@tangle-network/agent-eval 0.217.1 → 0.218.1

package/dist/eval-campaign-BUMIxKTE.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"eval-campaign-BUMIxKTE.js","names":[],"sources":["../src/eval-campaign.ts"],"sourcesContent":["/**\n * EvalCampaign — opinionated matrix runner that wires the four\n * capture-integrity directives by construction.\n *\n * The canonical benchmark shape — matrix runner → for each\n * (variant, scenario, seed) → start a TraceEmitter → call LLMs → end the\n * run → analyze — has a bug class at the integration boundary: raw\n * events not captured, route silently wrong, integrity not asserted,\n * analyst never run. The directives in `SKILL.md § Capture integrity`\n * are the mitigations.\n *\n * `EvalCampaign` is the structural fix — consumers don't wire the\n * integrity surface themselves; the campaign owns it. Specifically:\n *\n *   - constructs a per-run `TraceStore` and `RawProviderSink` via factories\n *   - builds the run's `ChatClient` through `chatFactory`, handing it the\n *     run's raw sink and trace context — a transport built any other way has\n *     no raw HTTP envelope, and `assertRunCaptured` says so\n *   - calls `assertRunCaptured` after every `endRun` and routes failures\n *     through a configurable policy (`throw` / `mark_failed` / `log`)\n *   - assembles per-run `RunRecord`s and runs `researchReport` at the end\n *     so the campaign artifact is launch-decision-grade by default\n *   - embeds the campaign fingerprint (a SHA-256 over the canonicalised\n *     run set) and optional `preregistrationHash` in the report\n *\n * The runner contract is intentionally narrow: produce a `CampaignRunOutcome`\n * given a fully-wired `CampaignRunContext`. Everything orchestration-shaped\n * lives in the campaign. This is the inversion-of-control point — consumers\n * stop writing matrix runners and start writing scenario-runners.\n *\n * Out of scope for v1 (tracked in `docs/research-report-methodology.md`):\n *\n *   - Distributed/cluster execution (concurrency is local async)\n *   - Adaptive sampling / sequential interim looks\n *   - Resume from partial state across crashes\n *   - LLM-call retry beyond what the caller's transport already does\n */\n\nimport {\n  type AgentProfileCell,\n  type AgentProfileCellInput,\n  buildAgentProfileCell,\n  verifyAgentProfileCell,\n} from './agent-profile-cell'\nimport type { ChatClient } from './analyst/chat-client'\nimport { hashJson } from './pre-registration'\nimport type {\n  JudgeScoresRecord,\n  RunCostProvenance,\n  RunJudgeMetadata,\n  RunOutcome,\n  RunRecord,\n  RunSplitTag,\n  RunTaskFailure,\n  RunTokenUsage,\n} from './run-record'\nimport { validateRunRecord } from './run-record'\nimport { type ResearchReport, type ResearchReportOptions, researchReport } from './summary-report'\nimport type { RunCompleteHook } from './trace/emitter'\nimport { TraceEmitter } from './trace/emitter'\nimport {\n  assertRunCaptured,\n  RunIntegrityError,\n  type RunIntegrityExpectations,\n  type RunIntegrityReport,\n} from './trace/integrity'\nimport { FileSystemRawProviderSink, type RawProviderSink } from './trace/raw-provider-sink'\nimport type { TraceStore } from './trace/store'\n\n// ── Public types ─────────────────────────────────────────────────────────\n\nexport interface CampaignVariant<V> {\n  id: string\n  payload: V\n}\n\nexport interface CampaignScenario {\n  scenarioId: string\n  /** Free-form metadata propagated to runs and reports. */\n  tags?: Record<string, string>\n}\n\nexport interface CampaignRunContext<V> {\n  /** Stable run id. The campaign generates this; the runner does not. */\n  runId: string\n  /** Logical experiment id (campaignId by default; overridable per-run via opts). */\n  experimentId: string\n  variant: V\n  variantId: string\n  scenarioId: string\n  scenarioTags: Record<string, string>\n  seed: number\n  splitTag: RunSplitTag\n  /**\n   * The TraceEmitter for this run, with `onRunComplete` hooks pre-wired\n   * (analyst auto-execution if configured, plus integrity check). The\n   * runner MUST call `emitter.startRun` before doing any work and either\n   * `emitter.endRun` or `emitter.abortRun` before returning.\n   */\n  emitter: TraceEmitter\n  store: TraceStore\n  rawSink: RawProviderSink\n  /**\n   * The run's model transport, built by `chatFactory` with this run's\n   * `rawSink` and `runId` already bound.\n   */\n  chat: ChatClient\n}\n\n/** What the campaign binds into the run's transport. */\nexport interface CampaignChatWiring {\n  /**\n   * Raw provider sink for this run. Bind it into the transport: the campaign's\n   * integrity check requires every LLM span to carry a matching raw request\n   * event, so a transport built without it fails `assertRunCaptured`.\n   */\n  rawSink: RawProviderSink\n  runId: string\n}\n\ninterface CampaignRunOutcomeFields {\n  /** Did the run pass? Mirrors `RunOutcome.pass` semantics. */\n  pass: boolean\n  /** Score for the run on its split. Maps to `searchScore` or `holdoutScore`. */\n  score: number\n  /** Cost in USD, or null when the runner could not capture it. */\n  costUsd: number | null\n  /** Source of the cost amount. */\n  costProvenance: RunCostProvenance\n  tokenUsage: RunTokenUsage\n  /** Snapshot model id (e.g. `claude-sonnet-4-6@2025-04-15`). */\n  model: string\n  /** sha256 of the effective prompt sent to the model. */\n  promptHash: string\n  /** sha256 of the effective config (model, temperature, tools, judges, splits). */\n  configHash: string\n  /** Optional extra numeric metrics to land in `outcome.raw`. */\n  raw?: Record<string, number>\n  /** Optional judge metadata when a judge was used. */\n  judgeMetadata?: RunJudgeMetadata\n  /**\n   * Optional per-judge / per-dim breakdown for ensemble-judged runs.\n   * Propagated to `outcome.judgeScores` on the resulting `RunRecord`.\n   * Single-judge or scalar-only runs leave this unset.\n   */\n  judgeScores?: JudgeScoresRecord\n  /**\n   * Agent profile cell observed by the runner. When supplied, it overrides\n   * `EvalCampaignOptions.agentProfile` for this run and must match the\n   * outcome's `model` and `promptHash`.\n   */\n  agentProfile?: AgentProfileCell | AgentProfileCellInput\n}\n\n/** Campaign result with the same task-failure invariant as `RunRecord`. */\nexport type CampaignRunOutcome = CampaignRunOutcomeFields & RunTaskFailure\n\nexport type CampaignRunner<V> = (ctx: CampaignRunContext<V>) => Promise<CampaignRunOutcome>\n\nexport type CampaignIntegrityPolicy = 'throw' | 'mark_failed' | 'log'\n\nexport interface EvalCampaignOptions<V> {\n  /**\n   * Stable id for the campaign. Used as the default `experimentId` on\n   * every run, and folded into the campaign fingerprint.\n   */\n  campaignId: string\n  variants: CampaignVariant<V>[]\n  scenarios: CampaignScenario[]\n  /** Default `[0, 1, 2]`. */\n  seeds?: number[]\n  /** Default `'holdout'` — the split that anchors a launch decision. */\n  splitTag?: RunSplitTag\n  /** Git SHA the campaign is run against. Mandatory; `RunRecord` rejects unset. */\n  commitSha: string\n  /**\n   * Build the model transport for one run. agent-eval executes no paid model:\n   * the caller owns the transport and the credential never enters this\n   * package. The campaign calls this once per run and passes the run's raw\n   * provider sink and `runId`, so a transport that binds them captures the\n   * raw HTTP envelope `assertRunCaptured` checks for.\n   */\n  chatFactory: (wiring: CampaignChatWiring) => ChatClient\n  /**\n   * Caller-declared identity of the execution route, folded into the campaign\n   * fingerprint so two campaigns run against different endpoints do not share\n   * one identity. agent-eval no longer knows the endpoint; the owner of\n   * execution names it.\n   */\n  executionRef?: string\n  /**\n   * Per-run TraceStore factory. Common shape: a fresh store per run keyed\n   * on `runId`. Implementations that share a store across the campaign\n   * are valid — the campaign only writes through `emitter`.\n   */\n  storeFactory: (params: CampaignFactoryParams) => TraceStore\n  /**\n   * Per-run RawProviderSink factory. Defaults to `FileSystemRawProviderSink`\n   * rooted at `${workDir}/raw-events/${runId}` if `workDir` is supplied;\n   * otherwise required. Forensic capture is non-negotiable in a campaign\n   * run — pass `NoopRawProviderSink` explicitly if you want to opt out.\n   */\n  rawSinkFactory?: (params: CampaignFactoryParams) => RawProviderSink\n  /**\n   * Filesystem root for default `rawSinkFactory`. Ignored if\n   * `rawSinkFactory` is supplied.\n   */\n  workDir?: string\n  /**\n   * Extra `onRunComplete` hooks the campaign appends (after its own\n   * integrity-check hook). Pass `traceAnalystOnRunComplete(...)` here.\n   */\n  onRunComplete?: RunCompleteHook[]\n  /**\n   * Per-run integrity expectations. Defaults to:\n   *   `{ llmSpansMin: 1, requireRawCoverageOfLlmSpans: true, requireOutcome: true }`.\n   * Override (e.g. `{ llmSpansMin: 0 }`) for runs that don't call LLMs.\n   */\n  integrity?: RunIntegrityExpectations\n  /** Behaviour when integrity fails. Default `'mark_failed'`. */\n  onIntegrityFailure?: CampaignIntegrityPolicy\n  /**\n   * Per-run runner. Receives a fully-wired context; produces an outcome\n   * the campaign converts into a `RunRecord`.