@tangle-network/agent-eval 0.144.11 → 0.144.13

package/dist/benchmarks/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","names":["clamp01","entries"],"sources":["../../src/benchmarks/calibration.ts","../../src/benchmarks/types.ts","../../src/benchmarks/routing/dataset.ts","../../src/benchmarks/routing/index.ts","../../src/benchmarks/runner.ts","../../src/benchmarks/standard-formats.ts"],"sourcesContent":["import type { BenchmarkAdapter, BenchmarkDatasetItem, BenchmarkEvaluation } from './types'\n\nexport interface BenchmarkMetricCalibrationOptions<TPayload = unknown, TArtifact = string> {\n  adapter: BenchmarkAdapter<BenchmarkDatasetItem<TPayload>, TPayload, TArtifact>\n  item: BenchmarkDatasetItem<TPayload>\n  weakArtifact: TArtifact\n  strongArtifact: TArtifact\n  maxWeakScore?: number\n  minStrongScore?: number\n  minGap?: number\n}\n\nexport interface BenchmarkMetricCalibrationResult {\n  passed: boolean\n  weak: BenchmarkEvaluation\n  strong: BenchmarkEvaluation\n  weakScore: number\n  strongScore: number\n  gap: number\n  reasons: string[]\n}\n\nexport async function calibrateBenchmarkMetric<TPayload = unknown, TArtifact = string>(\n  options: BenchmarkMetricCalibrationOptions<TPayload, TArtifact>,\n): Promise<BenchmarkMetricCalibrationResult> {\n  const weak = await options.adapter.evaluate(options.item, options.weakArtifact)\n  const strong = await options.adapter.evaluate(options.item, options.strongArtifact)\n  const weakScore = clamp01(weak.score)\n  const strongScore = clamp01(strong.score)\n  const maxWeakScore = options.maxWeakScore ?? 0.3\n  const minStrongScore = options.minStrongScore ?? 0.7\n  const minGap = options.minGap ?? 0.4\n  const gap = strongScore - weakScore\n  const reasons = [\n    weakScore <= maxWeakScore\n      ? undefined\n      : `weak score ${weakScore.toFixed(3)} exceeds max ${maxWeakScore.toFixed(3)}`,\n    strongScore >= minStrongScore\n      ? undefined\n      : `strong score ${strongScore.toFixed(3)} below min ${minStrongScore.toFixed(3)}`,\n    gap >= minGap ? undefined : `gap ${gap.toFixed(3)} below min ${minGap.toFixed(3)}`,\n  ].filter((reason): reason is string => Boolean(reason))\n\n  return {\n    passed: reasons.length === 0,\n    weak,\n    strong,\n    weakScore,\n    strongScore,\n    gap,\n    reasons,\n  }\n}\n\nfunction clamp01(value: number): number {\n  if (!Number.isFinite(value)) return 0\n  if (value < 0) return 0\n  if (value > 1) return 1\n  return value\n}\n","/**\n * Shared types for the reference benchmark wrappers under\n * `src/benchmarks/`. Each wrapper exports the three functions in\n * `BenchmarkAdapter` plus its own typed `DatasetItem` shape.\n */\n\nimport type { DispatchContext, Scenario } from '../campaign/types'\nimport type { RunSplitTag } from '../run-record'\n\nexport type BenchmarkTaskKind =\n  | 'retrieval'\n  | 'rag-answer'\n  | 'hallucination'\n  | 'kb-improvement'\n  | 'routing'\n  | 'custom'\n\nexport type BenchmarkFamily =\n  | 'beir'\n  | 'mteb-retrieval'\n  | 'msmarco'\n  | 'trec-dl'\n  | 'miracl'\n  | 'lotte'\n  | 'bright'\n  | 'crag'\n  | 'hotpotqa'\n  | 'kilt'\n  | 'ragtruth'\n  | 'faithbench'\n  | 'first-party'\n  | 'custom'\n\nexport interface BenchmarkDatasetItem<TPayload = unknown> {\n  /** Stable dataset-local item id (used for split assignment + paper\n   *  references). Unique within a benchmark. */\n  id: string\n  /** Free-form payload. Each benchmark defines its own shape. */\n  payload: TPayload\n  /** Optional precomputed split. When absent, adapters use assignSplit(id). */\n  split?: RunSplitTag\n  /** Benchmark family this row came from, e.g. `beir` or `crag`. */\n  family?: BenchmarkFamily | string\n  /** Benchmark-local task kind, e.g. retrieval vs answer quality. */\n  taskKind?: BenchmarkTaskKind | string\n  /** Slice labels such as language, domain, freshness, multihop, long-tail. */\n  tags?: string[]\n  /** Dataset provenance, version, URL, or license notes. */\n  source?: BenchmarkSource\n  metadata?: Record<string, unknown>\n}\n\nexport interface BenchmarkEvaluation {\n  /** [0, 1] score for the response on this item. Exact-match\n   *  benchmarks use 0/1; partial-credit benchmarks may return\n   *  fractional values. */\n  score: number\n  /** Optional pass/fail projection. Defaults to `score > 0` when absent. */\n  passed?: boolean\n  /** Numeric sub-metrics. These become report dimensions. */\n  dimensions?: Record<string, number>\n  /** Optional bag of raw scoring signals — e.g. parsed numeric\n   *  answer, regex match, judge sub-scores. */\n  raw: Record<string, unknown>\n  notes?: string\n}\n\nexport interface BenchmarkSource {\n  name?: string\n  url?: string\n  version?: string\n  license?: string\n  citation?: string\n}\n\n/** Common signature implemented by every adapter under `src/benchmarks/*`. */\n// `TPayload` is the per-item payload type; `_TItem` is preserved for\n// downstream type-narrowing extensions (a richer `BenchmarkDatasetItem`\n// subclass that adds e.g. provenance metadata) but is intentionally\n// unused here. `noUnusedLocals` requires the leading underscore.\nexport interface BenchmarkAdapter<_TItem = unknown, TPayload = unknown, TArtifact = string> {\n  /** Stable benchmark id such as `beir/nfcorpus` or `crag/smoke`. */\n  id?: string\n  family?: BenchmarkFamily | string\n  taskKind?: BenchmarkTaskKind | string\n  description?: string\n  source?: BenchmarkSource\n  defaultMetric?: string\n  /** Load the dataset for the given split. May hit the network on\n   *  first call but should be cache-friendly. Adapters that don't\n   *  ship the dataset itself MUST throw a clearly-marked error\n   *  pointing the caller at the loader script. */\n  loadDataset(split: RunSplitTag): Promise<BenchmarkDatasetItem<TPayload>[]>\n  /** Score a single response. Pure with respect to the inputs. */\n  evaluate(item: BenchmarkDatasetItem<TPayload>, artifact: TArtifact): Promise<BenchmarkEvaluation>\n  /** Deterministic split assignment via item id hashing. The\n   *  fraction of items in each split is implementation-defined but\n   *  MUST be stable across processes and platforms. */\n  assignSplit(itemId: string): RunSplitTag\n}\n\nexport interface BenchmarkScenario<TPayload = unknown> extends Scenario {\n  kind: 'benchmark'\n  benchmarkId: string\n  family: BenchmarkFamily | string\n  taskKind: BenchmarkTaskKind | string\n  splitTag: RunSplitTag\n  item: BenchmarkDatasetItem<TPayload>\n}\n\nexport type BenchmarkResponder<TPayload = unknown, TArtifact = string> = (input: {\n  scenario: BenchmarkScenario<TPayload>\n  item: BenchmarkDatasetItem<TPayload>\n  context: DispatchContext\n}) => Promise<TArtifact> | TArtifact\n\n// ── Deterministic split assignment ───────────────────────────────────\n\n/**\n * 32-bit FNV-1a hash. Stable, allocation-free, deterministic across\n * runtimes. We use it to assign items to splits rather than depending\n * on a polyfilled crypto.subtle path.