGG.js";
+import { f as studentTQuantile, i as assertFiniteSample$1, m as regularizedIncompleteBeta, p as lnGamma, r as meanOrZero, t as mean$1 } from "./mean-DDFhslBt.js";
+import { i as pearsonR, o as spearmanR } from "./descriptive-BTzLL-aj.js";
+import { r as continuousAgreement } from "./judge-calibration-D93UCm1k.js";
+import { o as pairedTTest, r as pairedBootstrap } from "./paired-tests-kASEKpc-.js";
+import { i as requiredPairedSampleSize, r as pairedMde } from "./power-and-mde-CRxepy8y.js";
+import { n as aggregatePairedHoldout } from "./statistical-heldout-DYF5f10K.js";
 import "./query-CGXzO9d4.js";
 // src/statistics/inference.ts
 /** Two-sided normal tail, evaluated without subtracting a near-one CDF. */
@@ -1184,4 +1183,4 @@ async function analyzeRuns(opts) {
 }
 export { welchsTTest as a, fisherExactTwoSided as c, studentTTwoSidedPValue as d, twoByTwoPValue as f, iqr as i, linearRegressionSlope as l, checkCanaries as n, chiSquared as o, summarizeExecution as r, chiSquaredPValue as s, analyzeRuns as t, normalTwoSidedPValue as u };
 
-//# sourceMappingURL=analyze-runs-Ck68mbgj.js.map
+//# sourceMappingURL=analyze-runs-Da7P_Mc5.js.map