\n   */\n  runner: CampaignRunner<V>\n  /**\n   * If set, the campaign computes `researchReport` at the end. `comparator`\n   * is a `variantId`. Other fields are forwarded verbatim.\n   */\n  report?: { comparator?: string } & Omit<\n    ResearchReportOptions,\n    'comparator' | 'preregistrationHash' | 'generatedAt'\n  >\n  /**\n   * Hash of a signed `HypothesisManifest` (see `pre-registration.ts`).\n   * Embedded in the campaign fingerprint and the research report.\n   */\n  preregistrationHash?: string\n  /** Local concurrency. Default `1` (sequential). */\n  concurrency?: number\n  /**\n   * Override the time source. Tests pass a mock to make wallMs deterministic.\n   */\n  now?: () => number\n  /** Override the runId generator. Tests pin this. */\n  runId?: (params: CampaignFactoryParams) => string\n  /**\n   * Agent profile cell for campaign runs. Static profiles can pass an object;\n   * routers or variant-specific harnesses can pass a factory. The campaign\n   * stamps the built cell onto every `RunRecord` and rejects profile/model or\n   * profile/prompt contradictions.\n   */\n  agentProfile?:\n    | AgentProfileCell\n    | AgentProfileCellInput\n    | ((\n        params: CampaignFactoryParams & {\n          variant: V\n          scenarioTags: Record<string, string>\n        },\n      ) =>\n        | AgentProfileCell\n        | AgentProfileCellInput\n        | Promise<AgentProfileCell | AgentProfileCellInput>)\n}\n\nexport interface CampaignFactoryParams {\n  campaignId: string\n  runId: string\n  variantId: string\n  scenarioId: string\n  seed: number\n}\n\nexport interface FailedRun {\n  runId: string\n  variantId: string\n  scenarioId: string\n  seed: number\n  reason: string\n  error?: string\n}\n\nexport interface EvalCampaignResult {\n  campaignId: string\n  /** SHA-256 over canonicalised `(variantIds, scenarioIds, seeds, comparator, splitTag, executionRef, preregistrationHash)`. */\n  campaignFingerprint: string\n  preregistrationHash: string | null\n  /** Successful runs only. Failed runs land in `failedRuns`. */\n  runs: RunRecord[]\n  /** Integrity reports for every successful run. */\n  integrityReports: RunIntegrityReport[]\n  failedRuns: FailedRun[]\n  /** Computed when `report` is set on options. */\n  report?: ResearchReport\n  startedAt: string\n  endedAt: string\n}\n\n// ── Implementation ───────────────────────────────────────────────────────\n\nconst DEFAULT_INTEGRITY: RunIntegrityExpectations = {\n  llmSpansMin: 1,\n  requireRawCoverageOfLlmSpans: true,\n  requireOutcome: true,\n}\n\ntype CampaignCell<V> = { variant: CampaignVariant<V>; scenario: CampaignScenario; seed: number }\n\ntype CampaignRunResult = { record: RunRecord; integrity: RunIntegrityReport }\n\n/** Resolved options shared by every step of one campaign. */\ninterface CampaignSettings {\n  seeds: number[]\n  splitTag: RunSplitTag\n  concurrency: number\n  integrity: RunIntegrityExpectations\n  onIntegrityFailure: CampaignIntegrityPolicy\n  now: () => number\n  executionRef: string | null\n  preregistrationHash: string | null\n  rawSinkFactory: (params: CampaignFactoryParams) => RawProviderSink\n}\n\n/**\n * Mutable state of one campaign's worker pool.\n *\n * A genuine (non-CellExecutionError) error from any worker is a bug, not a\n * run-level failure. We must NOT let `Promise.all` reject mid-flight and\n * orphan the other workers' in-progress runs (emitters never finalized,\n * sinks/handles leak, partial work silently discarded). Instead: capture the\n * first genuine error, stop dispatching new cells so in-flight workers wind\n * down, finalize every still-open run, then re-throw the aggregated error.\n */\ninterface CampaignPool<V> {\n  opts: EvalCampaignOptions<V>\n  settings: CampaignSettings\n  cells: CampaignCell<V>[]\n  cursor: number\n  aborting: boolean\n  genuineErrors: unknown[]\n  /**\n   * Emitters for runs that are currently mid-flight, keyed by runId. A worker\n   * registers its emitter before invoking the runner and removes it once the\n   * run is finalized (via endRun/abortRun, success or failure). Anything left\n   * here after the pool settles is an orphan we must abort.\n   */\n  openRuns: Map<string, TraceEmitter>\n  runs: RunRecord[]\n  integrityReports: RunIntegrityReport[]\n  failedRuns: FailedRun[]\n}\n\nexport async function runEvalCampaign<V>(\n  opts: EvalCampaignOptions<V>,\n): Promise<EvalCampaignResult> {\n  assertCampaignPreflight(opts)\n  const settings = resolveCampaignSettings(opts)\n  const campaignFingerprint = await fingerprintCampaign(opts, settings)\n\n  const startedAt = new Date(settings.now()).toISOString()\n  const pool: CampaignPool<V> = {\n    opts,\n    settings,\n    cells: planCampaignCells(opts, settings.seeds),\n    cursor: 0,\n    aborting: false,\n    genuineErrors: [],\n    openRuns: new Map(),\n    runs: [],\n    integrityReports: [],\n    failedRuns: [],\n  }\n  await drainCampaignPool(pool)\n\n  const report = await buildCampaignReport(opts, settings, pool.runs)\n  const endedAt = new Date(settings.now()).toISOString()\n\n  return {\n    campaignId: opts.campaignId,\n    campaignFingerprint,\n    preregistrationHash: settings.preregistrationHash,\n    runs: pool.runs,\n    integrityReports: pool.integrityReports,\n    failedRuns: pool.failedRuns,\n    report,\n    startedAt,\n    endedAt,\n  }\n}\n\n/** Reject structurally invalid campaigns before any work starts. */\nfunction assertCampaignPreflight<V>(opts: EvalCampaignOptions<V>): void {\n  if (opts.variants.length === 0) {\n    throw new Error('runEvalCampaign: variants must be non-empty.')\n  }\n  if (opts.scenarios.length === 0) {\n    throw new Error('runEvalCampaign: scenarios must be non-empty.')\n  }\n  const variantIds = new Set<string>()\n  for (const v of opts.variants) {\n    if (variantIds.has(v.id)) {\n      throw new Error(`runEvalCampaign: duplicate variant id \"${v.id}\".`)\n    }\n    variantIds.add(v.id)\n  }\n  const scenarioIds = new Set<string>()\n  for (const s of opts.scenarios) {\n    if (scenarioIds.has(s.scenarioId)) {\n      throw new Error(`runEvalCampaign: duplicate scenarioId \"${s.scenarioId}\".`)\n    }\n    scenarioIds.add(s.scenarioId)\n  }\n  if (opts.report?.comparator && !variantIds.has(opts.report.comparator)) {\n    throw new Error(\n      `runEvalCampaign: report.comparator \"${opts.report.comparator}\" is not a configured variantId.`,\n    )\n  }\n  if (!opts.commitSha) {\n    throw new Error('runEvalCampaign: commitSha is required (every RunRecord needs it).')\n  }\n}\n\n/** Apply defaults to the optional campaign options. */\nfunction resolveCampaignSettings<V>(opts: EvalCampaignOptions<V>): CampaignSettings {\n  return {\n    seeds: opts.seeds ?? [0, 1, 2],\n    splitTag: opts.splitTag ?? 'holdout',\n    concurrency: Math.max(1, opts.concurrency ?? 1),\n    integrity: { ...DEFAULT_INTEGRITY, ...(opts.integrity ?? {}) },\n    onIntegrityFailure: opts.onIntegrityFailure ?? 'mark_failed',\n    now: opts.now ?? (() => Date.now()),\n    executionRef: opts.executionRef ?? null,\n    preregistrationHash: opts.preregistrationHash ?? null,\n    rawSinkFactory: opts.rawSinkFactory ?? defaultRawSinkFactory(opts.workDir),\n  }\n}\n\n/** SHA-256 over the canonicalised matrix definition. */\nfunction fingerprintCampaign<V>(\n  opts: EvalCampaignOptions<V>,\n  settings: CampaignSettings,\n): Promise<string> {\n  return hashJson({\n    campaignId: opts.campaignId,\n    variants: opts.variants.map((v) => v.id).sort(),\n    scenarios: opts.scenarios.map((s) => s.scenarioId).sort(),\n    seeds: [...settings.seeds].sort((a, b) => a - b),\n    splitTag: settings.splitTag,\n    comparator: opts.report?.comparator ?? null,\n    executionRef: settings.executionRef,\n    preregistrationHash: settings.preregistrationHash,\n  })\n}\n\n/** Expand variants x scenarios x seeds, variant-major. */\nfunction planCampaignCells<V>(opts: EvalCampaignOptions<V>, seeds: number[]): CampaignCell<V>[] {\n  const cells: CampaignCell<V>[] = []\n  for (const variant of opts.variants) {\n    for (const scenario of opts.scenarios) {\n      for (const seed of seeds) {\n        cells.push({ variant, scenario, seed })\n      }\n    }\n  }\n  return cells\n}\n\n/**\n * Run the bounded-concurrency worker pool to completion, finalize any run left\n * open, and re-throw the genuine errors the workers captured.\n */\nasync function drainCampaignPool<V>(pool: CampaignPool<V>): Promise<void> {\n  const workers = Array.from(\n    { length: Math.min(pool.settings.concurrency, pool.cells.length) },\n    () => campaignWorker(pool),\n  )\n  // allSettled (not all): a genuine error in one worker must not reject the\n  // pool mid-flight and orphan the others. Each worker captures its own\n  // genuine error into `genuineErrors` and returns; we re-throw below.\n  await Promise.allSettled(workers)\n\n  // Finalize any run still open after the pool wound down. With the\n  // stop-dispatch flag these are the runs that were mid-flight in peer workers\n  // when the first genuine error fired — abort them so their emitters finalize\n  // (hooks fire, store records a terminal status) instead of leaking.