\n */\nfunction fnv1a32(input: string): number {\n  let h = 0x811c9dc5\n  for (let i = 0; i < input.length; i++) {\n    h ^= input.charCodeAt(i) & 0xff\n    h = (h + ((h << 1) + (h << 4) + (h << 7) + (h << 8) + (h << 24))) >>> 0\n  }\n  return h >>> 0\n}\n\n/** Split-assignment seed shared across all benchmarks. Bumping this\n *  value reshuffles every split — do NOT do that lightly. */\nexport const BENCHMARK_SPLIT_SEED = 'agent-eval-v1'\n\n/**\n * Assign an item id to one of `'search' | 'dev' | 'holdout'` using a\n * stable 32-bit hash of `${seed}::${id}`. Default proportions:\n *\n *   search: 60%   (optimization-readable)\n *   dev:    20%   (held-out for tuning, leak-on-purpose during dev)\n *   holdout:20%   (paper-grade held-out, gated reads)\n */\nexport function deterministicSplit(\n  itemId: string,\n  seed: string = BENCHMARK_SPLIT_SEED,\n): RunSplitTag {\n  const h = fnv1a32(`${seed}::${itemId}`)\n  const pos = h / 0x100000000\n  if (pos < 0.6) return 'search'\n  if (pos < 0.8) return 'dev'\n  return 'holdout'\n}\n","/**\n * Synthetic routing dataset. 16 tasks across 4 categories. Used as a\n * deterministic, dependency-free benchmark for any router that maps a\n * natural-language request to one of a fixed set of route labels.\n *\n * Format (see `routing/README.md` for prose):\n *\n *   {\n *     id:           stable per-task ID (matches across processes).\n *     category:     one of the four route labels.\n *     prompt:       the user-facing request the router must classify.\n *     route:        the ground-truth route the router should pick.\n *     synonyms:     other strings that count as a correct answer.\n *     hardNegatives:close-but-wrong route labels — used to detect the\n *                   \"always picks the popular route\" failure mode.\n *   }\n *\n * The four categories are intentionally cross-domain (file ops,\n * math, search, conversation) so a router that collapses to one\n * category is easy to spot.\n */\n\nexport interface RoutingItem {\n  id: string\n  category: 'file' | 'math' | 'search' | 'chat'\n  prompt: string\n  /** Canonical correct route label. */\n  route: string\n  /** Alternate route labels that also count as correct. */\n  synonyms: string[]\n  /** Wrong-but-tempting route labels (for analysis, not grading). */\n  hardNegatives: string[]\n}\n\nexport const ROUTING_DATASET: RoutingItem[] = [\n  {\n    id: 'file_001',\n    category: 'file',\n    prompt: 'Save the meeting notes to /tmp/notes-2025-04.md as markdown.',\n    route: 'fs.write',\n    synonyms: ['filesystem.write', 'write_file'],\n    hardNegatives: ['fs.read', 'chat.reply'],\n  },\n  {\n    id: 'file_002',\n    category: 'file',\n    prompt: 'Read the contents of /etc/hosts and summarize the entries.',\n    route: 'fs.read',\n    synonyms: ['filesystem.read', 'read_file'],\n    hardNegatives: ['fs.write', 'search.web'],\n  },\n  {\n    id: 'file_003',\n    category: 'file',\n    prompt: 'List every Python file under src/ recursively.',\n    route: 'fs.list',\n    synonyms: ['filesystem.list', 'list_files'],\n    hardNegatives: ['fs.read', 'search.code'],\n  },\n  {\n    id: 'file_004',\n    category: 'file',\n    prompt: 'Delete the cached build at .turbo/cache.',\n    route: 'fs.delete',\n    synonyms: ['filesystem.delete', 'remove_file'],\n    hardNegatives: ['fs.write', 'fs.list'],\n  },\n  {\n    id: 'math_001',\n    category: 'math',\n    prompt: 'What is the integral of 3x^2 + 2x from 0 to 5?',\n    route: 'math.integral',\n    synonyms: ['calculator.integral', 'math.solve'],\n    hardNegatives: ['math.derivative', 'chat.reply'],\n  },\n  {\n    id: 'math_002',\n    category: 'math',\n    prompt: 'Compute the derivative of sin(x) * cos(x).',\n    route: 'math.derivative',\n    synonyms: ['calculator.derivative', 'math.solve'],\n    hardNegatives: ['math.integral', 'math.algebra'],\n  },\n  {\n    id: 'math_003',\n    category: 'math',\n    prompt: 'Solve 2x + 7 = 19 for x.',\n    route: 'math.algebra',\n    synonyms: ['calculator.algebra', 'math.solve'],\n    hardNegatives: ['math.derivative', 'math.integral'],\n  },\n  {\n    id: 'math_004',\n    category: 'math',\n    prompt: 'What is the prime factorization of 360?',\n    route: 'math.numbertheory',\n    synonyms: ['calculator.factor', 'math.solve'],\n    hardNegatives: ['math.algebra', 'search.web'],\n  },\n  {\n    id: 'search_001',\n    category: 'search',\n    prompt: 'Find recent papers on agent prompt optimization with held-out promotion gates.',\n    route: 'search.web',\n    synonyms: ['web.search', 'search.papers'],\n    hardNegatives: ['search.code', 'chat.reply'],\n  },\n  {\n    id: 'search_002',\n    category: 'search',\n    prompt: 'Search the codebase for every call site of `runProposeReview`.',\n    route: 'search.code',\n    synonyms: ['code.search', 'grep'],\n    hardNegatives: ['search.web', 'fs.read'],\n  },\n  {\n    id: 'search_003',\n    category: 'search',\n    prompt: 'What is the latest release of the Tangle network on GitHub?',\n    route: 'search.web',\n    synonyms: ['web.search', 'github.releases'],\n    hardNegatives: ['search.code', 'chat.reply'],\n  },\n  {\n    id: 'search_004',\n    category: 'search',\n    prompt: 'Find all TODO comments in the agent-eval src tree.',\n    route: 'search.code',\n    synonyms: ['code.search', 'grep'],\n    hardNegatives: ['search.web', 'fs.list'],\n  },\n  {\n    id: 'chat_001',\n    category: 'chat',\n    prompt: 'Hi there, how are you doing today?',\n    route: 'chat.reply',\n    synonyms: ['conversation.reply'],\n    hardNegatives: ['search.web', 'fs.read'],\n  },\n  {\n    id: 'chat_002',\n    category: 'chat',\n    prompt: 'Please explain the difference between an LLM and a foundation model.',\n    route: 'chat.reply',\n    synonyms: ['conversation.reply', 'qa.answer'],\n    hardNegatives: ['search.web', 'math.algebra'],\n  },\n  {\n    id: 'chat_003',\n    category: 'chat',\n    prompt: 'Tell me a short joke about distributed systems.',\n    route: 'chat.reply',\n    synonyms: ['conversation.reply'],\n    hardNegatives: ['search.web', 'fs.read'],\n  },\n  {\n    id: 'chat_004',\n    category: 'chat',\n    prompt: 'Acknowledge my last message with a thumbs up.',\n    route: 'chat.reply',\n    synonyms: ['conversation.reply', 'react'],\n    hardNegatives: ['fs.write', 'search.web'],\n  },\n]\n","/**\n * Routing benchmark — synthetic, dependency-free, ships in the\n * package. 16 cross-category items in `dataset.ts`. See\n * `routing/README.md` for the format.\n *\n * `evaluate` does case-insensitive exact match against the canonical\n * route plus declared synonyms. The first valid route token in the\n * response wins; everything else is ignored. Wrong answers also\n * report whether they hit a hard negative — useful when triaging\n * \"always picks the popular route\" failure modes.