\n  for (const [runId, emitter] of pool.openRuns) {\n    await finalizeAbort(emitter, runId, 'campaign aborted: genuine error in a sibling run')\n  }\n  pool.openRuns.clear()\n\n  if (pool.genuineErrors.length > 0) {\n    throw pool.genuineErrors.length === 1\n      ? pool.genuineErrors[0]\n      : new AggregateError(\n          pool.genuineErrors,\n          `runEvalCampaign: ${pool.genuineErrors.length} runs failed with genuine (non-run-level) errors`,\n        )\n  }\n}\n\n/** Pull cells off the shared cursor until none remain or the pool is aborting. */\nasync function campaignWorker<V>(pool: CampaignPool<V>): Promise<void> {\n  while (!pool.aborting) {\n    const i = pool.cursor++\n    if (i >= pool.cells.length) return\n    const cell = pool.cells[i]!\n    try {\n      const result = await runCampaignCell(pool, cell)\n      pool.runs.push(result.record)\n      pool.integrityReports.push(result.integrity)\n    } catch (err) {\n      if (err instanceof CellExecutionError) {\n        pool.failedRuns.push(err.failed)\n        if (err.integrity) pool.integrityReports.push(err.integrity)\n      } else {\n        // Genuine bug — not a runner failure, not an integrity failure.\n        // Capture it and stop dispatching so peers wind down gracefully;\n        // re-thrown after the pool settles. Do not surface here — that would\n        // reject Promise.all and orphan the other workers' runs.\n        pool.genuineErrors.push(err)\n        pool.aborting = true\n        return\n      }\n    }\n  }\n}\n\n/** Run one matrix cell: runner, integrity check, then the validated RunRecord. */\nasync function runCampaignCell<V>(\n  pool: CampaignPool<V>,\n  cell: CampaignCell<V>,\n): Promise<CampaignRunResult> {\n  const { opts, settings } = pool\n  const runId = (opts.runId ?? defaultRunId)({\n    campaignId: opts.campaignId,\n    runId: '', // unused by default generator\n    variantId: cell.variant.id,\n    scenarioId: cell.scenario.scenarioId,\n    seed: cell.seed,\n  })\n  const factoryParams: CampaignFactoryParams = {\n    campaignId: opts.campaignId,\n    runId,\n    variantId: cell.variant.id,\n    scenarioId: cell.scenario.scenarioId,\n    seed: cell.seed,\n  }\n  const store = opts.storeFactory(factoryParams)\n  const rawSink = settings.rawSinkFactory(factoryParams)\n\n  const emitter = new TraceEmitter(store, {\n    runId,\n    now: opts.now,\n    onRunComplete: opts.onRunComplete,\n  })\n  // Track this run as open so a genuine error elsewhere in the pool can\n  // finalize it instead of orphaning it. Removed in the finally below.\n  pool.openRuns.set(runId, emitter)\n\n  const chat = opts.chatFactory({ rawSink, runId })\n\n  const ctx: CampaignRunContext<V> = {\n    runId,\n    experimentId: opts.campaignId,\n    variant: cell.variant.payload,\n    variantId: cell.variant.id,\n    scenarioId: cell.scenario.scenarioId,\n    scenarioTags: cell.scenario.tags ?? {},\n    seed: cell.seed,\n    splitTag: settings.splitTag,\n    emitter,\n    store,\n    rawSink,\n    chat,\n  }\n\n  try {\n    const wallStart = settings.now()\n    const outcome = await invokeCampaignRunner(opts, cell, ctx)\n    const wallMs = settings.now() - wallStart\n\n    const integrityReport = await assertRunCaptured(store, runId, {\n      ...settings.integrity,\n      rawSink,\n    })\n    if (!integrityReport.ok) admitOrRejectIntegrity(settings, cell, runId, integrityReport)\n\n    const record = buildCampaignRunRecord(opts, settings, cell, runId, outcome, wallMs)\n    await stampAgentProfile(opts, cell, record, runId, outcome)\n    return { record: validateRunRecord(record), integrity: integrityReport }\n  } finally {\n    // This run's worker has finished with it (success, run-level failure, or\n    // genuine error). It is no longer the pool's job to finalize — drop it so\n    // the post-settle sweep doesn't double-abort a finalized run.\n    pool.openRuns.delete(runId)\n  }\n}\n\n/** Invoke the runner; a throw aborts the run and becomes a run-level failure. */\nasync function invokeCampaignRunner<V>(\n  opts: EvalCampaignOptions<V>,\n  cell: CampaignCell<V>,\n  ctx: CampaignRunContext<V>,\n): Promise<CampaignRunOutcome> {\n  try {\n    return await opts.runner(ctx)\n  } catch (err) {\n    const message = err instanceof Error ? err.message : String(err)\n    // The runner threw mid-execution. Abort the run so the emitter\n    // finalizes. The only benign abortRun failure is \"the runner never\n    // started the run\" (nothing to finalize). A store-write failure (disk\n    // full, FS error) is a genuine diagnostic — surface it rather than\n    // masking it as a plain runner failure.\n    await finalizeAbort(ctx.emitter, ctx.runId, message)\n    throw new CellExecutionError({\n      runId: ctx.runId,\n      variantId: cell.variant.id,\n      scenarioId: cell.scenario.scenarioId,\n      seed: cell.seed,\n      reason: 'runner_threw',\n      error: message,\n    })\n  }\n}\n\n/** Apply the integrity-failure policy to a report that is not ok. */\nfunction admitOrRejectIntegrity<V>(\n  settings: CampaignSettings,\n  cell: CampaignCell<V>,\n  runId: string,\n  integrityReport: RunIntegrityReport,\n): void {\n  switch (settings.onIntegrityFailure) {\n    case 'throw':\n      throw new RunIntegrityError(integrityReport)\n    case 'mark_failed':\n      throw new CellExecutionError(\n        {\n          runId,\n          variantId: cell.variant.id,\n          scenarioId: cell.scenario.scenarioId,\n          seed: cell.seed,\n          reason: 'integrity_failed',\n          error: integrityReport.issues.map((i) => i.code).join(', '),\n        },\n        integrityReport,\n      )\n    case 'log':\n      // Caller wants the run admitted with a flagged report; fall through.\n      break\n  }\n}\n\n/** Assemble the RunRecord for a successful run, before profile stamping. */\nfunction buildCampaignRunRecord<V>(\n  opts: EvalCampaignOptions<V>,\n  settings: CampaignSettings,\n  cell: CampaignCell<V>,\n  runId: string,\n  outcome: CampaignRunOutcome,\n  wallMs: number,\n): RunRecord {\n  const recordOutcome: RunOutcome = {\n    raw: outcome.raw ?? {},\n  }\n  if (settings.splitTag === 'holdout') recordOutcome.holdoutScore = outcome.score\n  else recordOutcome.searchScore = outcome.score\n  if (outcome.judgeScores !== undefined) recordOutcome.judgeScores = outcome.judgeScores\n\n  return {\n    runId,\n    experimentId: opts.campaignId,\n    candidateId: cell.variant.id,\n    seed: cell.seed,\n    model: outcome.model,\n    promptHash: outcome.promptHash,\n    configHash: outcome.configHash,\n    commitSha: opts.commitSha,\n    wallMs,\n    costUsd: outcome.costUsd,\n    costProvenance: outcome.costProvenance,\n    tokenUsage: outcome.tokenUsage,\n    terminalOutcome: 'succeeded',\n    judgeMetadata: outcome.judgeMetadata,\n    outcome: recordOutcome,\n    ...(outcome.failureClass ? { failureClass: outcome.failureClass } : {}),\n    failureMode: outcome.failureMode,\n    splitTag: settings.splitTag,\n    scenarioId: cell.scenario.scenarioId,\n  }\n}\n\n/** Resolve the run's agent profile (outcome first, then options) and stamp it on the record. */\nasync function stampAgentProfile<V>(\n  opts: EvalCampaignOptions<V>,\n  cell: CampaignCell<V>,\n  record: RunRecord,\n  runId: string,\n  outcome: CampaignRunOutcome,\n): Promise<void> {\n  const profileSource =\n    outcome.agentProfile ??\n    (typeof opts.agentProfile === 'function'\n      ? await opts.agentProfile({\n          campaignId: opts.campaignId,\n          runId,\n          variantId: cell.variant.id,\n          scenarioId: cell.scenario.scenarioId,\n          seed: cell.seed,\n          variant: cell.variant.payload,\n          scenarioTags: cell.scenario.tags ?? {},\n        })\n      : opts.agentProfile)\n  if (profileSource !== undefined) {\n    const agentProfile = await resolveAgentProfileCell(profileSource)\n    assertAgentProfileMatchesRun(agentProfile, outcome.model, outcome.promptHash)\n    record.agentProfile = agentProfile\n  }\n}\n\n/** Build the optional research report over the successful runs. */\nasync function buildCampaignReport<V>(\n  opts: EvalCampaignOptions<V>,\n  settings: CampaignSettings,\n  runs: RunRecord[],\n): Promise<ResearchReport | undefined> {\n  if (!opts.report) return undefined\n  const reportOpts: ResearchReportOptions = {\n    ...opts.report,\n    comparator: opts.report.comparator,\n    split: settings.splitTag === 'dev' ? 'search' : settings.splitTag,\n    generatedAt: new Date(settings.now()).toISOString(),\n    preregistrationHash: settings.preregistrationHash ?? undefined,\n  }\n  return researchReport(runs, reportOpts)\n}\n\n// ── Internal ─────────────────────────────────────────────────────────────\n\nclass CellExecutionError extends Error {\n  readonly failed: FailedRun\n  readonly integrity?: RunIntegrityReport\n  constructor(failed: FailedRun, integrity?: RunIntegrityReport) {\n    super(`cell ${failed.variantId}/${failed.scenarioId}@${failed.seed} failed: ${failed.reason}`)\n    this.failed = failed\n    this.integrity = integrity\n  }\n}\n\n/**\n * Abort a run whose owning work threw or was orphaned by a sibling's genuine\n * error, finalizing the emitter so hooks fire and the store records a terminal\n * status. Safe to call unconditionally: it only aborts runs that are still\n * `running`. Two no-op cases are intentional and benign:\n *\n *   - the run was never started (absent from the store) — nothing to finalize\n *   - the run is already terminal (`completed` / `failed` / `aborted`) —\n *     finalizing again would overwrite the real outcome (e.g. flip a passed run\n *     to `aborted` with `{ pass: false, notes: reason }`), so we leave it alone\n *\n * Any OTHER failure of the store read or the abort write (disk full, FS fault,\n * backend down) is a genuine diagnostic and propagates rather than being\n * swallowed.