\n */\n\nimport type { RunSplitTag } from '../../run-record'\nimport type { BenchmarkAdapter, BenchmarkDatasetItem, BenchmarkEvaluation } from '../types'\nimport { deterministicSplit } from '../types'\nimport { ROUTING_DATASET, type RoutingItem } from './dataset'\n\nexport type { RoutingItem }\nexport type RoutingPayload = RoutingItem\nexport type RoutingDatasetItem = BenchmarkDatasetItem<RoutingPayload>\n\nclass RoutingAdapter implements BenchmarkAdapter<RoutingDatasetItem, RoutingPayload> {\n  readonly id = 'first-party/routing'\n  readonly family = 'first-party'\n  readonly taskKind = 'routing'\n  readonly description = 'Synthetic fixed-route classification smoke benchmark'\n  readonly defaultMetric = 'route_exact_match'\n\n  async loadDataset(split: RunSplitTag): Promise<RoutingDatasetItem[]> {\n    return ROUTING_DATASET.map((item) => ({ id: item.id, payload: item })).filter(\n      (it) => assignSplitImpl(it.id) === split,\n    )\n  }\n\n  async evaluate(item: RoutingDatasetItem, response: string): Promise<BenchmarkEvaluation> {\n    const tokens = extractRouteTokens(response)\n    const correct = new Set<string>(\n      [item.payload.route, ...item.payload.synonyms].map((s) => s.toLowerCase()),\n    )\n    const hardNeg = new Set<string>(item.payload.hardNegatives.map((s) => s.toLowerCase()))\n    const firstMatch = tokens.find((t) => correct.has(t.toLowerCase())) ?? null\n    const firstHardNeg = tokens.find((t) => hardNeg.has(t.toLowerCase())) ?? null\n    const score = firstMatch ? 1 : 0\n    return {\n      score,\n      raw: {\n        firstToken: tokens[0] ?? null,\n        matchedRoute: firstMatch,\n        hitHardNegative: Boolean(firstHardNeg),\n        hardNegativeRoute: firstHardNeg,\n        category: item.payload.category,\n      },\n    }\n  }\n\n  assignSplit(itemId: string): RunSplitTag {\n    return assignSplitImpl(itemId)\n  }\n}\n\nfunction assignSplitImpl(itemId: string): RunSplitTag {\n  return deterministicSplit(`routing::${itemId}`)\n}\n\n/**\n * Pull route-shaped tokens out of a model response. Routes look like\n * `category.action` (`fs.write`, `chat.reply`). Bare alphanumerics\n * are not routes, but `category.action` patterns are robust to most\n * model wrappers (JSON output, prose explanations, code fences).\n */\nexport function extractRouteTokens(response: string): string[] {\n  const matches = response.match(/[a-z][a-z0-9_]*\\.[a-z][a-z0-9_]*/gi)\n  return matches ?? []\n}\n\nconst adapter = new RoutingAdapter()\n\nexport const loadDataset = adapter.loadDataset.bind(adapter)\nexport const evaluate = adapter.evaluate.bind(adapter)\nexport const assignSplit = adapter.assignSplit.bind(adapter)\nexport { ROUTING_DATASET, RoutingAdapter }\n","import { join } from 'node:path'\nimport {\n  type CampaignResult,\n  type DispatchFn,\n  type JudgeConfig,\n  type JudgeScore,\n  runCampaign,\n  type Scenario,\n} from '../campaign'\nimport { type CampaignStorage, fsCampaignStorage } from '../campaign/storage'\nimport type { RunSplitTag } from '../run-record'\nimport type {\n  BenchmarkAdapter,\n  BenchmarkDatasetItem,\n  BenchmarkEvaluation,\n  BenchmarkFamily,\n  BenchmarkResponder,\n  BenchmarkScenario,\n  BenchmarkTaskKind,\n} from './types'\n\nexport interface BenchmarkRunOptions<TPayload = unknown, TArtifact = string> {\n  adapter: BenchmarkAdapter<BenchmarkDatasetItem<TPayload>, TPayload, TArtifact>\n  respond: BenchmarkResponder<TPayload, TArtifact>\n  splits?: readonly RunSplitTag[]\n  runDir: string\n  repo?: string\n  seed?: number\n  reps?: number\n  resumable?: boolean\n  costCeiling?: number\n  maxConcurrency?: number\n  dispatchTimeoutMs?: number\n  expectUsage?: 'assert' | 'warn' | 'off'\n  storage?: CampaignStorage\n  now?: () => Date\n}\n\nexport interface BenchmarkReport {\n  benchmarkId: string\n  family: BenchmarkFamily | string\n  taskKind: BenchmarkTaskKind | string\n  source?: BenchmarkAdapter['source']\n  runDir: string\n  manifestHash: string\n  seed: number\n  startedAt: string\n  endedAt: string\n  durationMs: number\n  totalItems: number\n  totalCells: number\n  cellsFailed: number\n  cellsCached: number\n  totalCostUsd: number\n  splits: Record<string, BenchmarkSliceSummary>\n  tags: Record<string, BenchmarkSliceSummary>\n  dimensions: Record<string, BenchmarkDistribution>\n  score: BenchmarkDistribution\n  costUsd: BenchmarkDistribution\n  latencyMs: BenchmarkDistribution\n}\n\nexport interface BenchmarkSliceSummary {\n  n: number\n  meanScore: number\n  passRate: number\n  score: BenchmarkDistribution\n  costUsd: BenchmarkDistribution\n  latencyMs: BenchmarkDistribution\n}\n\nexport interface BenchmarkDistribution {\n  n: number\n  min: number\n  mean: number\n  median: number\n  p90: number\n  max: number\n}\n\nexport interface BenchmarkRunResult<TPayload = unknown, TArtifact = string> {\n  scenarios: Array<BenchmarkScenario<TPayload>>\n  campaign: CampaignResult<TArtifact, BenchmarkScenario<TPayload>>\n  report: BenchmarkReport\n  reportJsonPath: string\n  reportMarkdownPath: string\n}\n\nexport async function runBenchmarkAdapter<TPayload = unknown, TArtifact = string>(\n  options: BenchmarkRunOptions<TPayload, TArtifact>,\n): Promise<BenchmarkRunResult<TPayload, TArtifact>> {\n  const storage = options.storage ?? fsCampaignStorage()\n  const benchmarkId = benchmarkIdFor(options.adapter)\n  const scenarios = await loadBenchmarkScenarios(options.adapter, options.splits)\n  const judge = benchmarkAdapterJudge(options.adapter)\n  const dispatch: DispatchFn<BenchmarkScenario<TPayload>, TArtifact> = async (\n    scenario,\n    context,\n  ) => {\n    return options.respond({ scenario, item: scenario.item, context })\n  }\n\n  const campaign = await runCampaign<BenchmarkScenario<TPayload>, TArtifact>({\n    scenarios,\n    dispatch,\n    dispatchRef: `benchmark:${benchmarkId}`,\n    judges: [judge],\n    seed: options.seed,\n    reps: options.reps,\n    resumable: options.resumable,\n    costCeiling: options.costCeiling,\n    maxConcurrency: options.maxConcurrency,\n    dispatchTimeoutMs: options.dispatchTimeoutMs,\n    expectUsage: options.expectUsage ?? 'off',\n    runDir: options.runDir,\n    repo: options.repo,\n    storage,\n    now: options.now,\n  })\n  const report = summarizeBenchmarkCampaign({\n    adapter: options.adapter,\n    scenarios,\n    campaign,\n  })\n  storage.ensureDir(campaign.runDir)\n  const reportJsonPath = join(campaign.runDir, 'benchmark-report.json')\n  const reportMarkdownPath = join(campaign.runDir, 'benchmark-report.md')\n  storage.write(reportJsonPath, `${JSON.stringify(report, null, 2)}\\n`)\n  storage.write(reportMarkdownPath, renderBenchmarkReportMarkdown(report))\n  return { scenarios, campaign, report, reportJsonPath, reportMarkdownPath }\n}\n\nexport async function loadBenchmarkScenarios<TPayload>(\n  adapter: BenchmarkAdapter<BenchmarkDatasetItem<TPayload>, TPayload, unknown>,\n  splits: readonly RunSplitTag[] = ['search', 'dev', 'holdout'],\n): Promise<Array<BenchmarkScenario<TPayload>>> {\n  const benchmarkId = benchmarkIdFor(adapter)\n  const family = adapter.family ?? 'custom'\n  const taskKind = adapter.taskKind ?? 'custom'\n  const out: Array<BenchmarkScenario<TPayload>> = []\n  const seen = new Set<string>()\n  for (const split of splits) {\n    const items = await adapter.loadDataset(split)\n    for (const item of items) {\n      const splitTag = item.split ?? adapter.assignSplit(item.id)\n      if (splitTag !== split) continue\n      const key = `${benchmarkId}:${item.id}`\n      if (seen.has(key)) continue\n      seen.add(key)\n      out.push({\n        id: key,\n        kind: 'benchmark',\n        benchmarkId,\n        family: item.family ?? family,\n        taskKind: item.taskKind ?? taskKind,\n        splitTag,\n        tags: [...new Set([splitTag, ...