\n */\nasync function finalizeAbort(emitter: TraceEmitter, runId: string, reason: string): Promise<void> {\n  const existing = await emitter.traceStore.getRun(runId)\n  if (existing === undefined) return // run never started; nothing to abort\n  if (existing.status !== 'running') return // already finalized; never overwrite a real outcome\n  // The emitter never throws on a failed store write; it counts it. A failed\n  // abort write is a genuine diagnostic here, so read the count and surface it.\n  const droppedBefore = emitter.captureStats().dropped\n  await emitter.abortRun(reason)\n  const after = emitter.captureStats()\n  if (after.dropped > droppedBefore) {\n    throw new Error(`finalizeAbort: run ${runId} could not be finalized: ${after.lastError}`)\n  }\n}\n\nfunction defaultRawSinkFactory(workDir: string | undefined) {\n  return (params: CampaignFactoryParams): RawProviderSink => {\n    if (!workDir) {\n      throw new Error(\n        'runEvalCampaign: rawSinkFactory not supplied and workDir not set. Pass either to enable raw provider capture, or pass `new NoopRawProviderSink()` via rawSinkFactory to opt out explicitly.',\n      )\n    }\n    return new FileSystemRawProviderSink({\n      dir: `${workDir}/raw-events/${params.runId}`,\n    })\n  }\n}\n\nasync function resolveAgentProfileCell(\n  input: AgentProfileCell | AgentProfileCellInput,\n): Promise<AgentProfileCell> {\n  if (isAgentProfileCell(input)) {\n    if (!(await verifyAgentProfileCell(input))) {\n      throw new Error(`runEvalCampaign: agentProfile.cellId does not match its content`)\n    }\n    return input\n  }\n  return buildAgentProfileCell(input)\n}\n\nfunction isAgentProfileCell(\n  input: AgentProfileCell | AgentProfileCellInput,\n): input is AgentProfileCell {\n  return 'schemaVersion' in input && 'cellId' in input\n}\n\nfunction assertAgentProfileMatchesRun(\n  profile: AgentProfileCell,\n  model: string,\n  promptHash: string,\n): void {\n  if (profile.model !== undefined && profile.model !== model) {\n    throw new Error(\n      `runEvalCampaign: agentProfile.model \"${profile.model}\" does not match outcome.model \"${model}\"`,\n    )\n  }\n  if (profile.promptHash !== undefined && profile.promptHash !== promptHash) {\n    throw new Error(\n      `runEvalCampaign: agentProfile.promptHash \"${profile.promptHash}\" does not match outcome.promptHash \"${promptHash}\"`,\n    )\n  }\n}\n\nfunction defaultRunId(params: CampaignFactoryParams): string {\n  // Stable across re-runs: fingerprint of (campaignId, variantId, scenarioId, seed).\n  // Caller can override via opts.runId for non-deterministic IDs.\n  const base = `${params.campaignId}::${params.variantId}::${params.scenarioId}::${params.seed}`\n  // TWO 32-bit FNV-1a accumulators with different offsets, concatenated for a\n  // 64-bit-wide id. Not crypto-grade; stability and uniqueness are what this\n  // needs. Frozen: the value names a persisted cell, and the second\n  // accumulator's offset is part of the identity, so no single-accumulator\n  // helper can replace it.\n  let h1 = 0x811c9dc5\n  let h2 = 0x12345678\n  for (let i = 0; i < base.length; i++) {\n    const c = base.charCodeAt(i)\n    h1 = Math.imul(h1 ^ c, 0x01000193) >>> 0\n    h2 = Math.imul(h2 ^ c, 0x9e3779b1) >>> 0\n  }\n  return `run-${h1.toString(16).padStart(8, '0')}${h2.toString(16).padStart(8, '0')}`\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AA8SA,MAAM,oBAA8C;CAClD,aAAa;CACb,8BAA8B;CAC9B,gBAAgB;AAClB;AAgDA,eAAsB,gBACpB,MAC6B;CAC7B,wBAAwB,IAAI;CAC5B,MAAM,WAAW,wBAAwB,IAAI;CAC7C,MAAM,sBAAsB,MAAM,oBAAoB,MAAM,QAAQ;CAEpE,MAAM,YAAY,IAAI,KAAK,SAAS,IAAI,CAAC,CAAC,CAAC,YAAY;CACvD,MAAM,OAAwB;EAC5B;EACA;EACA,OAAO,kBAAkB,MAAM,SAAS,KAAK;EAC7C,QAAQ;EACR,UAAU;EACV,eAAe,CAAC;EAChB,0BAAU,IAAI,IAAI;EAClB,MAAM,CAAC;EACP,kBAAkB,CAAC;EACnB,YAAY,CAAC;CACf;CACA,MAAM,kBAAkB,IAAI;CAE5B,MAAM,SAAS,MAAM,oBAAoB,MAAM,UAAU,KAAK,IAAI;CAClE,MAAM,UAAU,IAAI,KAAK,SAAS,IAAI,CAAC,CAAC,CAAC,YAAY;CAErD,OAAO;EACL,YAAY,KAAK;EACjB;EACA,qBAAqB,SAAS;EAC9B,MAAM,KAAK;EACX,kBAAkB,KAAK;EACvB,YAAY,KAAK;EACjB;EACA;EACA;CACF;AACF;;AAGA,SAAS,wBAA2B,MAAoC;CACtE,IAAI,KAAK,SAAS,WAAW,GAC3B,MAAM,IAAI,MAAM,8CAA8C;CAEhE,IAAI,KAAK,UAAU,WAAW,GAC5B,MAAM,IAAI,MAAM,+CAA+C;CAEjE,MAAM,6BAAa,IAAI,IAAY;CACnC,KAAK,MAAM,KAAK,KAAK,UAAU;EAC7B,IAAI,WAAW,IAAI,EAAE,EAAE,GACrB,MAAM,IAAI,MAAM,0CAA0C,EAAE,GAAG,GAAG;EAEpE,WAAW,IAAI,EAAE,EAAE;CACrB;CACA,MAAM,8BAAc,IAAI,IAAY;CACpC,KAAK,MAAM,KAAK,KAAK,WAAW;EAC9B,IAAI,YAAY,IAAI,EAAE,UAAU,GAC9B,MAAM,IAAI,MAAM,0CAA0C,EAAE,WAAW,GAAG;EAE5E,YAAY,IAAI,EAAE,UAAU;CAC9B;CACA,IAAI,KAAK,QAAQ,cAAc,CAAC,WAAW,IAAI,KAAK,OAAO,UAAU,GACnE,MAAM,IAAI,MACR,uCAAuC,KAAK,OAAO,WAAW,iCAChE;CAEF,IAAI,CAAC,KAAK,WACR,MAAM,IAAI,MAAM,oEAAoE;AAExF;;AAGA,SAAS,wBAA2B,MAAgD;CAClF,OAAO;EACL,OAAO,KAAK,SAAS;GAAC;GAAG;GAAG;EAAC;EAC7B,UAAU,KAAK,YAAY;EAC3B,aAAa,KAAK,IAAI,GAAG,KAAK,eAAe,CAAC;EAC9C,WAAW;GAAE,GAAG;GAAmB,GAAI,KAAK,aAAa,CAAC;EAAG;EAC7D,oBAAoB,KAAK,sBAAsB;EAC/C,KAAK,KAAK,cAAc,KAAK,IAAI;EACjC,cAAc,KAAK,gBAAgB;EACnC,qBAAqB,KAAK,uBAAuB;EACjD,gBAAgB,KAAK,kBAAkB,sBAAsB,KAAK,OAAO;CAC3E;AACF;;AAGA,SAAS,oBACP,MACA,UACiB;CACjB,OAAO,SAAS;EACd,YAAY,KAAK;EACjB,UAAU,KAAK,SAAS,KAAK,MAAM,EAAE,EAAE,CAAC,CAAC,KAAK;EAC9C,WAAW,KAAK,UAAU,KAAK,MAAM,EAAE,UAAU,CAAC,CAAC,KAAK;EACxD,OAAO,CAAC,GAAG,SAAS,KAAK,CAAC,CAAC,MAAM,GAAG,MAAM,IAAI,CAAC;EAC/C,UAAU,SAAS;EACnB,YAAY,KAAK,QAAQ,cAAc;EACvC,cAAc,SAAS;EACvB,qBAAqB,SAAS;CAChC,CAAC;AACH;;AAGA,SAAS,kBAAqB,MAA8B,OAAoC;CAC9F,MAAM,QAA2B,CAAC;CAClC,KAAK,MAAM,WAAW,KAAK,UACzB,KAAK,MAAM,YAAY,KAAK,WAC1B,KAAK,MAAM,QAAQ,OACjB,MAAM,KAAK;EAAE;EAAS;EAAU;CAAK,CAAC;CAI5C,OAAO;AACT;;;;;AAMA,eAAe,kBAAqB,MAAsC;CACxE,MAAM,UAAU,MAAM,KACpB,EAAE,QAAQ,KAAK,IAAI,KAAK,SAAS,aAAa,KAAK,MAAM,MAAM,EAAE,SAC3D,eAAe,IAAI,CAC3B;CAIA,MAAM,QAAQ,WAAW,OAAO;CAMhC,KAAK,MAAM,CAAC,OAAO,YAAY,KAAK,UAClC,MAAM,cAAc,SAAS,OAAO,kDAAkD;CAExF,KAAK,SAAS,MAAM;CAEpB,IAAI,KAAK,cAAc,SAAS,GAC9B,MAAM,KAAK,cAAc,WAAW,IAChC,KAAK,cAAc,KACnB,IAAI,eACF,KAAK,eACL,oBAAoB,KAAK,cAAc,OAAO,iDAChD;AAER;;AAGA,eAAe,eAAkB,MAAsC;CACrE,OAAO,CAAC,KAAK,UAAU;EACrB,MAAM,IAAI,KAAK;EACf,IAAI,KAAK,KAAK,MAAM,QAAQ;EAC5B,MAAM,OAAO,KAAK,MAAM;EACxB,IAAI;GACF,MAAM,SAAS,MAAM,gBAAgB,MAAM,IAAI;GAC/C,KAAK,KAAK,KAAK,OAAO,MAAM;GAC5B,KAAK,iBAAiB,KAAK,OAAO,SAAS;EAC7C,SAAS,KAAK;GACZ,IAAI,eAAe,oBAAoB;IACrC,KAAK,WAAW,KAAK,IAAI,MAAM;IAC/B,IAAI,IAAI,WAAW,KAAK,iBAAiB,KAAK,IAAI,SAAS;GAC7D,OAAO;IAKL,KAAK,cAAc,KAAK,GAAG;IAC3B,KAAK,WAAW;IAChB;GACF;EACF;CACF;AACF;;AAGA,eAAe,gBACb,MACA,MAC4B;CAC5B,MAAM,EAAE,MAAM,aAAa;CAC3B,MAAM,SAAS,KAAK,SAAS,aAAA,CAAc;EACzC,YAAY,KAAK;EACjB,OAAO;EACP,WAAW,KAAK,QAAQ;EACxB,YAAY,KAAK,SAAS;EAC1B,MAAM,KAAK;CACb,CAAC;CACD,MAAM,gBAAuC;EAC3C,YAAY,KAAK;EACjB;EACA,WAAW,KAAK,QAAQ;EACxB,YAAY,KAAK,SAAS;EAC1B,MAAM,KAAK;CACb;CACA,MAAM,QAAQ,KAAK,aAAa,aAAa;CAC7C,MAAM,UAAU,SAAS,eAAe,aAAa;CAErD,MAAM,UAAU,IAAI,aAAa,OAAO;EACtC;EACA,KAAK,KAAK;EACV,eAAe,KAAK;CACtB,CAAC;CAGD,KAAK,SAAS,IAAI,OAAO,OAAO;CAEhC,MAAM,OAAO,KAAK,YAAY;EAAE;EAAS;CAAM,CAAC;CAEhD,MAAM,MAA6B;EACjC;EACA,cAAc,KAAK;EACnB,SAAS,KAAK,QAAQ;EACtB,WAAW,KAAK,QAAQ;EACxB,YAAY,KAAK,SAAS;EAC1B,cAAc,KAAK,SAAS,QAAQ,CAAC;EACrC,MAAM,KAAK;EACX,UAAU,SAAS;EACnB;EACA;EACA;EACA;CACF;CAEA,IAAI;EACF,MAAM,YAAY,SAAS,IAAI;EAC/B,MAAM,UAAU,MAAM,qBAAqB,MAAM,MAAM,GAAG;EAC1D,MAAM,SAAS,SAAS,IAAI,IAAI;EAEhC,MAAM,kBAAkB,MAAM,kBAAkB,OAAO,OAAO;GAC5D,GAAG,SAAS;GACZ;EACF,CAAC;EACD,IAAI,CAAC,gBAAgB,IAAI,uBAAuB,UAAU,MAAM,OAAO,eAAe;EAEtF,MAAM,SAAS,uBAAuB,MAAM,UAAU,MAAM,OAAO,SAAS,MAAM;EAClF,MAAM,kBAAkB,MAAM,MAAM,QAAQ,OAAO,OAAO;EAC1D,OAAO;GAAE,QAAQ,kBAAkB,MAAM;GAAG,WAAW;EAAgB;CACzE,UAAU;EAIR,KAAK,SAAS,OAAO,KAAK;CAC5B;AACF;;AAGA,eAAe,qBACb,MACA,MACA,KAC6B;CAC7B,IAAI;EACF,OAAO,MAAM,KAAK,OAAO,GAAG;CAC9B,SAAS,KAAK;EACZ,MAAM,UAAU,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;EAM/D,MAAM,cAAc,IAAI,SAAS,IAAI,OAAO,OAAO;EACnD,MAAM,IAAI,mBAAmB;GAC3B,OAAO,IAAI;GACX,WAAW,KAAK,QAAQ;GACxB,YAAY,KAAK,SAAS;GAC1B,MAAM,KAAK;GACX,QAAQ;GACR,OAAO;EACT,CAAC;CACH;AACF;;AAGA,SAAS,uBACP,UACA,MACA,OACA,iBACM;CACN,QAAQ,SAAS,oBAAjB;EACE,KAAK,SACH,MAAM,IAAI,kBAAkB,eAAe;EAC7C,KAAK,eACH,MAAM,IAAI,mBACR;GACE;GACA,WAAW,KAAK,QAAQ;GACxB,YAAY,KAAK,SAAS;GAC1B,MAAM,KAAK;GACX,QAAQ;GACR,OAAO,gBAAgB,OAAO,KAAK,MAAM,EAAE,IAAI,CAAC,CAAC,KAAK,IAAI;EAC5D,GACA,eACF;CAIJ;AACF;;AAGA,SAAS,uBACP,MACA,UACA,MACA,OACA,SACA,QACW;CACX,MAAM,gBAA4B,EAChC,KAAK,QAAQ,OAAO,CAAC,EACvB;CACA,IAAI,SAAS,aAAa,WAAW,cAAc,eAAe,QAAQ;MACrE,cAAc,cAAc,QAAQ;CACzC,IAAI,QAAQ,gBAAgB,KAAA,GAAW,cAAc,cAAc,QAAQ;CAE3E,OAAO;EACL;EACA,cAAc,KAAK;EACnB,aAAa,KAAK,QAAQ;EAC1B,MAAM,KAAK;EACX,OAAO,QAAQ;EACf,YAAY,QAAQ;EACpB,YAAY,QAAQ;EACpB,WAAW,KAAK;EAChB;EACA,SAAS,QAAQ;EACjB,gBAAgB,QAAQ;EACxB,YAAY,QAAQ;EACpB,iBAAiB;EACjB,eAAe,QAAQ;EACvB,SAAS;EACT,GAAI,QAAQ,eAAe,EAAE,cAAc,QAAQ,aAAa,IAAI,CAAC;EACrE,aAAa,QAAQ;EACrB,UAAU,SAAS;EACnB,YAAY,KAAK,SAAS;CAC5B;AACF;;AAGA,eAAe,kBACb,MACA,MACA,QACA,OACA,SACe;CACf,MAAM,gBACJ,QAAQ,iBACP,OAAO,KAAK,iBAAiB,aAC1B,MAAM,KAAK,aAAa;EACtB,YAAY,KAAK;EACjB;EACA,WAAW,KAAK,QAAQ;EACxB,YAAY,KAAK,SAAS;EAC1B,MAAM,KAAK;EACX,SAAS,KAAK,QAAQ;EACtB,cAAc,KAAK,SAAS,QAAQ,CAAC;CACvC,CAAC,IACD,KAAK;CACX,IAAI,kBAAkB,KAAA,GAAW;EAC/B,MAAM,eAAe,MAAM,wBAAwB,aAAa;EAChE,6BAA6B,cAAc,QAAQ,OAAO,QAAQ,UAAU;EAC5E,OAAO,eAAe;CACxB;AACF;;AAGA,eAAe,oBACb,MACA,UACA,MACqC;CACrC,IAAI,CAAC,KAAK,QAAQ,OAAO,KAAA;CACzB,MAAM,aAAoC;EACxC,GAAG,KAAK;EACR,YAAY,KAAK,OAAO;EACxB,OAAO,SAAS,aAAa,QAAQ,WAAW,SAAS;EACzD,aAAa,IAAI,KAAK,SAAS,IAAI,CAAC,CAAC,CAAC,YAAY;EAClD,qBAAqB,SAAS,uBAAuB,KAAA;CACvD;CACA,OAAO,eAAe,MAAM,UAAU;AACxC;AAIA,IAAM,qBAAN,cAAiC,MAAM;CACrC;CACA;CACA,YAAY,QAAmB,WAAgC;EAC7D,MAAM,QAAQ,OAAO,UAAU,GAAG,OAAO,WAAW,GAAG,OAAO,KAAK,WAAW,OAAO,QAAQ;EAC7F,KAAK,SAAS;EACd,KAAK,YAAY;CACnB;AACF;;;;;;;;;;;;;;;;AAiBA,eAAe,cAAc,SAAuB,OAAe,QAA+B;CAChG,MAAM,WAAW,MAAM,QAAQ,WAAW,OAAO,KAAK;CACtD,IAAI,aAAa,KAAA,GAAW;CAC5B,IAAI,SAAS,WAAW,WAAW;CAGnC,MAAM,gBAAgB,QAAQ,aAAa,CAAC,CAAC;CAC7C,MAAM,QAAQ,SAAS,MAAM;CAC7B,MAAM,QAAQ,QAAQ,aAAa;CACnC,IAAI,MAAM,UAAU,eAClB,MAAM,IAAI,MAAM,sBAAsB,MAAM,2BAA2B,MAAM,WAAW;AAE5F;AAEA,SAAS,sBAAsB,SAA6B;CAC1D,QAAQ,WAAmD;EACzD,IAAI,CAAC,SACH,MAAM,IAAI,MACR,6LACF;EAEF,OAAO,IAAI,0BAA0B,EACnC,KAAK,GAAG,QAAQ,cAAc,OAAO,QACvC,CAAC;CACH;AACF;AAEA,eAAe,wBACb,OAC2B;CAC3B,IAAI,mBAAmB,KAAK,GAAG;EAC7B,IAAI,CAAE,MAAM,uBAAuB,KAAK,GACtC,MAAM,IAAI,MAAM,iEAAiE;EAEnF,OAAO;CACT;CACA,OAAO,sBAAsB,KAAK;AACpC;AAEA,SAAS,mBACP,OAC2B;CAC3B,OAAO,mBAAmB,SAAS,YAAY;AACjD;AAEA,SAAS,6BACP,SACA,OACA,YACM;CACN,IAAI,QAAQ,UAAU,KAAA,KAAa,QAAQ,UAAU,OACnD,MAAM,IAAI,MACR,wCAAwC,QAAQ,MAAM,kCAAkC,MAAM,EAChG;CAEF,IAAI,QAAQ,eAAe,KAAA,KAAa,QAAQ,eAAe,YAC7D,MAAM,IAAI,MACR,6CAA6C,QAAQ,WAAW,uCAAuC,WAAW,EACpH;AAEJ;AAEA,SAAS,aAAa,QAAuC;CAG3D,MAAM,OAAO,GAAG,OAAO,WAAW,IAAI,OAAO,UAAU,IAAI,OAAO,WAAW,IAAI,OAAO;CAMxF,IAAI,KAAK;CACT,IAAI,KAAK;CACT,KAAK,IAAI,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK;EACpC,MAAM,IAAI,KAAK,WAAW,CAAC;EAC3B,KAAK,KAAK,KAAK,KAAK,GAAG,QAAU,MAAM;EACvC,KAAK,KAAK,KAAK,KAAK,GAAG,UAAU,MAAM;CACzC;CACA,OAAO,OAAO,GAAG,SAAS,EAAE,CAAC,CAAC,SAAS,GAAG,GAAG,IAAI,GAAG,SAAS,EAAE,CAAC,CAAC,SAAS,GAAG,GAAG;AAClF"}

package/dist/{evaluation-jTjC_QQD.d.ts → evaluation-BOZ38vOH.d.ts}

@@ -1,6 +1,6 @@
 import { a as CostLedgerHandle, d as CostReceiptInput, m as MaximumCharge, r as CostChannel, u as CostReceipt } from "./cost-ledger-Jk62OBfF.js";
 import { i as AnalystFinding, n as AnalystContext, t as Analyst } from "./types-CW8ewATZ.js";
-import { R as Scenario, S as JudgeConfig, w as JudgeScore } from "./types-D8GhJoHk.js";
+import { R as Scenario, S as JudgeConfig, w as JudgeScore } from "./types-DtkO1G9L.js";
 //#region src/evaluation.d.ts
 interface EvaluationResult<T> {
   value: T;
@@ -72,4 +72,4 @@ interface EvaluationAnalystOptions<I, O> {
 declare function asAnalyst<I, O>(options: EvaluationAnalystOptions<I, O>): Analyst<I>;
 //#endregion
 export { EvaluationResult as a, asJudge as c, EvaluationJudgeOptions as i, createEvaluator as l, EvaluationAnalystOptions as n, EvaluatorOptions as o, EvaluationContext as r, asAnalyst as s, CostedEvaluator as t };
-//# sourceMappingURL=evaluation-jTjC_QQD.d.ts.map
+//# sourceMappingURL=evaluation-BOZ38vOH.d.ts.map

package/dist/{evaluation-jTjC_QQD.d.ts.map → evaluation-BOZ38vOH.d.ts.map}

@@ -1 +1 @@
-{"version":3,"file":"evaluation-jTjC_QQD.d.ts","names":[],"sources":["../src/evaluation.ts"],"mappings":";;;;UAWiB,iBAAiB;EAChC,OAAO;EACP,SAAS;EACT;;UAGe;;;EAGf;EACA,SAAS;EACT,aAAa;EACb;EACA,UAAU;EACV;EACA,WAAW;EACX,gBAAgB;;EAEhB,aAAa,SAAS,uBAAuB;;KAGnC,gBAAgB,GAAG,MAC7B,OAAO,GACP,UAAU,sBACP,QAAQ,iBAAiB;UAEb,iBAAiB,GAAG;EACnC,UAAU,OAAO,GAAG;IAAW,QAAQ;IAAa;QAA6B,QAAQ;EACzF,UAAU,OAAO,MAAM;EACvB,oBAAoB,OAAO,UAAU;EACrC,mBAAmB,OAAO;EAC1B,aAAa;EACb,gBAAgB;;EAEhB,YAAY,OAAO,GAAG,OAAO,aAAa;;;iBAI5B,gBAAgB,GAAG,GAAG,SAAS,iBAAiB,GAAG,KAAK,gBAAgB,GAAG;UA6B1E,uBAAuB,GAAG,UAAU,UAAU;EAC7D;EACA;EACA,YAAY,YAAY,GAAG;EAC3B,UAAU;IAAkB,UAAU;IAAG,UAAU;KAAK;EACxD,MAAM,OAAO,GAAG;IAAS,UAAU;IAAG,UAAU;QAAQ,aAAa,QAAQ;;EAE7E,UACE,QAAQ,iBAAiB,IACzB;IAAS,UAAU;IAAG,UAAU;eACtB;EACZ,aAAa,UAAU;;;iBAIT,QAAQ,GAAG,UAAU,UAAU,GAC7C,SAAS,uBAAuB,GAAG,GAAG,KACrC,YAAY,GAAG;UA4BD,yBAAyB,GAAG;EAC3C;EACA;EACA;EACA,WAAW,QAAQ;EACnB,MAAM,QAAQ;EACd,WACE,OAAO,GACP,SAAS,mBACT,gBAAgB,mBACb,QAAQ,iBAAiB;EAC9B,MAAM,OAAO,GAAG,OAAO,GAAG,SAAS,mBAAmB,mBAAmB,QAAQ;EACjF,UAAU,QAAQ,iBAAiB,IAAI,OAAO,GAAG,SAAS,0BAA0B;;;iBAItE,UAAU,GAAG,GAAG,SAAS,yBAAyB,GAAG,KAAK,QAAQ"}
+{"version":3,"file":"evaluation-BOZ38vOH.d.ts","names":[],"sources":["../src/evaluation.ts"],"mappings":";;;;UAWiB,iBAAiB;EAChC,OAAO;EACP,SAAS;EACT;;UAGe;;;EAGf;EACA,SAAS;EACT,aAAa;EACb;EACA,UAAU;EACV;EACA,WAAW;EACX,gBAAgB;;EAEhB,aAAa,SAAS,uBAAuB;;KAGnC,gBAAgB,GAAG,MAC7B,OAAO,GACP,UAAU,sBACP,QAAQ,iBAAiB;UAEb,iBAAiB,GAAG;EACnC,UAAU,OAAO,GAAG;IAAW,QAAQ;IAAa;QAA6B,QAAQ;EACzF,UAAU,OAAO,MAAM;EACvB,oBAAoB,OAAO,UAAU;EACrC,mBAAmB,OAAO;EAC1B,aAAa;EACb,gBAAgB;;EAEhB,YAAY,OAAO,GAAG,OAAO,aAAa;;;iBAI5B,gBAAgB,GAAG,GAAG,SAAS,iBAAiB,GAAG,KAAK,gBAAgB,GAAG;UA6B1E,uBAAuB,GAAG,UAAU,UAAU;EAC7D;EACA;EACA,YAAY,YAAY,GAAG;EAC3B,UAAU;IAAkB,UAAU;IAAG,UAAU;KAAK;EACxD,MAAM,OAAO,GAAG;IAAS,UAAU;IAAG,UAAU;QAAQ,aAAa,QAAQ;;EAE7E,UACE,QAAQ,iBAAiB,IACzB;IAAS,UAAU;IAAG,UAAU;eACtB;EACZ,aAAa,UAAU;;;iBAIT,QAAQ,GAAG,UAAU,UAAU,GAC7C,SAAS,uBAAuB,GAAG,GAAG,KACrC,YAAY,GAAG;UA4BD,yBAAyB,GAAG;EAC3C;EACA;EACA;EACA,WAAW,QAAQ;EACnB,MAAM,QAAQ;EACd,WACE,OAAO,GACP,SAAS,mBACT,gBAAgB,mBACb,QAAQ,iBAAiB;EAC9B,MAAM,OAAO,GAAG,OAAO,GAAG,SAAS,mBAAmB,mBAAmB,QAAQ;EACjF,UAAU,QAAQ,iBAAiB,IAAI,OAAO,GAAG,SAAS,0BAA0B;;;iBAItE,UAAU,GAAG,GAAG,SAAS,yBAAyB,GAAG,KAAK,QAAQ"}

package/dist/evaluation.d.ts

@@ -1,2 +1,2 @@
-import { a as EvaluationResult, c as asJudge, i as EvaluationJudgeOptions, l as createEvaluator, n as EvaluationAnalystOptions, o as EvaluatorOptions, r as EvaluationContext, s as asAnalyst, t as CostedEvaluator } from "./evaluation-jTjC_QQD.js";
+import { a as EvaluationResult, c as asJudge, i as EvaluationJudgeOptions, l as createEvaluator, n as EvaluationAnalystOptions, o as EvaluatorOptions, r as EvaluationContext, s as asAnalyst, t as CostedEvaluator } from "./evaluation-BOZ38vOH.js";
 export { CostedEvaluator, EvaluationAnalystOptions, EvaluationContext, EvaluationJudgeOptions, EvaluationResult, EvaluatorOptions, asAnalyst, asJudge, createEvaluator };

package/dist/experiment/index.d.ts

@@ -1,5 +1,5 @@
 import { n as bonferroni, r as holm, t as benjaminiHochberg } from "../multiplicity-DIWHvysC.js";
-import { At as EProcessStep, B as powerPreflight, Bt as pairedRiskDifferenceScore, C as signManifest, Ct as FinalEvidenceOutcome, Dt as EProcess, E as MatchedPair, Et as openFinalEvidenceLedger, F as comparePairedArms, Ht as wilson, I as pairArms, J as PromotionPolicy, L as pairRunRecords, Lt as mcnemar, O as PairArmsOptions, Ot as EProcessOptions, Pt as ProportionInterval, R as PowerPreflight, Rt as pairedRiskDifference, S as manifestContentDigest, St as FinalEvidenceMeasurement, Tt as FinalEvidenceReservation, W as EvidenceVector, X as paretoPolicy, Y as buildEvidenceVector, Z as paretoSignificanceGate, _ as HypothesisManifest, at as EvidenceAuthority, b as SignedManifestAlgo, bt as FinalEvidenceError, ct as EvidenceReceipt, d as SequentialObservation, dt as isIndependentEvidence, f as SequentialPairedGate, ft as verifyEvidenceReceipt, g as sequentialPairedGate, h as sequentialDecide, i as PairedHoldout, it as CreateEvidenceReceiptInput, j as PairedArmRow, jt as eProcess, k as PairArmsResult, kt as EProcessState, lt as EvidenceReceiptVerification, n as HeldoutSignificance, nt as CampaignEvidenceContext, o as heldoutSignificance, ot as EvidenceAuthorityKind, p as SequentialPairedGateOptions, r as HeldoutSignificanceOptions, rt as createCampaignEvidenceReceipt, s as pairHoldout, st as EvidenceBinding, u as SequentialGateDecision, ut as createEvidenceReceipt, v as HypothesisResult, w as verifyManifest, wt as FinalEvidenceRecord, x as hashJson, xt as FinalEvidenceLedger, y as SignedManifest, yt as FinalEvidenceConflictError, z as PowerPreflightOptions, zt as pairedRiskDifferenceExact } from "../statistical-heldout-C9AbgREN.js";
+import { At as EProcessStep, B as powerPreflight, Bt as pairedRiskDifferenceScore, C as signManifest, Ct as FinalEvidenceOutcome, Dt as EProcess, E as MatchedPair, Et as openFinalEvidenceLedger, F as comparePairedArms, Ht as wilson, I as pairArms, J as PromotionPolicy, L as pairRunRecords, Lt as mcnemar, O as PairArmsOptions, Ot as EProcessOptions, Pt as ProportionInterval, R as PowerPreflight, Rt as pairedRiskDifference, S as manifestContentDigest, St as FinalEvidenceMeasurement, Tt as FinalEvidenceReservation, W as EvidenceVector, X as paretoPolicy, Y as buildEvidenceVector, Z as paretoSignificanceGate, _ as HypothesisManifest, at as EvidenceAuthority, b as SignedManifestAlgo, bt as FinalEvidenceError, ct as EvidenceReceipt, d as SequentialObservation, dt as isIndependentEvidence, f as SequentialPairedGate, ft as verifyEvidenceReceipt, g as sequentialPairedGate, h as sequentialDecide, i as PairedHoldout, it as CreateEvidenceReceiptInput, j as PairedArmRow, jt as eProcess, k as PairArmsResult, kt as EProcessState, lt as EvidenceReceiptVerification, n as HeldoutSignificance, nt as CampaignEvidenceContext, o as heldoutSignificance, ot as EvidenceAuthorityKind, p as SequentialPairedGateOptions, r as HeldoutSignificanceOptions, rt as createCampaignEvidenceReceipt, s as pairHoldout, st as EvidenceBinding, u as SequentialGateDecision, ut as createEvidenceReceipt, v as HypothesisResult, w as verifyManifest, wt as FinalEvidenceRecord, x as hashJson, xt as FinalEvidenceLedger, y as SignedManifest, yt as FinalEvidenceConflictError, z as PowerPreflightOptions, zt as pairedRiskDifferenceExact } from "../statistical-heldout-BHfLZHrm.js";
 import { c as BOOTSTRAP_GATE_MIN_N, l as PairedBootstrapOptions, m as pairedBootstrap, u as PairedBootstrapResult } from "../paired-promotion-decision-CQMJ5xGN.js";
 import { a as requiredPairedSampleSize, i as pairedMde, n as mcnemarPower, o as requiredSampleSize, r as mcnemarRequiredN, t as mulberry32 } from "../random-DGEyr3zG.js";