(item.tags ?? [])])],\n        item,\n      })\n    }\n  }\n  return out\n}\n\nexport function benchmarkAdapterJudge<TPayload, TArtifact>(\n  adapter: BenchmarkAdapter<BenchmarkDatasetItem<TPayload>, TPayload, TArtifact>,\n): JudgeConfig<TArtifact, BenchmarkScenario<TPayload>> {\n  const benchmarkId = benchmarkIdFor(adapter)\n  return {\n    name: `${benchmarkId}:score`,\n    dimensions: [\n      { key: 'score', description: `Primary ${benchmarkId} benchmark score` },\n      { key: 'passed', description: 'Binary pass projection for aggregate reporting' },\n    ],\n    appliesTo: (scenario: Scenario): scenario is BenchmarkScenario<TPayload> => {\n      return scenario.kind === 'benchmark' && scenario.id.startsWith(`${benchmarkId}:`)\n    },\n    async score({ artifact, scenario }) {\n      const evaluation = await adapter.evaluate(scenario.item, artifact)\n      const dimensions = normalizeEvaluationDimensions(evaluation)\n      return {\n        dimensions,\n        composite: clamp01(evaluation.score),\n        notes: evaluation.notes ?? '',\n      }\n    },\n  }\n}\n\nexport function summarizeBenchmarkCampaign<TPayload, TArtifact>(input: {\n  adapter: BenchmarkAdapter<BenchmarkDatasetItem<TPayload>, TPayload, TArtifact>\n  scenarios: Array<BenchmarkScenario<TPayload>>\n  campaign: CampaignResult<TArtifact, BenchmarkScenario<TPayload>>\n}): BenchmarkReport {\n  const scenarioById = new Map(input.scenarios.map((scenario) => [scenario.id, scenario]))\n  const rows = input.campaign.cells.map((cell) => {\n    const scenario = scenarioById.get(cell.scenarioId)\n    const judge = firstJudgeScore(cell.judgeScores)\n    const score = judge?.composite ?? 0\n    return {\n      cell,\n      scenario,\n      score,\n      passed: (judge?.dimensions.passed ?? (score > 0 ? 1 : 0)) >= 1,\n      dimensions: judge?.dimensions ?? {},\n    }\n  })\n  const successful = rows.filter((row) => !row.cell.error)\n  const scoreValues = successful.map((row) => row.score)\n  return {\n    benchmarkId: benchmarkIdFor(input.adapter),\n    family: input.adapter.family ?? 'custom',\n    taskKind: input.adapter.taskKind ?? 'custom',\n    ...(input.adapter.source ? { source: input.adapter.source } : {}),\n    runDir: input.campaign.runDir,\n    manifestHash: input.campaign.manifestHash,\n    seed: input.campaign.seed,\n    startedAt: input.campaign.startedAt,\n    endedAt: input.campaign.endedAt,\n    durationMs: input.campaign.durationMs,\n    totalItems: input.scenarios.length,\n    totalCells: input.campaign.cells.length,\n    cellsFailed: input.campaign.aggregates.cellsFailed,\n    cellsCached: input.campaign.aggregates.cellsCached,\n    totalCostUsd: input.campaign.aggregates.cost.totalCostUsd,\n    splits: summarizeSlices(successful, (row) => row.scenario?.splitTag ?? 'unknown', [\n      'search',\n      'dev',\n      'holdout',\n    ]),\n    tags: summarizeSlices(successful, (row) => row.scenario?.tags ?? []),\n    dimensions: summarizeDimensions(successful.map((row) => row.dimensions)),\n    score: distribution(scoreValues),\n    costUsd: distribution(successful.map((row) => row.cell.costUsd)),\n    latencyMs: distribution(successful.map((row) => row.cell.durationMs)),\n  }\n}\n\nexport function renderBenchmarkReportMarkdown(report: BenchmarkReport): string {\n  const splitRows = Object.entries(report.splits)\n    .map(([split, summary]) => {\n      return `| ${split} | ${summary.n} | ${fmt(summary.meanScore)} | ${fmt(summary.passRate)} | ${fmt(summary.score.p90)} | ${fmt(summary.costUsd.mean)} | ${fmt(summary.latencyMs.p90)} |`\n    })\n    .join('\\n')\n  const dimRows = Object.entries(report.dimensions)\n    .sort(([a], [b]) => a.localeCompare(b))\n    .map(([key, dist]) => `| ${key} | ${dist.n} | ${fmt(dist.mean)} | ${fmt(dist.p90)} |`)\n    .join('\\n')\n  return [\n    `# Benchmark Report: ${report.benchmarkId}`,\n    '',\n    `- family: ${report.family}`,\n    `- task kind: ${report.taskKind}`,\n    `- run dir: ${report.runDir}`,\n    `- manifest: ${report.manifestHash}`,\n    `- cells: ${report.totalCells} total, ${report.cellsFailed} failed, ${report.cellsCached} cached`,\n    `- cost: $${fmt(report.totalCostUsd)}`,\n    `- score: mean ${fmt(report.score.mean)}, median ${fmt(report.score.median)}, p90 ${fmt(report.score.p90)}, n=${report.score.n}`,\n    '',\n    '## Splits',\n    '',\n    '| split | n | mean score | pass rate | score p90 | mean cost | latency p90 ms |',\n    '| --- | ---: | ---: | ---: | ---: | ---: | ---: |',\n    splitRows || '| none | 0 | 0 | 0 | 0 | 0 | 0 |',\n    '',\n    '## Dimensions',\n    '',\n    '| dimension | n | mean | p90 |',\n    '| --- | ---: | ---: | ---: |',\n    dimRows || '| none | 0 | 0 | 0 |',\n    '',\n  ].join('\\n')\n}\n\nfunction benchmarkIdFor(adapter: Pick<BenchmarkAdapter, 'id' | 'family' | 'taskKind'>): string {\n  return adapter.id ?? `${adapter.family ?? 'custom'}/${adapter.taskKind ?? 'custom'}`\n}\n\nfunction normalizeEvaluationDimensions(evaluation: BenchmarkEvaluation): Record<string, number> {\n  const dimensions: Record<string, number> = { score: clamp01(evaluation.score) }\n  for (const [key, value] of Object.entries(evaluation.dimensions ?? {})) {\n    if (Number.isFinite(value)) dimensions[key] = value\n  }\n  dimensions.passed = (evaluation.passed ?? evaluation.score > 0) ? 1 : 0\n  return dimensions\n}\n\nfunction summarizeDimensions(\n  rows: Array<Record<string, number>>,\n): Record<string, BenchmarkDistribution> {\n  const values = new Map<string, number[]>()\n  for (const row of rows) {\n    for (const [key, value] of Object.entries(row)) {\n      if (!Number.isFinite(value)) continue\n      const list = values.get(key) ?? []\n      list.push(value)\n      values.set(key, list)\n    }\n  }\n  return Object.fromEntries([...values.entries()].map(([key, vals]) => [key, distribution(vals)]))\n}\n\nfunction summarizeSlices<T>(\n  rows: T[],\n  keyOf: (row: T) => string | string[],\n  knownKeys: readonly string[] = [],\n): Record<string, BenchmarkSliceSummary> {\n  const grouped = new Map<string, T[]>()\n  for (const key of knownKeys) grouped.set(key, [])\n  for (const row of rows) {\n    const keys = keyOf(row)\n    for (const key of Array.isArray(keys) ? keys : [keys]) {\n      const list = grouped.get(key) ?? []\n      list.push(row)\n      grouped.set(key, list)\n    }\n  }\n  const out: Record<string, BenchmarkSliceSummary> = {}\n  for (const [key, list] of grouped) {\n    const withShape = list as Array<{\n      score: number\n      passed: boolean\n      cell: { costUsd: number; durationMs: number }\n    }>\n    out[key] = {\n      n: list.length,\n      meanScore: mean(withShape.map((row) => row.score)),\n      passRate: mean(withShape.map((row) => (row.passed ? 1 : 0))),\n      score: distribution(withShape.map((row) => row.score)),\n      costUsd: distribution(withShape.map((row) => row.cell.costUsd)),\n      latencyMs: distribution(withShape.map((row) => row.cell.durationMs)),\n    }\n  }\n  return out\n}\n\nfunction firstJudgeScore(\n  judgeScores: CampaignResult<unknown>['cells'][number]['judgeScores'],\n): JudgeScore | undefined {\n  return Object.values(judgeScores)[0]\n}\n\nfunction distribution(values: readonly number[]): BenchmarkDistribution {\n  const finite = [...values].filter(Number.isFinite).sort((a, b) => a - b)\n  if (finite.length === 0) return { n: 0, min: 0, mean: 0, median: 0, p90: 0, max: 0 }\n  return {\n    n: finite.length,\n    min: finite[0]!,\n    mean: mean(finite),\n    median: percentile(finite, 0.5),\n    p90: percentile(finite, 0.9),\n    max: finite[finite.length - 1]!,\n  }\n}\n\nfunction percentile(sortedValues: readonly number[], p: number): number {\n  if (sortedValues.length === 0) return 0\n  const index = Math.min(\n    sortedValues.length - 1,\n    Math.max(0, Math.ceil(p * sortedValues.length) - 1),\n  )\n  return sortedValues[index]!\n}\n\nfunction mean(values: readonly number[]): number {\n  const finite = values.filter(Number.isFinite)\n  if (finite.length === 0) return 0\n  return finite.reduce((sum, value) => sum + value, 0) / finite.length\n}\n\nfunction clamp01(value: number): number {\n  if (!Number.isFinite(value)) return 0\n  if (value < 0) return 0\n  if (value > 1) return 1\n  return value\n}\n\nfunction fmt(value: number): string {\n  if (!Number.isFinite(value)) return '0'\n  return value.toFixed(value === 0 || Math.abs(value) >= 10 ? 0 : 3)\n}\n","import type { RunSplitTag } from '../run-record'\nimport type {\n  BenchmarkAdapter,\n  BenchmarkDatasetItem,\n  BenchmarkFamily,\n  BenchmarkSource,\n  BenchmarkTaskKind,\n} from './types'\nimport { deterministicSplit } from './types'\n\nexport interface StandardRetrievalDocument {\n  id: string\n  title?: string\n  text: string\n  metadata?: Record<string, unknown>\n}\n\nexport interface StandardRetrievalQuery {\n  id: string\n  text: string\n  metadata?: Record<string, unknown>\n}\n\nexport interface StandardRetrievalQrel {\n  queryId: string\n  documentId: string\n  score: number\n}\n\nexport interface StandardRetrievalPayload {\n  queryId: string\n  query: string\n  expectedDocumentIds: string[]\n  expectedScores: Record<string, number>\n  corpus?: Record<string, StandardRetrievalDocument>\n  metadata?: Record<string, unknown>\n}\n\nexport interface BuildStandardRetrievalItemsOptions {\n  benchmarkId: string\n  family: BenchmarkFamily | string\n  queries: readonly StandardRetrievalQuery[]\n  qrels: readonly StandardRetrievalQrel[]\n  corpus?: readonly StandardRetrievalDocument[]\n  includeCorpusInPayload?: boolean\n  source?: BenchmarkSource\n  tags?: readonly string[]\n  splitOf?: (queryId: string) => RunSplitTag\n}\n\nexport interface RetrievalIdAdapterOptions extends BuildStandardRetrievalItemsOptions {\n  responseIdPattern?: RegExp\n  cutoffs?: readonly number[]\n  primaryMetric?: string\n  passMetric?: string\n  passThreshold?: number\n}\n\nexport interface StandardRetrievalResult {\n  id?: string\n  documentId?: string\n  docId?: string\n  score?: number\n}\n\nexport type StandardRetrievalArtifact =\n  | string\n  | readonly string[]\n  | readonly StandardRetrievalResult[]\n  | {\n      ids?: readonly string[]\n      documentIds?: readonly string[]\n      results?: readonly StandardRetrievalResult[]\n    }\n\nexport interface StandardRetrievalEvaluationOptions {\n  responseIdPattern?: RegExp\n  cutoffs?: readonly number[]\n  primaryMetric?: string\n  passMetric?: string\n  passThreshold?: number\n}\n\nexport function parseJsonlRows<T = unknown>(text: string): T[] {\n  return text\n    .split(/\\r?\\n/)\n    .map((line) => line.trim())\n    .filter(Boolean)\n    .map((line, index) => {\n      try {\n        return JSON.parse(line) as T\n      } catch (error) {\n        throw new Error(`invalid JSONL row ${index + 1}: ${(error as Error).message}`)\n      }\n    })\n}\n\nexport function parseTsvRows(text: string): string[][] {\n  return text\n    .split(/\\r?\\n/)\n    .map((line) => line.trim())\n    .filter(Boolean)\n    .filter((line) => !line.startsWith('#'))\n    .map((line) => line.split(/\\t|\\s+/))\n}\n\nexport function parseQrels(text: string): StandardRetrievalQrel[] {\n  return parseTsvRows(text).flatMap((parts, index) => {\n    if (parts.length < 3) return []\n    const [queryId, maybeZeroOrDocId, maybeDocIdOrScore, maybeScore] = parts\n    if (!queryId || !maybeZeroOrDocId || !maybeDocIdOrScore) return []\n    if (queryId.toLowerCase() === 'query-id' || queryId.toLowerCase() === 'qid') return []\n    const documentId = maybeScore === undefined ? maybeZeroOrDocId : maybeDocIdOrScore\n    const scoreText = maybeScore === undefined ? maybeDocIdOrScore : maybeScore\n    const score = Number(scoreText)\n    if (!documentId || !Number.isFinite(score)) {\n      throw new Error(`invalid qrels row ${index + 1}: expected query id, doc id, score`)\n    }\n    return [{ queryId, documentId, score }]\n  })\n}\n\nexport function parseBeirCorpusJsonl(text: string): StandardRetrievalDocument[] {\n  return parseJsonlRows<Record<string, unknown>>(text).map((row, index) => {\n    const id = stringField(row, '_id') ?? stringField(row, 'id')\n    const body = stringField(row, 'text') ?? stringField(row, 'contents')\n    if (!id || body === undefined) {\n      throw new Error(`invalid BEIR corpus row ${index + 1}: expected _id/id and text/contents`)\n    }\n    return {\n      id,\n      title: stringField(row, 'title'),\n      text: body,\n      metadata: stripKnown(row, ['_id', 'id', 'title', 'text', 'contents']),\n    }\n  })\n}\n\nexport function parseBeirQueriesJsonl(text: string): StandardRetrievalQuery[] {\n  return parseJsonlRows<Record<string, unknown>>(text).map((row, index) => {\n    const id = stringField(row, '_id') ?? stringField(row, 'id') ?? stringField(row, 'query_id')\n    const query = stringField(row, 'text') ?? stringField(row, 'query')\n    if (!id || !query) {\n      throw new Error(`invalid BEIR query row ${index + 1}: expected _id/id and text/query`)\n    }\n    return {\n      id,\n      text: query,\n      metadata: stripKnown(row, ['_id', 'id', 'query_id', 'text', 'query']),\n    }\n  })\n}\n\nexport function buildStandardRetrievalItems(\n  options: BuildStandardRetrievalItemsOptions,\n): Array<BenchmarkDatasetItem<StandardRetrievalPayload>> {\n  const qrelsByQuery = new Map<string, StandardRetrievalQrel[]>()\n  for (const qrel of options.qrels) {\n    if (qrel.score <= 0) continue\n    const list = qrelsByQuery.get(qrel.queryId) ?? []\n    list.push(qrel)\n    qrelsByQuery.set(qrel.queryId, list)\n  }\n  const corpus =\n    options.includeCorpusInPayload && options.corpus\n      ? Object.fromEntries(options.corpus.map((document) => [document.id, document]))\n      : undefined\n  return options.queries.flatMap((query) => {\n    const qrels = qrelsByQuery.get(query.id) ?? []\n    if (qrels.length === 0) return []\n    const split =\n      options.splitOf?.(query.id) ?? deterministicSplit(`${options.benchmarkId}:${query.id}`)\n    return [\n      {\n        id: query.id,\n        split,\n        family: options.family,\n        taskKind: 'retrieval',\n        tags: [...new Set([...(options.tags ?? []), split])],\n        ...(options.source ? { source: options.source } : {}),\n        ...(query.metadata ? { metadata: query.metadata } : {}),\n        payload: {\n          queryId: query.id,\n          query: query.text,\n          expectedDocumentIds: qrels.map((qrel) => qrel.documentId),\n          expectedScores: Object.fromEntries(qrels.map((qrel) => [qrel.documentId, qrel.score])),\n          ...(corpus ? { corpus } : {}),\n          ...