 import { at as summarizeEvaluationUnits, it as defineEvaluationClaim, nt as EvaluationClaim, rt as EvaluationUnitSummary } from "../search-ledger-types-p2xKfiDJ.js";

package/dist/experiment/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/experiment/ast/predicate.ts","../../src/experiment/ast/validity.ts","../../src/experiment/ast/budget.ts","../../src/experiment/ast/decision.ts","../../src/experiment/ast/estimand.ts","../../src/experiment/ast/reissue.ts","../../src/experiment/ast/selection.ts","../../src/experiment/budget.ts","../../src/experiment/funnel.ts","../../src/experiment/define.ts","../../src/experiment/evidence-record.ts","../../src/experiment/power.ts","../../src/experiment/randomized-canary.ts","../../src/experiment/prospective-canary-join.ts"],"mappings":";;;;;;;;;;;;;KAQY;EAEN;EACA;EACA;EACA,OAAO;;EAEP;EAAY;EAAe,QAAQ;;EACnC;EAAa,IAAI;;EACjB;EAAa,IAAI;;EACjB;EAAa,IAAI;;;wBAGP,UAAU,QAAQ,gBAAgB;;;;UChBjC;EACf;EACA;EACA;;;KAMU;EAEN;EACA;EACA;EACA;;EAGA;EACA;EACA;EACA;;EAEA;EAA8B,OAAO;;EAErC;EACA;;EAEA;EACA;EACA;IAAO;IAAgB;IAAmB;;;EAE1C;EAAkB;EAAuB;EAAmB;;UAEjD;EACf;EACA;EACA,UAAU;;;;;;;wBA+EI,uBACd,YACA,MAAM,QAAQ;EAAgB;IAC9B;EAAkB;EAAgB;MACjC;;;;UCxHc;EACf;IAAQ;IAAyB;;EACjC;EACA;;UAGe;EACf;EACA;EACA;;;;;;KAuBU;EAEN;EACA;EACA;EACA;EACA;EACA;IAAU;IAAY;;EACtB;;EAGA;EACA;EACA;EACA;EACA;;UAGW;EACf;EACA;EACA;EACA;;UAGe;EACf,WAAW;EACX;;UAmCe;EACf;EACA;IAAe;IAAW;IAAsB;;EAChD;IAAkB;IAAuC;;;;UA+B1C;EACf;EACA;EACA;;;;;KC9HU;EACN;EAAgC;EAAkB;;EAClD;EAAgC;;EAChC;EAA4B;EAAkB;EAAiC;;EAC/E;EAAwB;;EACxB;EAAa,IAAI;;EACjB;EAAa,IAAI;;EACjB;EAAa,IAAI;;;UAGN;EACf,WAAW;IAAiB;IAAe;;EAC3C,YAAY;EACZ,gBAAgB;;wBAGF,kBAAkB,WAAW,WAAW,UAAU;UAyCjD;EACf,MAAM;EACN;EACA;;;;;;;;;KAUU;EACN;EAAe,UAAU;;EAEzB;EACA;EACA;EACA;EACA;IAAmB;IAAuB;;;UAG/B;EACf;EACA;;wBAGc,oBACd,MAAM,cACN,UAAU,oBACT;;;;KC1FS;EACN;EAAoB;EAAa,OAAO;;EACxC;EAAmB,IAAI;;;;;;KAMjB;EACN;EAAc,OAAO;EAAW;;EAChC;EAA4B,OAAO;EAAW;;EAE9C;EACA;EACA;EACA;EACA;;;;;EAKA;;EAEA;;EAGA;EACA;EACA;EACA,WAAW;EACX,aAAa;;UAGF;EACf;EACA;EACA;;;;;UC1Ce;EACf;EACA;EACA;;KAGU;;;;;;;UCGK;EACf;EACA,MAAM;EACN;;;UAIe;EACf;;EAEA;EACA,MAAM;EACN;;;UAIe;EACf;EACA,QAAQ;EACR,aAAa;;;;;;;KAUH;EAEN;EACA;EACA;EACA;IAAe;IAAe;;EAC9B;EACA;;EAGA;;EAEA;EACA,MAAM;EACN;IAAS;IAAe;;;;;UCtCb;EACf;;EAEA;;UAGe;EACf,MAAM;EACN,MAAM;;EAEN;;EAEA;EACA;;EAEA;IAAkB;IAA2B;;;;;;;;wBAQ/B,qBACd,MAAM,mBACN,eAAe,sBACd;;wBAsCa,qBACd,MAAM,mBACN,eAAe,sBACd;;;UC3Dc;EACf;EACA;EACA;EACA;;EAEA,aAAa;;EAEb;;UAGe;EACf;;EAEA;EACA;;EAEA;;UAGe;EACf;EACA;EACA,QAAQ;EACR;EACA,YAAY;;UAGG;EACf;;EAEA;EACA,aAAa;EACb;;;;;;;;;;;wBAYc,YAAY;EAC1B;EACA;EACA,QAAQ;EACR;IAAe;IAAY;IAAc;;IACvC;UA8Fa;EACf,QAAQ;EACR,WAAW;;EAEX,eAAe,eAAe;;;;;;;;wBAShB,qBACd,MAAM,eACN,kBAAkB,mBACjB;;;;;;wBA8Ca,eACd,OAAO,kBACP,QAAQ,mBACP;;;;;;wBA4Ba,kBAAkB,QAAQ;;;UC5LzB;EACf;EACA;;EAEA;;EAEA;;EAEA,OAAO,eAAe;;KAGZ;EAEN;;EAEA;EACA;;EAEA;;EAEA;;EAGA;EACA;EACA;EACA;;;;;;UAOW;EACf;;UAGe;EACf;;EAEA,QAAQ;;EAER;EACA,MAAM;EACN,SAAS;EACT,YAAY;;EAEZ,aAAa,eAAe;;;;;EAK5B,gBAAgB;EAChB,YAAY,eAAe;EAC3B,YAAY,eAAe;EAC3B,UAAU;EACV,cAAc;;EAEd,QAAQ,eAAe;EACvB,OAAO;EACP,SAAS;EACT,gBAAgB;EAChB,UAAU;EACV,iBAAiB;EACjB;;;;;;;;;;wBAwCc,iBAAiB,MAAM,iBAAiB;UA0IvC;;EAEf;EACA;;EAEA;;EAEA;;;KAIG;UAEY;EACf,MAAM;;EAEN;;EAEA,MAAM;EACN;;EAEA;EACA,YAAY;;;wBAIQ,eACpB,MAAM,gBACN;EAAW;IACV,QAAQ;;;;;;wBAkBW,gBACpB,QAAQ,kBACR;EAAa,MAAM;EAAgB;EAAgB;EAAiB;IACnE,QAAQ;;wBAyBW,uBAAuB,QAAQ,mBAAmB;;KAwB5D;EACN;EAA4B,aAAa;;EAEzC;EACA,eAAe;EACf,gBAAgB;;EAEhB;EAA8B,YAAY;;EAC1C;EAAkB;EAAgB;;EAClC;EAAqB;IAAkB;IAAgB;;;;;;;UAM5C;WACN,QAAQ;;EAEjB,MAAM,kBAAkB,mBAAmB;;EAE3C,OAAO,UAAU,oBAAoB;;EAErC,MAAM,kBAAkB,mBAAmB;;EAE3C,OAAO,cAAc,kBAAkB,kBAAkB;IAAW;;;EAEpE,KAAK,cAAc,UAAU,yBAAyB;;EAEtD,KAAK,gBAAgB,uBAAuB;;EAE5C,qBAAqB,uCAAuC;;EAE5D,qBAAqB;IAAY;IAAqB;MAAiB;;EAEvE,eAAe,eAAe,sBAAsB;;EAEpD,SAAS,cAAc,eAAe,mBAAmB;;EAEzD,SACE,cACA;IACM;IAAc,eAAe;;IAC7B;IAAkB;IAAmB;IAAgB;MAC1D;IAAqB,QAAQ;;;;;;;;wBAQZ,qBACpB,QAAQ,mBACP,QAAQ;;;;;;;;;;;;;;;qBC7aE;KAQD,wBAAwB;;cAO9B,2BAAyB,EAAA;;;;GAM7B,EAAA,KAAA;KAEU,sBAAsB,EAAE,aAAa;qBAEpC,8BAA4B,EAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAsCvC,EAAA,KAAA;KAEU,yBAAyB,EAAE,aAAa;;;UCxEnC;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;EAMA;IAAkB;IAAe;;;UAGlB;EACf;EACA;EACA;;UAGe;;EAEf;;EAEA;EACA;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,OAAO;;EAEP;EACA,eAAe;;EAEf;;EAEA,SAAS;;;;;;wBAOK,eAAe,SAAS,wBAAwB;;wBAgKhD,qBAAqB,QAAQ;;;;KCvQjC;;UAGK;EACf;EACA;EACA;EACA,KAAK;;UAGU,0BAA0B;;EAEzC;EACA;;EAEA;EACA;;EAEA;;EAEA;;EAEA;EACA;EACA;EACA;;;UAIe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA;EACA;EACA;EACA;EACA;;EAEA;EACA;;EAEA;;EAEA;;EAEA;;EAEA;;;UAIe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA;EACA;EACA;EACA;EACA;;;KAIU,wBACV,SAAS,qBACT,UAAU;EAGN;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA;EAAiB;;UAEN;EACf,MAAM;EACN;EACA;;KAGU;UASK;EACf;EACA;EACA;EACA;EACA;;EAEA;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,UAAU,OACR;IAEE;IACA;IACA;IACA;;EAGJ;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;IAAqB;IAAa;;EAClC;EACA;IAA0B;IAAa;;EACvC;;EAEA;EACA;EACA;EACA;EACA;EACA,SAAS;;EAET;;;wBAwBc,6BAA6B,eAAe;;wBA6B5C,kCAAkC,eAAe;;wBAiFjD,yBAAyB,MAAM,uBAAuB;;wBAOtD,uBACd,QAAQ,4BACR,QAAQ,qBACR,eAAe,qBACf,cAAc,uBACb;;;;UC3Sc,oCAAoC;EACnD;EACA;;;UAIe;EACf;EACA;EACA;EACA;EACA;;;UAIe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;;UAIe;EACf;EACA;EACA;EACA;EACA;EACA;;UAGe;;EAEf;EACA,QAAQ;;EAER;;EAEA;EACA,sBAAsB;EACtB,mBAAmB;EACnB,qBAAqB;EACrB,kBAAkB;;;;;;;wBA4DJ,8BACd,OAAO,6BACN"}
+{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/experiment/ast/predicate.ts","../../src/experiment/ast/validity.ts","../../src/experiment/ast/budget.ts","../../src/experiment/ast/decision.ts","../../src/experiment/ast/estimand.ts","../../src/experiment/ast/reissue.ts","../../src/experiment/ast/selection.ts","../../src/experiment/budget.ts","../../src/experiment/funnel.ts","../../src/experiment/define.ts","../../src/experiment/evidence-record.ts","../../src/experiment/power.ts","../../src/experiment/randomized-canary.ts","../../src/experiment/prospective-canary-join.ts"],"mappings":";;;;;;;;;;;;;KAQY;EAEN;EACA;EACA;EACA,OAAO;;EAEP;EAAY;EAAe,QAAQ;;EACnC;EAAa,IAAI;;EACjB;EAAa,IAAI;;EACjB;EAAa,IAAI;;;wBAGP,UAAU,QAAQ,gBAAgB;;;;UChBjC;EACf;EACA;EACA;;;KAMU;EAEN;EACA;EACA;EACA;;EAGA;EACA;EACA;EACA;;EAEA;EAA8B,OAAO;;EAErC;EACA;;EAEA;EACA;EACA;IAAO;IAAgB;IAAmB;;;EAE1C;EAAkB;EAAuB;EAAmB;;UAEjD;EACf;EACA;EACA,UAAU;;;;;;;wBA+EI,uBACd,YACA,MAAM,QAAQ;EAAgB;IAC9B;EAAkB;EAAgB;MACjC;;;;UCxHc;EACf;IAAQ;IAAyB;;EACjC;EACA;;UAGe;EACf;EACA;EACA;;;;;;KAuBU;EAEN;EACA;EACA;EACA;EACA;EACA;IAAU;IAAY;;EACtB;;EAGA;EACA;EACA;EACA;EACA;;UAGW;EACf;EACA;EACA;EACA;;UAGe;EACf,WAAW;EACX;;UAmCe;EACf;EACA;IAAe;IAAW;IAAsB;;EAChD;IAAkB;IAAuC;;;;UA+B1C;EACf;EACA;EACA;;;;;KC9HU;EACN;EAAgC;EAAkB;;EAClD;EAAgC;;EAChC;EAA4B;EAAkB;EAAiC;;EAC/E;EAAwB;;EACxB;EAAa,IAAI;;EACjB;EAAa,IAAI;;EACjB;EAAa,IAAI;;;UAGN;EACf,WAAW;IAAiB;IAAe;;EAC3C,YAAY;EACZ,gBAAgB;;wBAGF,kBAAkB,WAAW,WAAW,UAAU;UAyCjD;EACf,MAAM;EACN;EACA;;;;;;;;;KAUU;EACN;EAAe,UAAU;;EAEzB;EACA;EACA;EACA;EACA;IAAmB;IAAuB;;;UAG/B;EACf;EACA;;wBAGc,oBACd,MAAM,cACN,UAAU,oBACT;;;;KC1FS;EACN;EAAoB;EAAa,OAAO;;EACxC;EAAmB,IAAI;;;;;;KAMjB;EACN;EAAc,OAAO;EAAW;;EAChC;EAA4B,OAAO;EAAW;;EAE9C;EACA;EACA;EACA;EACA;;;;;EAKA;;EAEA;;EAGA;EACA;EACA;EACA,WAAW;EACX,aAAa;;UAGF;EACf;EACA;EACA;;;;;UC1Ce;EACf;EACA;EACA;;KAGU;;;;;;;UCGK;EACf;EACA,MAAM;EACN;;;UAIe;EACf;;EAEA;EACA,MAAM;EACN;;;UAIe;EACf;EACA,QAAQ;EACR,aAAa;;;;;;;KAUH;EAEN;EACA;EACA;EACA;IAAe;IAAe;;EAC9B;EACA;;EAGA;;EAEA;EACA,MAAM;EACN;IAAS;IAAe;;;;;UCtCb;EACf;;EAEA;;UAGe;EACf,MAAM;EACN,MAAM;;EAEN;;EAEA;EACA;;EAEA;IAAkB;IAA2B;;;;;;;;wBAQ/B,qBACd,MAAM,mBACN,eAAe,sBACd;;wBAsCa,qBACd,MAAM,mBACN,eAAe,sBACd;;;UC3Dc;EACf;EACA;EACA;EACA;;EAEA,aAAa;;EAEb;;UAGe;EACf;;EAEA;EACA;;EAEA;;UAGe;EACf;EACA;EACA,QAAQ;EACR;EACA,YAAY;;UAGG;EACf;;EAEA;EACA,aAAa;EACb;;;;;;;;;;;wBAYc,YAAY;EAC1B;EACA;EACA,QAAQ;EACR;IAAe;IAAY;IAAc;;IACvC;UA8Fa;EACf,QAAQ;EACR,WAAW;;EAEX,eAAe,eAAe;;;;;;;;wBAShB,qBACd,MAAM,eACN,kBAAkB,mBACjB;;;;;;wBA8Ca,eACd,OAAO,kBACP,QAAQ,mBACP;;;;;;wBA4Ba,kBAAkB,QAAQ;;;UC5LzB;EACf;EACA;;EAEA;;EAEA;;EAEA,OAAO,eAAe;;KAGZ;EAEN;;EAEA;EACA;;EAEA;;EAEA;;EAGA;EACA;EACA;EACA;;;;;;UAOW;EACf;;UAGe;EACf;;EAEA,QAAQ;;EAER;EACA,MAAM;EACN,SAAS;EACT,YAAY;;EAEZ,aAAa,eAAe;;;;;EAK5B,gBAAgB;EAChB,YAAY,eAAe;EAC3B,YAAY,eAAe;EAC3B,UAAU;EACV,cAAc;;EAEd,QAAQ,eAAe;EACvB,OAAO;EACP,SAAS;EACT,gBAAgB;EAChB,UAAU;EACV,iBAAiB;EACjB;;;;;;;;;;wBAwCc,iBAAiB,MAAM,iBAAiB;UA0IvC;;EAEf;EACA;;EAEA;;EAEA;;;KAIG;UAEY;EACf,MAAM;;EAEN;;EAEA,MAAM;EACN;;EAEA;EACA,YAAY;;;wBAIQ,eACpB,MAAM,gBACN;EAAW;IACV,QAAQ;;;;;;wBAkBW,gBACpB,QAAQ,kBACR;EAAa,MAAM;EAAgB;EAAgB;EAAiB;IACnE,QAAQ;;wBAyBW,uBAAuB,QAAQ,mBAAmB;;KAwB5D;EACN;EAA4B,aAAa;;EAEzC;EACA,eAAe;EACf,gBAAgB;;EAEhB;EAA8B,YAAY;;EAC1C;EAAkB;EAAgB;;EAClC;EAAqB;IAAkB;IAAgB;;;;;;;UAM5C;WACN,QAAQ;;EAEjB,MAAM,kBAAkB,mBAAmB;;EAE3C,OAAO,UAAU,oBAAoB;;EAErC,MAAM,kBAAkB,mBAAmB;;EAE3C,OAAO,cAAc,kBAAkB,kBAAkB;IAAW;;;EAEpE,KAAK,cAAc,UAAU,yBAAyB;;EAEtD,KAAK,gBAAgB,uBAAuB;;EAE5C,qBAAqB,uCAAuC;;EAE5D,qBAAqB;IAAY;IAAqB;MAAiB;;EAEvE,eAAe,eAAe,sBAAsB;;EAEpD,SAAS,cAAc,eAAe,mBAAmB;;EAEzD,SACE,cACA;IACM;IAAc,eAAe;;IAC7B;IAAkB;IAAmB;IAAgB;MAC1D;IAAqB,QAAQ;;;;;;;;wBAQZ,qBACpB,QAAQ,mBACP,QAAQ;;;;;;;;;;;;;;;qBC7aE;KAQD,wBAAwB;;cAO9B,2BAAyB,EAAA;;;;GAM7B,EAAA,KAAA;KAEU,sBAAsB,EAAE,aAAa;qBAEpC,8BAA4B,EAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAsCvC,EAAA,KAAA;KAEU,yBAAyB,EAAE,aAAa;;;UCxEnC;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;EAMA;IAAkB;IAAe;;;UAGlB;EACf;EACA;EACA;;UAGe;;EAEf;;EAEA;EACA;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,OAAO;;EAEP;EACA,eAAe;;EAEf;;EAEA,SAAS;;;;;;wBAOK,eAAe,SAAS,wBAAwB;;wBAgKhD,qBAAqB,QAAQ;;;;KCvQjC;;UAGK;EACf;EACA;EACA;EACA,KAAK;;UAGU,0BAA0B;;EAEzC;EACA;;EAEA;EACA;;EAEA;;EAEA;;EAEA;EACA;EACA;EACA;;;UAIe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA;EACA;EACA;EACA;EACA;;EAEA;EACA;;EAEA;;EAEA;;EAEA;;EAEA;;;UAIe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA;EACA;EACA;EACA;EACA;;;KAIU,wBACV,SAAS,qBACT,UAAU;EAGN;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA;EAAiB;;UAEN;EACf,MAAM;EACN;EACA;;KAGU;UASK;EACf;EACA;EACA;EACA;EACA;;EAEA;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,UAAU,OACR;IAEE;IACA;IACA;IACA;;EAGJ;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;IAAqB;IAAa;;EAClC;EACA;IAA0B;IAAa;;EACvC;;EAEA;EACA;EACA;EACA;EACA;EACA,SAAS;;EAET;;;wBAwBc,6BAA6B,eAAe;;wBA6B5C,kCAAkC,eAAe;;wBAiFjD,yBAAyB,MAAM,uBAAuB;;wBAoYtD,uBACd,QAAQ,4BACR,QAAQ,qBACR,eAAe,qBACf,cAAc,uBACb;;;;UCxqBc,oCAAoC;EACnD;EACA;;;UAIe;EACf;EACA;EACA;EACA;EACA;;;UAIe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;;UAIe;EACf;EACA;EACA;EACA;EACA;EACA;;UAGe;;EAEf;EACA,QAAQ;;EAER;;EAEA;EACA,sBAAsB;EACtB,mBAAmB;EACnB,qBAAqB;EACrB,kBAAkB;;;;;;;wBA4DJ,8BACd,OAAO,6BACN"}