(query.metadata ? { metadata: query.metadata } : {}),\n        },\n      },\n    ]\n  })\n}\n\nexport function createRetrievalIdBenchmarkAdapter(\n  options: RetrievalIdAdapterOptions,\n): BenchmarkAdapter<\n  BenchmarkDatasetItem<StandardRetrievalPayload>,\n  StandardRetrievalPayload,\n  StandardRetrievalArtifact\n> {\n  const items = buildStandardRetrievalItems(options)\n  return {\n    id: options.benchmarkId,\n    family: options.family,\n    taskKind: 'retrieval' satisfies BenchmarkTaskKind,\n    source: options.source,\n    defaultMetric: options.primaryMetric ?? 'ndcg@10',\n    async loadDataset(split) {\n      return items.filter((item) => item.split === split)\n    },\n    async evaluate(item, artifact) {\n      return evaluateStandardRetrieval(item.payload, artifact, options)\n    },\n    assignSplit(itemId) {\n      return options.splitOf?.(itemId) ?? deterministicSplit(`${options.benchmarkId}:${itemId}`)\n    },\n  }\n}\n\nexport function evaluateStandardRetrieval(\n  payload: StandardRetrievalPayload,\n  artifact: StandardRetrievalArtifact,\n  options: StandardRetrievalEvaluationOptions = {},\n) {\n  const rankedDocumentIds = normalizeRetrievedDocumentIds(\n    artifact,\n    options.responseIdPattern ?? /[A-Za-z0-9_.:/-]+/g,\n  )\n  const cutoffs = normalizeCutoffs(options.cutoffs ?? [1, 3, 5, 10])\n  const dimensions: Record<string, number> = {\n    expected_count: payload.expectedDocumentIds.length,\n    returned_count: rankedDocumentIds.length,\n  }\n  for (const cutoff of cutoffs) {\n    Object.assign(\n      dimensions,\n      retrievalMetricsAtCutoff({\n        rankedDocumentIds,\n        expectedScores: payload.expectedScores,\n        cutoff,\n      }),\n    )\n  }\n  const primaryMetric = options.primaryMetric ?? 'ndcg@10'\n  const passMetric = options.passMetric ?? 'hit@10'\n  const score = dimensions[primaryMetric] ?? dimensions[`ndcg@${cutoffs[cutoffs.length - 1]}`] ?? 0\n  const passScore =\n    dimensions[passMetric] ?? dimensions[`hit@${cutoffs[cutoffs.length - 1]}`] ?? score\n  return {\n    score,\n    passed: passScore >= (options.passThreshold ?? 1),\n    dimensions,\n    raw: {\n      rankedDocumentIds,\n      expectedDocumentIds: payload.expectedDocumentIds,\n      expectedScores: payload.expectedScores,\n    },\n  }\n}\n\nexport function normalizeRetrievedDocumentIds(\n  artifact: StandardRetrievalArtifact,\n  responseIdPattern: RegExp = /[A-Za-z0-9_.:/-]+/g,\n): string[] {\n  if (typeof artifact === 'string') {\n    return uniqueOrdered((artifact.match(responseIdPattern) ?? []).map((value) => value.trim()))\n  }\n  if (Array.isArray(artifact)) {\n    const entries = artifact as readonly (string | StandardRetrievalResult)[]\n    return uniqueOrdered(\n      entries.flatMap((entry) => {\n        if (typeof entry === 'string') return [entry]\n        return [entry.documentId ?? entry.docId ?? entry.id ?? '']\n      }),\n    )\n  }\n  const objectArtifact = artifact as Exclude<StandardRetrievalArtifact, string | readonly unknown[]>\n  return uniqueOrdered(\n    [\n      ...(objectArtifact.documentIds ?? []),\n      ...(objectArtifact.ids ?? []),\n      ...(objectArtifact.results ?? []).map(\n        (entry) => entry.documentId ?? entry.docId ?? entry.id ?? '',\n      ),\n    ].map((value) => value.trim()),\n  )\n}\n\nexport function retrievalMetricsAtCutoff(input: {\n  rankedDocumentIds: readonly string[]\n  expectedScores: Record<string, number>\n  cutoff: number\n}): Record<string, number> {\n  const top = input.rankedDocumentIds.slice(0, input.cutoff)\n  const relevantIds = Object.keys(input.expectedScores).filter(\n    (id) => input.expectedScores[id]! > 0,\n  )\n  const relevant = new Set(relevantIds)\n  const hits = top.filter((id) => relevant.has(id)).length\n  const firstRelevantRank = top.findIndex((id) => relevant.has(id))\n  const prefix = `@${input.cutoff}`\n  return {\n    [`hit${prefix}`]: hits > 0 ? 1 : 0,\n    [`recall${prefix}`]: relevant.size === 0 ? 1 : hits / relevant.size,\n    [`precision${prefix}`]: hits / input.cutoff,\n    [`mrr${prefix}`]: firstRelevantRank === -1 ? 0 : 1 / (firstRelevantRank + 1),\n    [`ndcg${prefix}`]: ndcgAt(input.rankedDocumentIds, input.expectedScores, input.cutoff),\n  }\n}\n\nfunction stringField(row: Record<string, unknown>, key: string): string | undefined {\n  const value = row[key]\n  return typeof value === 'string' ? value : undefined\n}\n\nfunction stripKnown(\n  row: Record<string, unknown>,\n  keys: readonly string[],\n): Record<string, unknown> {\n  const known = new Set(keys)\n  return Object.fromEntries(Object.entries(row).filter(([key]) => !known.has(key)))\n}\n\nfunction normalizeCutoffs(cutoffs: readonly number[]): number[] {\n  return [\n    ...new Set(cutoffs.map((cutoff) => Math.trunc(cutoff)).filter((cutoff) => cutoff > 0)),\n  ].sort((a, b) => a - b)\n}\n\nfunction uniqueOrdered(values: readonly string[]): string[] {\n  const seen = new Set<string>()\n  const out: string[] = []\n  for (const value of values) {\n    const trimmed = value.trim()\n    if (!trimmed || seen.has(trimmed)) continue\n    seen.add(trimmed)\n    out.push(trimmed)\n  }\n  return out\n}\n\nfunction ndcgAt(\n  rankedDocumentIds: readonly string[],\n  expectedScores: Record<string, number>,\n  cutoff: number,\n): number {\n  const dcg = rankedDocumentIds\n    .slice(0, cutoff)\n    .reduce(\n      (sum, documentId, index) => sum + discountedGain(expectedScores[documentId] ?? 0, index),\n      0,\n    )\n  const ideal = Object.values(expectedScores)\n    .filter((score) => score > 0)\n    .sort((a, b) => b - a)\n    .slice(0, cutoff)\n    .reduce((sum, score, index) => sum + discountedGain(score, index), 0)\n  return ideal === 0 ? 1 : dcg / ideal\n}\n\nfunction discountedGain(relevance: number, zeroBasedRank: number): number {\n  if (relevance <= 0) return 0\n  return (2 ** relevance - 1) / Math.log2(zeroBasedRank + 2)\n}\n"],"mappings":";;;;;;AAsBA,eAAsB,yBACpB,SAC2C;CAC3C,MAAM,OAAO,MAAM,QAAQ,QAAQ,SAAS,QAAQ,MAAM,QAAQ,YAAY;CAC9E,MAAM,SAAS,MAAM,QAAQ,QAAQ,SAAS,QAAQ,MAAM,QAAQ,cAAc;CAClF,MAAM,YAAYA,UAAQ,KAAK,KAAK;CACpC,MAAM,cAAcA,UAAQ,OAAO,KAAK;CACxC,MAAM,eAAe,QAAQ,gBAAgB;CAC7C,MAAM,iBAAiB,QAAQ,kBAAkB;CACjD,MAAM,SAAS,QAAQ,UAAU;CACjC,MAAM,MAAM,cAAc;CAC1B,MAAM,UAAU;EACd,aAAa,eACT,KAAA,IACA,cAAc,UAAU,QAAQ,CAAC,EAAE,eAAe,aAAa,QAAQ,CAAC;EAC5E,eAAe,iBACX,KAAA,IACA,gBAAgB,YAAY,QAAQ,CAAC,EAAE,aAAa,eAAe,QAAQ,CAAC;EAChF,OAAO,SAAS,KAAA,IAAY,OAAO,IAAI,QAAQ,CAAC,EAAE,aAAa,OAAO,QAAQ,CAAC;CACjF,CAAC,CAAC,QAAQ,WAA6B,QAAQ,MAAM,CAAC;CAEtD,OAAO;EACL,QAAQ,QAAQ,WAAW;EAC3B;EACA;EACA;EACA;EACA;EACA;CACF;AACF;AAEA,SAASA,UAAQ,OAAuB;CACtC,IAAI,CAAC,OAAO,SAAS,KAAK,GAAG,OAAO;CACpC,IAAI,QAAQ,GAAG,OAAO;CACtB,IAAI,QAAQ,GAAG,OAAO;CACtB,OAAO;AACT;;;;;;;;ACgEA,SAAS,QAAQ,OAAuB;CACtC,IAAI,IAAI;CACR,KAAK,IAAI,IAAI,GAAG,IAAI,MAAM,QAAQ,KAAK;EACrC,KAAK,MAAM,WAAW,CAAC,IAAI;EAC3B,IAAK,MAAM,KAAK,MAAM,KAAK,MAAM,KAAK,MAAM,KAAK,MAAM,KAAK,SAAU;CACxE;CACA,OAAO,MAAM;AACf;;;AAIA,MAAa,uBAAuB;;;;;;;;;AAUpC,SAAgB,mBACd,QACA,OAAe,sBACF;CAEb,MAAM,MADI,QAAQ,GAAG,KAAK,IAAI,QAClB,IAAI;CAChB,IAAI,MAAM,IAAK,OAAO;CACtB,IAAI,MAAM,IAAK,OAAO;CACtB,OAAO;AACT;;;ACvHA,MAAa,kBAAiC;CAC5C;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,oBAAoB,YAAY;EAC3C,eAAe,CAAC,WAAW,YAAY;CACzC;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,mBAAmB,WAAW;EACzC,eAAe,CAAC,YAAY,YAAY;CAC1C;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,mBAAmB,YAAY;EAC1C,eAAe,CAAC,WAAW,aAAa;CAC1C;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,qBAAqB,aAAa;EAC7C,eAAe,CAAC,YAAY,SAAS;CACvC;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,uBAAuB,YAAY;EAC9C,eAAe,CAAC,mBAAmB,YAAY;CACjD;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,yBAAyB,YAAY;EAChD,eAAe,CAAC,iBAAiB,cAAc;CACjD;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,sBAAsB,YAAY;EAC7C,eAAe,CAAC,mBAAmB,eAAe;CACpD;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,qBAAqB,YAAY;EAC5C,eAAe,CAAC,gBAAgB,YAAY;CAC9C;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,cAAc,eAAe;EACxC,eAAe,CAAC,eAAe,YAAY;CAC7C;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,eAAe,MAAM;EAChC,eAAe,CAAC,cAAc,SAAS;CACzC;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,cAAc,iBAAiB;EAC1C,eAAe,CAAC,eAAe,YAAY;CAC7C;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,eAAe,MAAM;EAChC,eAAe,CAAC,cAAc,SAAS;CACzC;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,oBAAoB;EAC/B,eAAe,CAAC,cAAc,SAAS;CACzC;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,sBAAsB,WAAW;EAC5C,eAAe,CAAC,cAAc,cAAc;CAC9C;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,oBAAoB;EAC/B,eAAe,CAAC,cAAc,SAAS;CACzC;CACA;EACE,IAAI;EACJ,UAAU;EACV,QAAQ;EACR,OAAO;EACP,UAAU,CAAC,sBAAsB,OAAO;EACxC,eAAe,CAAC,YAAY,YAAY;CAC1C;AACF;;;;;;;;;;;AC9IA,IAAM,iBAAN,MAAqF;CACnF,KAAc;CACd,SAAkB;CAClB,WAAoB;CACpB,cAAuB;CACvB,gBAAyB;CAEzB,MAAM,YAAY,OAAmD;EACnE,OAAO,gBAAgB,KAAK,UAAU;GAAE,IAAI,KAAK;GAAI,SAAS;EAAK,EAAE,CAAC,CAAC,QACpE,OAAO,gBAAgB,GAAG,EAAE,MAAM,KACrC;CACF;CAEA,MAAM,SAAS,MAA0B,UAAgD;EACvF,MAAM,SAAS,mBAAmB,QAAQ;EAC1C,MAAM,UAAU,IAAI,IAClB,CAAC,KAAK,QAAQ,OAAO,GAAG,KAAK,QAAQ,QAAQ,CAAC,CAAC,KAAK,MAAM,EAAE,YAAY,CAAC,CAC3E;EACA,MAAM,UAAU,IAAI,IAAY,KAAK,QAAQ,cAAc,KAAK,MAAM,EAAE,YAAY,CAAC,CAAC;EACtF,MAAM,aAAa,OAAO,MAAM,MAAM,QAAQ,IAAI,EAAE,YAAY,CAAC,CAAC,KAAK;EACvE,MAAM,eAAe,OAAO,MAAM,MAAM,QAAQ,IAAI,EAAE,YAAY,CAAC,CAAC,KAAK;EAEzE,OAAO;GACL,OAFY,aAAa,IAAI;GAG7B,KAAK;IACH,YAAY,OAAO,MAAM;IACzB,cAAc;IACd,iBAAiB,QAAQ,YAAY;IACrC,mBAAmB;IACnB,UAAU,KAAK,QAAQ;GACzB;EACF;CACF;CAEA,YAAY,QAA6B;EACvC,OAAO,gBAAgB,MAAM;CAC/B;AACF;AAEA,SAAS,gBAAgB,QAA6B;CACpD,OAAO,mBAAmB,YAAY,QAAQ;AAChD;;;;;;;AAQA,SAAgB,mBAAmB,UAA4B;CAE7D,OADgB,SAAS,MAAM,oCAClB,KAAK,CAAC;AACrB;AAEA,MAAM,UAAU,IAAI,eAAe;AAEnC,MAAa,cAAc,QAAQ,YAAY,KAAK,OAAO;AAC3D,MAAa,WAAW,QAAQ,SAAS,KAAK,OAAO;AACrD,MAAa,cAAc,QAAQ,YAAY,KAAK,OAAO;;;ACS3D,eAAsB,oBACpB,SACkD;CAClD,MAAM,UAAU,QAAQ,WAAW,kBAAkB;CACrD,MAAM,cAAc,eAAe,QAAQ,OAAO;CAClD,MAAM,YAAY,MAAM,uBAAuB,QAAQ,SAAS,QAAQ,MAAM;CAC9E,MAAM,QAAQ,sBAAsB,QAAQ,OAAO;CACnD,MAAM,WAA+D,OACnE,UACA,YACG;EACH,OAAO,QAAQ,QAAQ;GAAE;GAAU,MAAM,SAAS;GAAM;EAAQ,CAAC;CACnE;CAEA,MAAM,WAAW,MAAM,YAAoD;EACzE;EACA;EACA,aAAa,aAAa;EAC1B,QAAQ,CAAC,KAAK;EACd,MAAM,QAAQ;EACd,MAAM,QAAQ;EACd,WAAW,QAAQ;EACnB,aAAa,QAAQ;EACrB,gBAAgB,QAAQ;EACxB,mBAAmB,QAAQ;EAC3B,aAAa,QAAQ,eAAe;EACpC,QAAQ,QAAQ;EAChB,MAAM,QAAQ;EACd;EACA,KAAK,QAAQ;CACf,CAAC;CACD,MAAM,SAAS,2BAA2B;EACxC,SAAS,QAAQ;EACjB;EACA;CACF,CAAC;CACD,QAAQ,UAAU,SAAS,MAAM;CACjC,MAAM,iBAAiB,KAAK,SAAS,QAAQ,uBAAuB;CACpE,MAAM,qBAAqB,KAAK,SAAS,QAAQ,qBAAqB;CACtE,QAAQ,MAAM,gBAAgB,GAAG,KAAK,UAAU,QAAQ,MAAM,CAAC,EAAE,GAAG;CACpE,QAAQ,MAAM,oBAAoB,8BAA8B,MAAM,CAAC;CACvE,OAAO;EAAE;EAAW;EAAU;EAAQ;EAAgB;CAAmB;AAC3E;AAEA,eAAsB,uBACpB,SACA,SAAiC;CAAC;CAAU;CAAO;AAAS,GACf;CAC7C,MAAM,cAAc,eAAe,OAAO;CAC1C,MAAM,SAAS,QAAQ,UAAU;CACjC,MAAM,WAAW,QAAQ,YAAY;CACrC,MAAM,MAA0C,CAAC;CACjD,MAAM,uBAAO,IAAI,IAAY;CAC7B,KAAK,MAAM,SAAS,QAAQ;EAC1B,MAAM,QAAQ,MAAM,QAAQ,YAAY,KAAK;EAC7C,KAAK,MAAM,QAAQ,OAAO;GACxB,MAAM,WAAW,KAAK,SAAS,QAAQ,YAAY,KAAK,EAAE;GAC1D,IAAI,aAAa,OAAO;GACxB,MAAM,MAAM,GAAG,YAAY,GAAG,KAAK;GACnC,IAAI,KAAK,IAAI,GAAG,GAAG;GACnB,KAAK,IAAI,GAAG;GACZ,IAAI,KAAK;IACP,IAAI;IACJ,MAAM;IACN;IACA,QAAQ,KAAK,UAAU;IACvB,UAAU,KAAK,YAAY;IAC3B;IACA,MAAM,CAAC,mBAAG,IAAI,IAAI,CAAC,UAAU,GAAI,KAAK,QAAQ,CAAC,CAAE,CAAC,CAAC;IACnD;GACF,CAAC;EACH;CACF;CACA,OAAO;AACT;AAEA,SAAgB,sBACd,SACqD;CACrD,MAAM,cAAc,eAAe,OAAO;CAC1C,OAAO;EACL,MAAM,GAAG,YAAY;EACrB,YAAY,CACV;GAAE,KAAK;GAAS,aAAa,WAAW,YAAY;EAAkB,GACtE;GAAE,KAAK;GAAU,aAAa;EAAiD,CACjF;EACA,YAAY,aAAgE;GAC1E,OAAO,SAAS,SAAS,eAAe,SAAS,GAAG,WAAW,GAAG,YAAY,EAAE;EAClF;EACA,MAAM,MAAM,EAAE,UAAU,YAAY;GAClC,MAAM,aAAa,MAAM,QAAQ,SAAS,SAAS,MAAM,QAAQ;GAEjE,OAAO;IACL,YAFiB,8BAA8B,UAEtC;IACT,WAAW,QAAQ,WAAW,KAAK;IACnC,OAAO,WAAW,SAAS;GAC7B;EACF;CACF;AACF;AAEA,SAAgB,2BAAgD,OAI5C;CAClB,MAAM,eAAe,IAAI,IAAI,MAAM,UAAU,KAAK,aAAa,CAAC,SAAS,IAAI,QAAQ,CAAC,CAAC;CAavF,MAAM,aAZO,MAAM,SAAS,MAAM,KAAK,SAAS;EAC9C,MAAM,WAAW,aAAa,IAAI,KAAK,UAAU;EACjD,MAAM,QAAQ,gBAAgB,KAAK,WAAW;EAC9C,MAAM,QAAQ,OAAO,aAAa;EAClC,OAAO;GACL;GACA;GACA;GACA,SAAS,OAAO,WAAW,WAAW,QAAQ,IAAI,IAAI,OAAO;GAC7D,YAAY,OAAO,cAAc,CAAC;EACpC;CACF,CACsB,CAAC,CAAC,QAAQ,QAAQ,CAAC,IAAI,KAAK,KAAK;CACvD,MAAM,cAAc,WAAW,KAAK,QAAQ,IAAI,KAAK;CACrD,OAAO;EACL,aAAa,eAAe,MAAM,OAAO;EACzC,QAAQ,MAAM,QAAQ,UAAU;EAChC,UAAU,MAAM,QAAQ,YAAY;EACpC,GAAI,MAAM,QAAQ,SAAS,EAAE,QAAQ,MAAM,QAAQ,OAAO,IAAI,CAAC;EAC/D,QAAQ,MAAM,SAAS;EACvB,cAAc,MAAM,SAAS;EAC7B,MAAM,MAAM,SAAS;EACrB,WAAW,MAAM,SAAS;EAC1B,SAAS,MAAM,SAAS;EACxB,YAAY,MAAM,SAAS;EAC3B,YAAY,MAAM,UAAU;EAC5B,YAAY,MAAM,SAAS,MAAM;EACjC,aAAa,MAAM,SAAS,WAAW;EACvC,aAAa,MAAM,SAAS,WAAW;EACvC,cAAc,MAAM,SAAS,WAAW,KAAK;EAC7C,QAAQ,gBAAgB,aAAa,QAAQ,IAAI,UAAU,YAAY,WAAW;GAChF;GACA;GACA;EACF,CAAC;EACD,MAAM,gBAAgB,aAAa,QAAQ,IAAI,UAAU,QAAQ,CAAC,CAAC;EACnE,YAAY,oBAAoB,WAAW,KAAK,QAAQ,IAAI,UAAU,CAAC;EACvE,OAAO,aAAa,WAAW;EAC/B,SAAS,aAAa,WAAW,KAAK,QAAQ,IAAI,KAAK,OAAO,CAAC;EAC/D,WAAW,aAAa,WAAW,KAAK,QAAQ,IAAI,KAAK,UAAU,CAAC;CACtE;AACF;AAEA,SAAgB,8BAA8B,QAAiC;CAC7E,MAAM,YAAY,OAAO,QAAQ,OAAO,MAAM,CAAC,CAC5C,KAAK,CAAC,OAAO,aAAa;EACzB,OAAO,KAAK,MAAM,KAAK,QAAQ,EAAE,KAAK,IAAI,QAAQ,SAAS,EAAE,KAAK,IAAI,QAAQ,QAAQ,EAAE,KAAK,IAAI,QAAQ,MAAM,GAAG,EAAE,KAAK,IAAI,QAAQ,QAAQ,IAAI,EAAE,KAAK,IAAI,QAAQ,UAAU,GAAG,EAAE;CACrL,CAAC,CAAC,CACD,KAAK,IAAI;CACZ,MAAM,UAAU,OAAO,QAAQ,OAAO,UAAU,CAAC,CAC9C,MAAM,CAAC,IAAI,CAAC,OAAO,EAAE,cAAc,CAAC,CAAC,CAAC,CACtC,KAAK,CAAC,KAAK,UAAU,KAAK,IAAI,KAAK,KAAK,EAAE,KAAK,IAAI,KAAK,IAAI,EAAE,KAAK,IAAI,KAAK,GAAG,EAAE,GAAG,CAAC,CACrF,KAAK,IAAI;CACZ,OAAO;EACL,uBAAuB,OAAO;EAC9B;EACA,aAAa,OAAO;EACpB,gBAAgB,OAAO;EACvB,cAAc,OAAO;EACrB,eAAe,OAAO;EACtB,YAAY,OAAO,WAAW,UAAU,OAAO,YAAY,WAAW,OAAO,YAAY;EACzF,YAAY,IAAI,OAAO,YAAY;EACnC,iBAAiB,IAAI,OAAO,MAAM,IAAI,EAAE,WAAW,IAAI,OAAO,MAAM,MAAM,EAAE,QAAQ,IAAI,OAAO,MAAM,GAAG,EAAE,MAAM,OAAO,MAAM;EAC7H;EACA;EACA;EACA;EACA;EACA,aAAa;EACb;EACA;EACA;EACA;EACA;EACA,WAAW;EACX;CACF,CAAC,CAAC,KAAK,IAAI;AACb;AAEA,SAAS,eAAe,SAAuE;CAC7F,OAAO,QAAQ,MAAM,GAAG,QAAQ,UAAU,SAAS,GAAG,QAAQ,YAAY;AAC5E;AAEA,SAAS,8BAA8B,YAAyD;CAC9F,MAAM,aAAqC,EAAE,OAAO,QAAQ,WAAW,KAAK,EAAE;CAC9E,KAAK,MAAM,CAAC,KAAK,UAAU,OAAO,QAAQ,WAAW,cAAc,CAAC,CAAC,GACnE,IAAI,OAAO,SAAS,KAAK,GAAG,WAAW,OAAO;CAEhD,WAAW,SAAU,WAAW,UAAU,WAAW,QAAQ,IAAK,IAAI;CACtE,OAAO;AACT;AAEA,SAAS,oBACP,MACuC;CACvC,MAAM,yBAAS,IAAI,IAAsB;CACzC,KAAK,MAAM,OAAO,MAChB,KAAK,MAAM,CAAC,KAAK,UAAU,OAAO,QAAQ,GAAG,GAAG;EAC9C,IAAI,CAAC,OAAO,SAAS,KAAK,GAAG;EAC7B,MAAM,OAAO,OAAO,IAAI,GAAG,KAAK,CAAC;EACjC,KAAK,KAAK,KAAK;EACf,OAAO,IAAI,KAAK,IAAI;CACtB;CAEF,OAAO,OAAO,YAAY,CAAC,GAAG,OAAO,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,UAAU,CAAC,KAAK,aAAa,IAAI,CAAC,CAAC,CAAC;AACjG;AAEA,SAAS,gBACP,MACA,OACA,YAA+B,CAAC,GACO;CACvC,MAAM,0BAAU,IAAI,IAAiB;CACrC,KAAK,MAAM,OAAO,WAAW,QAAQ,IAAI,KAAK,CAAC,CAAC;CAChD,KAAK,MAAM,OAAO,MAAM;