package/dist/experiment/index.js

@@ -9,7 +9,7 @@ import { c as studentTCdf, i as pairedBootstrap, l as studentTQuantile, n as BOO
 import { a as requiredSampleSize, i as requiredPairedSampleSize, n as mcnemarRequiredN, r as pairedMde, t as mcnemarPower } from "../power-and-mde-N5nU4_Ly.js";
 import { t as eProcess } from "../sequential-eprocess-DodI6RoF.js";
 import { a as heldoutSignificance, o as pairHoldout } from "../statistical-heldout-Bksv92Th.js";
-import { S as openFinalEvidenceLedger, b as FinalEvidenceConflictError, i as verifyEvidenceReceipt, n as createEvidenceReceipt, r as isIndependentEvidence, t as createCampaignEvidenceReceipt, x as FinalEvidenceError, y as powerPreflight } from "../campaign-evidence-CQYw8_0H.js";
+import { S as openFinalEvidenceLedger, b as FinalEvidenceConflictError, i as verifyEvidenceReceipt, n as createEvidenceReceipt, r as isIndependentEvidence, t as createCampaignEvidenceReceipt, x as FinalEvidenceError, y as powerPreflight } from "../campaign-evidence-DunzcJGv.js";
 import { n as evaluatePredicate, r as readField, t as compareValues } from "../predicate-FdKmElA_.js";
 import { r as summarizeEvaluationUnits, t as defineEvaluationClaim } from "../claim-BPVuyHuy.js";
 import { n as pairedEvalueSequence } from "../sequential-BVv1sjsw.js";
@@ -1261,8 +1261,8 @@ function sealRandomizedCanaryRule(rule) {
 		algo: "sha256-rfc8785"
 	};
 }
-/** Apply one registered analysis to the entire frozen assignment roster. */
-function decideRandomizedCanary(sealed, cohort, rows, verifyCohort) {
+/** Check the registration seal, the cohort receipt, the frozen roster and the independent verifier. */
+function verifyCohortEvidence(sealed, cohort, rows, verifyCohort) {
 	if (sealed.algo !== "sha256-rfc8785") throw new ValidationError("randomized canary: unsupported registration digest scheme");
 	validateRule(sealed.rule);
 	if (hashCanonical(sealed.rule) !== sealed.digest) throw new ValidationError("randomized canary: registration digest mismatch");
@@ -1287,57 +1287,83 @@ function decideRandomizedCanary(sealed, cohort, rows, verifyCohort) {
 	if (!Number.isSafeInteger(cohort.assignmentCount) || cohort.assignmentCount < 1 || rows.length !== cohort.assignmentCount) throw new ValidationError("randomized canary: assignment count differs from complete ledger receipt");
 	if (randomizedCanaryRosterDigest(rows) !== cohort.assignmentRosterDigest) throw new ValidationError("randomized canary: assignment roster digest mismatch");
 	if (randomizedCanaryObservationDigest(rows) !== cohort.observationSnapshotDigest) throw new ValidationError("randomized canary: frozen observation snapshot digest mismatch");
+	verifyCohortTimes(sealed, cohort);
+	if (typeof verifyCohort !== "function") throw new ValidationError("randomized canary: independent cohort verifier is required");
+	const verification = verifyCohort(cohort, sealed);
+	if (verification?.verified !== true || verification.authorityId !== cohort.authorityId || verification.protocolWitnessedBeforeTraffic !== true || verification.familySlotReservedBeforeTraffic !== true || verification.eligibilityAndDispositionVerified !== true || verification.randomizationVerified !== true || verification.rosterCompleteAtCutoff !== true || verification.outcomeAndBillingSnapshotFrozen !== true || verification.sourceJoinsVerified !== true || verification.armIsolationVerified !== true) throw new ValidationError(`randomized canary: cohort attestation or completeness unverified${verification?.verified === false ? `: ${verification.reason}` : ""}`);
+}
+/** Require the ledger closure, observation freeze and attestation times to match the registered cutoff. */
+function verifyCohortTimes(sealed, cohort) {
 	const cutoff = Date.parse(sealed.rule.analysisCutoff);
 	const closedAt = Date.parse(cohort.closedAt);
 	const frozenAt = Date.parse(cohort.observationFrozenAt);
 	const attestedAt = Date.parse(cohort.attestedAt);
 	const expectedFreezeAt = cutoff + sealed.rule.outcomeMaturityMs;
 	if (!Number.isFinite(closedAt) || closedAt < cutoff || !cohort.closedAt.endsWith("Z") || !Number.isFinite(frozenAt) || frozenAt !== expectedFreezeAt || closedAt > frozenAt || !cohort.observationFrozenAt.endsWith("Z") || !Number.isFinite(attestedAt) || attestedAt < frozenAt || !cohort.attestedAt.endsWith("Z")) throw new ValidationError("randomized canary: ledger closure or frozen observation time differs from registered cutoffs");
-	if (typeof verifyCohort !== "function") throw new ValidationError("randomized canary: independent cohort verifier is required");
-	const verification = verifyCohort(cohort, sealed);
-	if (verification?.verified !== true || verification.authorityId !== cohort.authorityId || verification.protocolWitnessedBeforeTraffic !== true || verification.familySlotReservedBeforeTraffic !== true || verification.eligibilityAndDispositionVerified !== true || verification.randomizationVerified !== true || verification.rosterCompleteAtCutoff !== true || verification.outcomeAndBillingSnapshotFrozen !== true || verification.sourceJoinsVerified !== true || verification.armIsolationVerified !== true) throw new ValidationError(`randomized canary: cohort attestation or completeness unverified${verification?.verified === false ? `: ${verification.reason}` : ""}`);
-	const clusters = /* @__PURE__ */ new Map();
-	const arms = {
-		control: [],
-		candidate: []
+}
+/** Validate one row's evidence fields and count its gaps; returns whether it is a recorded no-execution. */
+function checkObservationRow(row, index, tally) {
+	if (row.checkedOutcome !== 0 && row.checkedOutcome !== 1 && row.checkedOutcome !== null) throw new ValidationError(`randomized canary: assignment ${index} outcome must be 0, 1, or null`);
+	if (row.checkedOutcome === null) tally.missingOutcomes++;
+	if (row.outcomeSourceId === null) {
+		if (row.checkedOutcome !== null) throw new ValidationError(`randomized canary: assignment ${index} has an outcome without its source`);
+	} else nonEmpty(row.outcomeSourceId, `assignment ${index} outcome source id`);
+	if (row.billedCostUsd === null) tally.missingBilling++;
+	else if (!Number.isFinite(row.billedCostUsd) || row.billedCostUsd < 0) throw new ValidationError(`randomized canary: assignment ${index} billed cost must be nonnegative`);
+	if (row.billingSourceId === null) {
+		if (row.billedCostUsd !== null) throw new ValidationError(`randomized canary: assignment ${index} has billed cost without its source`);
+	} else nonEmpty(row.billingSourceId, `assignment ${index} billing source id`);
+	if (row.turns === null) tally.missingTurns++;
+	else if (!Number.isSafeInteger(row.turns) || row.turns < 0) throw new ValidationError(`randomized canary: assignment ${index} turns must be a nonnegative integer`);
+	if (row.traceSourceId !== null) nonEmpty(row.traceSourceId, `assignment ${index} trace source id`);
+	if (row.noExecutionSourceId !== null) nonEmpty(row.noExecutionSourceId, `assignment ${index} no-execution source id`);
+	if (row.traceSourceId !== null && row.noExecutionSourceId !== null) throw new ValidationError(`randomized canary: assignment ${index} has conflicting trace and no-execution receipts`);
+	const noExecution = row.noExecutionSourceId !== null && row.turns === 0 && row.billedCostUsd === 0 && row.checkedOutcome === 0;
+	if (row.traceSourceId === null && !noExecution) tally.missingTraces++;
+	for (const field of [
+		"servedProfileDigest",
+		"servedCodeRevisionDigest",
+		"outcomeCheckerDigest"
+	]) if (row[field] !== null) nonEmpty(row[field], `assignment ${index} ${field}`);
+	return noExecution;
+}
+/** Whether a row was served by, or checked with, anything other than the registered profile, code and checker. */
+function drifted(row, noExecution, rule) {
+	const expectedProfile = row.arm === "control" ? rule.controlProfileDigest : rule.candidateProfileDigest;
+	return row.traceSourceId !== null && (row.servedProfileDigest !== expectedProfile || row.servedCodeRevisionDigest !== rule.servedCodeRevisionDigest) || row.checkedOutcome !== null && row.outcomeCheckerDigest !== rule.outcomeCheckerDigest || noExecution && (row.servedProfileDigest !== null || row.servedCodeRevisionDigest !== null);
+}
+/** Validate every row against the registered rule, then group by arm and cluster. */
+function tallyObservations(rule, rows) {
+	const tally = {
+		clusters: /* @__PURE__ */ new Map(),
+		arms: {
+			control: [],
+			candidate: []
+		},
+		missingOutcomes: 0,
+		missingBilling: 0,
+		missingTurns: 0,
+		missingTraces: 0,
+		executionDrift: 0
 	};
-	let missingOutcomes = 0;
-	let missingBilling = 0;
-	let missingTurns = 0;
-	let missingTraces = 0;
-	let executionDrift = 0;
 	for (const [index, row] of rows.entries()) {
-		if (row.checkedOutcome !== 0 && row.checkedOutcome !== 1 && row.checkedOutcome !== null) throw new ValidationError(`randomized canary: assignment ${index} outcome must be 0, 1, or null`);
-		if (row.checkedOutcome === null) missingOutcomes++;
-		if (row.outcomeSourceId === null) {
-			if (row.checkedOutcome !== null) throw new ValidationError(`randomized canary: assignment ${index} has an outcome without its source`);
-		} else nonEmpty(row.outcomeSourceId, `assignment ${index} outcome source id`);
-		if (row.billedCostUsd === null) missingBilling++;
-		else if (!Number.isFinite(row.billedCostUsd) || row.billedCostUsd < 0) throw new ValidationError(`randomized canary: assignment ${index} billed cost must be nonnegative`);
-		if (row.billingSourceId === null) {
-			if (row.billedCostUsd !== null) throw new ValidationError(`randomized canary: assignment ${index} has billed cost without its source`);
-		} else nonEmpty(row.billingSourceId, `assignment ${index} billing source id`);
-		if (row.turns === null) missingTurns++;
-		else if (!Number.isSafeInteger(row.turns) || row.turns < 0) throw new ValidationError(`randomized canary: assignment ${index} turns must be a nonnegative integer`);
-		if (row.traceSourceId !== null) nonEmpty(row.traceSourceId, `assignment ${index} trace source id`);
-		if (row.noExecutionSourceId !== null) nonEmpty(row.noExecutionSourceId, `assignment ${index} no-execution source id`);
-		if (row.traceSourceId !== null && row.noExecutionSourceId !== null) throw new ValidationError(`randomized canary: assignment ${index} has conflicting trace and no-execution receipts`);
-		const noExecution = row.noExecutionSourceId !== null && row.turns === 0 && row.billedCostUsd === 0 && row.checkedOutcome === 0;
-		if (row.traceSourceId === null && !noExecution) missingTraces++;
-		for (const field of [
-			"servedProfileDigest",
-			"servedCodeRevisionDigest",
-			"outcomeCheckerDigest"
-		]) if (row[field] !== null) nonEmpty(row[field], `assignment ${index} ${field}`);
-		const expectedProfile = row.arm === "control" ? sealed.rule.controlProfileDigest : sealed.rule.candidateProfileDigest;
-		if (row.traceSourceId !== null && (row.servedProfileDigest !== expectedProfile || row.servedCodeRevisionDigest !== sealed.rule.servedCodeRevisionDigest) || row.checkedOutcome !== null && row.outcomeCheckerDigest !== sealed.rule.outcomeCheckerDigest || noExecution && (row.servedProfileDigest !== null || row.servedCodeRevisionDigest !== null)) executionDrift++;
-		arms[row.arm].push(row);
-		const bucket = clusters.get(row.clusterId) ?? [];
+		if (drifted(row, checkObservationRow(row, index, tally), rule)) tally.executionDrift++;
+		tally.arms[row.arm].push(row);
+		const bucket = tally.clusters.get(row.clusterId) ?? [];
 		bucket.push(row);
-		clusters.set(row.clusterId, bucket);
+		tally.clusters.set(row.clusterId, bucket);
 	}
-	const controlClusters = [...clusters.values()].filter((bucket) => bucket.some((r) => r.arm === "control")).length;
-	const candidateClusters = [...clusters.values()].filter((bucket) => bucket.some((r) => r.arm === "candidate")).length;
+	return tally;
+}
+/** Count the clusters that hold at least one row of the arm. */
+function clustersWithArm(tally, arm) {
+	return [...tally.clusters.values()].filter((bucket) => bucket.some((r) => r.arm === arm)).length;
+}
+/** The decision with every count and total filled in and no interval or pass yet. */
+function baseDecision(sealed, cohort, rowCount, tally) {
+	const { clusters, arms, missingOutcomes, missingBilling, missingTurns } = tally;
+	const controlClusters = clustersWithArm(tally, "control");
+	const candidateClusters = clustersWithArm(tally, "candidate");
 	if (sealed.rule.assignmentUnit === sealed.rule.clusterUnit) {
 		for (const [clusterId, bucket] of clusters) if (bucket.length !== 1) throw new ValidationError(`randomized canary: ${sealed.rule.assignmentUnit} cluster '${clusterId}' has multiple assignments`);
 	}
@@ -1358,7 +1384,7 @@ function decideRandomizedCanary(sealed, cohort, rows, verifyCohort) {
 	const p0 = outcomesComplete ? sum("control", "checkedOutcome") / n0 : null;
 	const p1 = outcomesComplete ? sum("candidate", "checkedOutcome") / n1 : null;
 	const delta = p0 === null || p1 === null ? null : p1 - p0;
-	const result = {
+	return {
 		registrationDigest: sealed.digest,
 		cohortReceiptSourceId: cohort.attestationSourceId,
 		assignmentRosterDigest: cohort.assignmentRosterDigest,
@@ -1366,7 +1392,7 @@ function decideRandomizedCanary(sealed, cohort, rows, verifyCohort) {
 		method: "cluster-robust-t",
 		confidence: .95,
 		familyAdjustedConfidence: 1 - .05 / sealed.rule.confirmatoryFamilySize,
-		assignments: rows.length,
+		assignments: rowCount,
 		controlAssignments: n0,
 		candidateAssignments: n1,
 		clusters: clusters.size,
@@ -1376,8 +1402,8 @@ function decideRandomizedCanary(sealed, cohort, rows, verifyCohort) {
 		missingOutcomes,
 		missingBilling,
 		missingTurns,
-		missingTraces,
-		executionDrift,
+		missingTraces: tally.missingTraces,
+		executionDrift: tally.executionDrift,
 		controlSuccessRate: p0,
 		candidateSuccessRate: p1,
 		delta,
@@ -1393,32 +1419,33 @@ function decideRandomizedCanary(sealed, cohort, rows, verifyCohort) {
 		refusal: null,
 		successCriterionMet: false
 	};
-	if (missingOutcomes > 0 || missingBilling > 0 || missingTurns > 0 || n0 === 0 || n1 === 0) {
-		result.refusal = "missing-evidence";
-		return result;
-	}
-	if (missingTraces > 0) {
-		result.refusal = "missing-trace";
-		return result;
-	}
-	if (executionDrift > 0) {
-		result.refusal = "execution-drift";
-		return result;
-	}
-	if (clusters.size < sealed.rule.minimumClusters || controlClusters < 20 || candidateClusters < 20) {
-		result.refusal = "insufficient-clusters";
-		return result;
-	}
-	for (const bucket of clusters.values()) if (bucket.filter((r) => r.arm === "control").length / n0 > .1 || bucket.filter((r) => r.arm === "candidate").length / n1 > .1) {
-		result.refusal = "dominant-cluster";
-		return result;
-	}
+}
+/** The first reason the evidence cannot support an estimate, or null when it can. */
+function refusalReason(rule, tally, result) {
+	const { clusters } = tally;
+	const n0 = result.controlAssignments;
+	const n1 = result.candidateAssignments;
+	if (tally.missingOutcomes > 0 || tally.missingBilling > 0 || tally.missingTurns > 0 || n0 === 0 || n1 === 0) return "missing-evidence";
+	if (tally.missingTraces > 0) return "missing-trace";
+	if (tally.executionDrift > 0) return "execution-drift";
+	if (clusters.size < rule.minimumClusters || result.controlClusters < 20 || result.candidateClusters < 20) return "insufficient-clusters";
+	for (const bucket of clusters.values()) if (bucket.filter((r) => r.arm === "control").length / n0 > .1 || bucket.filter((r) => r.arm === "candidate").length / n1 > .1) return "dominant-cluster";
+	return null;
+}
+/** Fill the intervals, p-value and passes, or refuse on zero variance. */
+function estimateRiskDifference(rule, tally, result) {
+	const { clusters } = tally;
+	const p0 = result.controlSuccessRate;
+	const p1 = result.candidateSuccessRate;
+	const delta = result.delta;
+	const n0 = result.controlAssignments;
+	const n1 = result.candidateAssignments;
 	const scores = [...clusters.values()].map((bucket) => bucket.reduce((score, row) => score + (row.arm === "candidate" ? (row.checkedOutcome - p1) / n1 : -(row.checkedOutcome - p0) / n0), 0));
 	const G = clusters.size;
-	const variance = G / (G - 1) * (rows.length - 1) / (rows.length - 2) * scores.reduce((total, score) => total + score * score, 0);
+	const variance = G / (G - 1) * (result.assignments - 1) / (result.assignments - 2) * scores.reduce((total, score) => total + score * score, 0);
 	if (!Number.isFinite(variance) || variance <= 0) {
 		result.refusal = "zero-variance";
-		return result;
+		return;
 	}
 	const se = Math.sqrt(variance);
 	const t = delta / se;
@@ -1434,8 +1461,20 @@ function decideRandomizedCanary(sealed, cohort, rows, verifyCohort) {
 	};
 	result.unadjustedPValue = Math.min(1, 2 * (1 - studentTCdf(Math.abs(t), G - 1)));
 	result.headline95Pass = result.nominal95Interval.low > 0;
-	result.familywisePass = result.familyAdjustedInterval.low > sealed.rule.minimumLift;
+	result.familywisePass = result.familyAdjustedInterval.low > rule.minimumLift;
 	result.successCriterionMet = result.headline95Pass && result.familywisePass;
+}
+/** Apply one registered analysis to the entire frozen assignment roster. */
+function decideRandomizedCanary(sealed, cohort, rows, verifyCohort) {
+	verifyCohortEvidence(sealed, cohort, rows, verifyCohort);
+	const tally = tallyObservations(sealed.rule, rows);
+	const result = baseDecision(sealed, cohort, rows.length, tally);
+	const refusal = refusalReason(sealed.rule, tally, result);
+	if (refusal !== null) {
+		result.refusal = refusal;
+		return result;
+	}
+	estimateRiskDifference(sealed.rule, tally, result);
 	return result;
 }
 //#endregion