EACtB,MAAM,OAAO,MAAM,GAAG;EACtB,KAAK,MAAM,OAAO,MAAM,QAAQ,IAAI,IAAI,OAAO,CAAC,IAAI,GAAG;GACrD,MAAM,OAAO,QAAQ,IAAI,GAAG,KAAK,CAAC;GAClC,KAAK,KAAK,GAAG;GACb,QAAQ,IAAI,KAAK,IAAI;EACvB;CACF;CACA,MAAM,MAA6C,CAAC;CACpD,KAAK,MAAM,CAAC,KAAK,SAAS,SAAS;EACjC,MAAM,YAAY;EAKlB,IAAI,OAAO;GACT,GAAG,KAAK;GACR,WAAW,KAAK,UAAU,KAAK,QAAQ,IAAI,KAAK,CAAC;GACjD,UAAU,KAAK,UAAU,KAAK,QAAS,IAAI,SAAS,IAAI,CAAE,CAAC;GAC3D,OAAO,aAAa,UAAU,KAAK,QAAQ,IAAI,KAAK,CAAC;GACrD,SAAS,aAAa,UAAU,KAAK,QAAQ,IAAI,KAAK,OAAO,CAAC;GAC9D,WAAW,aAAa,UAAU,KAAK,QAAQ,IAAI,KAAK,UAAU,CAAC;EACrE;CACF;CACA,OAAO;AACT;AAEA,SAAS,gBACP,aACwB;CACxB,OAAO,OAAO,OAAO,WAAW,CAAC,CAAC;AACpC;AAEA,SAAS,aAAa,QAAkD;CACtE,MAAM,SAAS,CAAC,GAAG,MAAM,CAAC,CAAC,OAAO,OAAO,QAAQ,CAAC,CAAC,MAAM,GAAG,MAAM,IAAI,CAAC;CACvE,IAAI,OAAO,WAAW,GAAG,OAAO;EAAE,GAAG;EAAG,KAAK;EAAG,MAAM;EAAG,QAAQ;EAAG,KAAK;EAAG,KAAK;CAAE;CACnF,OAAO;EACL,GAAG,OAAO;EACV,KAAK,OAAO;EACZ,MAAM,KAAK,MAAM;EACjB,QAAQ,WAAW,QAAQ,EAAG;EAC9B,KAAK,WAAW,QAAQ,EAAG;EAC3B,KAAK,OAAO,OAAO,SAAS;CAC9B;AACF;AAEA,SAAS,WAAW,cAAiC,GAAmB;CACtE,IAAI,aAAa,WAAW,GAAG,OAAO;CAKtC,OAAO,aAJO,KAAK,IACjB,aAAa,SAAS,GACtB,KAAK,IAAI,GAAG,KAAK,KAAK,IAAI,aAAa,MAAM,IAAI,CAAC,CAE5B;AAC1B;AAEA,SAAS,KAAK,QAAmC;CAC/C,MAAM,SAAS,OAAO,OAAO,OAAO,QAAQ;CAC5C,IAAI,OAAO,WAAW,GAAG,OAAO;CAChC,OAAO,OAAO,QAAQ,KAAK,UAAU,MAAM,OAAO,CAAC,IAAI,OAAO;AAChE;AAEA,SAAS,QAAQ,OAAuB;CACtC,IAAI,CAAC,OAAO,SAAS,KAAK,GAAG,OAAO;CACpC,IAAI,QAAQ,GAAG,OAAO;CACtB,IAAI,QAAQ,GAAG,OAAO;CACtB,OAAO;AACT;AAEA,SAAS,IAAI,OAAuB;CAClC,IAAI,CAAC,OAAO,SAAS,KAAK,GAAG,OAAO;CACpC,OAAO,MAAM,QAAQ,UAAU,KAAK,KAAK,IAAI,KAAK,KAAK,KAAK,IAAI,CAAC;AACnE;;;ACzSA,SAAgB,eAA4B,MAAmB;CAC7D,OAAO,KACJ,MAAM,OAAO,CAAC,CACd,KAAK,SAAS,KAAK,KAAK,CAAC,CAAC,CAC1B,OAAO,OAAO,CAAC,CACf,KAAK,MAAM,UAAU;EACpB,IAAI;GACF,OAAO,KAAK,MAAM,IAAI;EACxB,SAAS,OAAO;GACd,MAAM,IAAI,MAAM,qBAAqB,QAAQ,EAAE,IAAK,MAAgB,SAAS;EAC/E;CACF,CAAC;AACL;AAEA,SAAgB,aAAa,MAA0B;CACrD,OAAO,KACJ,MAAM,OAAO,CAAC,CACd,KAAK,SAAS,KAAK,KAAK,CAAC,CAAC,CAC1B,OAAO,OAAO,CAAC,CACf,QAAQ,SAAS,CAAC,KAAK,WAAW,GAAG,CAAC,CAAC,CACvC,KAAK,SAAS,KAAK,MAAM,QAAQ,CAAC;AACvC;AAEA,SAAgB,WAAW,MAAuC;CAChE,OAAO,aAAa,IAAI,CAAC,CAAC,SAAS,OAAO,UAAU;EAClD,IAAI,MAAM,SAAS,GAAG,OAAO,CAAC;EAC9B,MAAM,CAAC,SAAS,kBAAkB,mBAAmB,cAAc;EACnE,IAAI,CAAC,WAAW,CAAC,oBAAoB,CAAC,mBAAmB,OAAO,CAAC;EACjE,IAAI,QAAQ,YAAY,MAAM,cAAc,QAAQ,YAAY,MAAM,OAAO,OAAO,CAAC;EACrF,MAAM,aAAa,eAAe,KAAA,IAAY,mBAAmB;EAEjE,MAAM,QAAQ,OADI,eAAe,KAAA,IAAY,oBAAoB,UACnC;EAC9B,IAAI,CAAC,cAAc,CAAC,OAAO,SAAS,KAAK,GACvC,MAAM,IAAI,MAAM,qBAAqB,QAAQ,EAAE,mCAAmC;EAEpF,OAAO,CAAC;GAAE;GAAS;GAAY;EAAM,CAAC;CACxC,CAAC;AACH;AAEA,SAAgB,qBAAqB,MAA2C;CAC9E,OAAO,eAAwC,IAAI,CAAC,CAAC,KAAK,KAAK,UAAU;EACvE,MAAM,KAAK,YAAY,KAAK,KAAK,KAAK,YAAY,KAAK,IAAI;EAC3D,MAAM,OAAO,YAAY,KAAK,MAAM,KAAK,YAAY,KAAK,UAAU;EACpE,IAAI,CAAC,MAAM,SAAS,KAAA,GAClB,MAAM,IAAI,MAAM,2BAA2B,QAAQ,EAAE,oCAAoC;EAE3F,OAAO;GACL;GACA,OAAO,YAAY,KAAK,OAAO;GAC/B,MAAM;GACN,UAAU,WAAW,KAAK;IAAC;IAAO;IAAM;IAAS;IAAQ;GAAU,CAAC;EACtE;CACF,CAAC;AACH;AAEA,SAAgB,sBAAsB,MAAwC;CAC5E,OAAO,eAAwC,IAAI,CAAC,CAAC,KAAK,KAAK,UAAU;EACvE,MAAM,KAAK,YAAY,KAAK,KAAK,KAAK,YAAY,KAAK,IAAI,KAAK,YAAY,KAAK,UAAU;EAC3F,MAAM,QAAQ,YAAY,KAAK,MAAM,KAAK,YAAY,KAAK,OAAO;EAClE,IAAI,CAAC,MAAM,CAAC,OACV,MAAM,IAAI,MAAM,0BAA0B,QAAQ,EAAE,iCAAiC;EAEvF,OAAO;GACL;GACA,MAAM;GACN,UAAU,WAAW,KAAK;IAAC;IAAO;IAAM;IAAY;IAAQ;GAAO,CAAC;EACtE;CACF,CAAC;AACH;AAEA,SAAgB,4BACd,SACuD;CACvD,MAAM,+BAAe,IAAI,IAAqC;CAC9D,KAAK,MAAM,QAAQ,QAAQ,OAAO;EAChC,IAAI,KAAK,SAAS,GAAG;EACrB,MAAM,OAAO,aAAa,IAAI,KAAK,OAAO,KAAK,CAAC;EAChD,KAAK,KAAK,IAAI;EACd,aAAa,IAAI,KAAK,SAAS,IAAI;CACrC;CACA,MAAM,SACJ,QAAQ,0BAA0B,QAAQ,SACtC,OAAO,YAAY,QAAQ,OAAO,KAAK,aAAa,CAAC,SAAS,IAAI,QAAQ,CAAC,CAAC,IAC5E,KAAA;CACN,OAAO,QAAQ,QAAQ,SAAS,UAAU;EACxC,MAAM,QAAQ,aAAa,IAAI,MAAM,EAAE,KAAK,CAAC;EAC7C,IAAI,MAAM,WAAW,GAAG,OAAO,CAAC;EAChC,MAAM,QACJ,QAAQ,UAAU,MAAM,EAAE,KAAK,mBAAmB,GAAG,QAAQ,YAAY,GAAG,MAAM,IAAI;EACxF,OAAO,CACL;GACE,IAAI,MAAM;GACV;GACA,QAAQ,QAAQ;GAChB,UAAU;GACV,MAAM,CAAC,mBAAG,IAAI,IAAI,CAAC,GAAI,QAAQ,QAAQ,CAAC,GAAI,KAAK,CAAC,CAAC;GACnD,GAAI,QAAQ,SAAS,EAAE,QAAQ,QAAQ,OAAO,IAAI,CAAC;GACnD,GAAI,MAAM,WAAW,EAAE,UAAU,MAAM,SAAS,IAAI,CAAC;GACrD,SAAS;IACP,SAAS,MAAM;IACf,OAAO,MAAM;IACb,qBAAqB,MAAM,KAAK,SAAS,KAAK,UAAU;IACxD,gBAAgB,OAAO,YAAY,MAAM,KAAK,SAAS,CAAC,KAAK,YAAY,KAAK,KAAK,CAAC,CAAC;IACrF,GAAI,SAAS,EAAE,OAAO,IAAI,CAAC;IAC3B,GAAI,MAAM,WAAW,EAAE,UAAU,MAAM,SAAS,IAAI,CAAC;GACvD;EACF,CACF;CACF,CAAC;AACH;AAEA,SAAgB,kCACd,SAKA;CACA,MAAM,QAAQ,4BAA4B,OAAO;CACjD,OAAO;EACL,IAAI,QAAQ;EACZ,QAAQ,QAAQ;EAChB,UAAU;EACV,QAAQ,QAAQ;EAChB,eAAe,QAAQ,iBAAiB;EACxC,MAAM,YAAY,OAAO;GACvB,OAAO,MAAM,QAAQ,SAAS,KAAK,UAAU,KAAK;EACpD;EACA,MAAM,SAAS,MAAM,UAAU;GAC7B,OAAO,0BAA0B,KAAK,SAAS,UAAU,OAAO;EAClE;EACA,YAAY,QAAQ;GAClB,OAAO,QAAQ,UAAU,MAAM,KAAK,mBAAmB,GAAG,QAAQ,YAAY,GAAG,QAAQ;EAC3F;CACF;AACF;AAEA,SAAgB,0BACd,SACA,UACA,UAA8C,CAAC,GAC/C;CACA,MAAM,oBAAoB,8BACxB,UACA,QAAQ,qBAAqB,oBAC/B;CACA,MAAM,UAAU,iBAAiB,QAAQ,WAAW;EAAC;EAAG;EAAG;EAAG;CAAE,CAAC;CACjE,MAAM,aAAqC;EACzC,gBAAgB,QAAQ,oBAAoB;EAC5C,gBAAgB,kBAAkB;CACpC;CACA,KAAK,MAAM,UAAU,SACnB,OAAO,OACL,YACA,yBAAyB;EACvB;EACA,gBAAgB,QAAQ;EACxB;CACF,CAAC,CACH;CAEF,MAAM,gBAAgB,QAAQ,iBAAiB;CAC/C,MAAM,aAAa,QAAQ,cAAc;CACzC,MAAM,QAAQ,WAAW,kBAAkB,WAAW,QAAQ,QAAQ,QAAQ,SAAS,SAAS;CAGhG,OAAO;EACL;EACA,SAHA,WAAW,eAAe,WAAW,OAAO,QAAQ,QAAQ,SAAS,SAAS,WAGxD,QAAQ,iBAAiB;EAC/C;EACA,KAAK;GACH;GACA,qBAAqB,QAAQ;GAC7B,gBAAgB,QAAQ;EAC1B;CACF;AACF;AAEA,SAAgB,8BACd,UACA,oBAA4B,sBAClB;CACV,IAAI,OAAO,aAAa,UACtB,OAAO,eAAe,SAAS,MAAM,iBAAiB,KAAK,CAAC,EAAA,CAAG,KAAK,UAAU,MAAM,KAAK,CAAC,CAAC;CAE7F,IAAI,MAAM,QAAQ,QAAQ,GAExB,OAAO,cACLC,SAAQ,SAAS,UAAU;EACzB,IAAI,OAAO,UAAU,UAAU,OAAO,CAAC,KAAK;EAC5C,OAAO,CAAC,MAAM,cAAc,MAAM,SAAS,MAAM,MAAM,EAAE;CAC3D,CAAC,CACH;CAEF,MAAM,iBAAiB;CACvB,OAAO,cACL;EACE,GAAI,eAAe,eAAe,CAAC;EACnC,GAAI,eAAe,OAAO,CAAC;EAC3B,IAAI,eAAe,WAAW,CAAC,EAAA,CAAG,KAC/B,UAAU,MAAM,cAAc,MAAM,SAAS,MAAM,MAAM,EAC5D;CACF,CAAC,CAAC,KAAK,UAAU,MAAM,KAAK,CAAC,CAC/B;AACF;AAEA,SAAgB,yBAAyB,OAId;CACzB,MAAM,MAAM,MAAM,kBAAkB,MAAM,GAAG,MAAM,MAAM;CACzD,MAAM,cAAc,OAAO,KAAK,MAAM,cAAc,CAAC,CAAC,QACnD,OAAO,MAAM,eAAe,MAAO,CACtC;CACA,MAAM,WAAW,IAAI,IAAI,WAAW;CACpC,MAAM,OAAO,IAAI,QAAQ,OAAO,SAAS,IAAI,EAAE,CAAC,CAAC,CAAC;CAClD,MAAM,oBAAoB,IAAI,WAAW,OAAO,SAAS,IAAI,EAAE,CAAC;CAChE,MAAM,SAAS,IAAI,MAAM;CACzB,OAAO;GACJ,MAAM,WAAW,OAAO,IAAI,IAAI;GAChC,SAAS,WAAW,SAAS,SAAS,IAAI,IAAI,OAAO,SAAS;GAC9D,YAAY,WAAW,OAAO,MAAM;GACpC,MAAM,WAAW,sBAAsB,KAAK,IAAI,KAAK,oBAAoB;GACzE,OAAO,WAAW,OAAO,MAAM,mBAAmB,MAAM,gBAAgB,MAAM,MAAM;CACvF;AACF;AAEA,SAAS,YAAY,KAA8B,KAAiC;CAClF,MAAM,QAAQ,IAAI;CAClB,OAAO,OAAO,UAAU,WAAW,QAAQ,KAAA;AAC7C;AAEA,SAAS,WACP,KACA,MACyB;CACzB,MAAM,QAAQ,IAAI,IAAI,IAAI;CAC1B,OAAO,OAAO,YAAY,OAAO,QAAQ,GAAG,CAAC,CAAC,QAAQ,CAAC,SAAS,CAAC,MAAM,IAAI,GAAG,CAAC,CAAC;AAClF;AAEA,SAAS,iBAAiB,SAAsC;CAC9D,OAAO,CACL,GAAG,IAAI,IAAI,QAAQ,KAAK,WAAW,KAAK,MAAM,MAAM,CAAC,CAAC,CAAC,QAAQ,WAAW,SAAS,CAAC,CAAC,CACvF,CAAC,CAAC,MAAM,GAAG,MAAM,IAAI,CAAC;AACxB;AAEA,SAAS,cAAc,QAAqC;CAC1D,MAAM,uBAAO,IAAI,IAAY;CAC7B,MAAM,MAAgB,CAAC;CACvB,KAAK,MAAM,SAAS,QAAQ;EAC1B,MAAM,UAAU,MAAM,KAAK;EAC3B,IAAI,CAAC,WAAW,KAAK,IAAI,OAAO,GAAG;EACnC,KAAK,IAAI,OAAO;EAChB,IAAI,KAAK,OAAO;CAClB;CACA,OAAO;AACT;AAEA,SAAS,OACP,mBACA,gBACA,QACQ;CACR,MAAM,MAAM,kBACT,MAAM,GAAG,MAAM,CAAC,CAChB,QACE,KAAK,YAAY,UAAU,MAAM,eAAe,eAAe,eAAe,GAAG,KAAK,GACvF,CACF;CACF,MAAM,QAAQ,OAAO,OAAO,cAAc,CAAC,CACxC,QAAQ,UAAU,QAAQ,CAAC,CAAC,CAC5B,MAAM,GAAG,MAAM,IAAI,CAAC,CAAC,CACrB,MAAM,GAAG,MAAM,CAAC,CAChB,QAAQ,KAAK,OAAO,UAAU,MAAM,eAAe,OAAO,KAAK,GAAG,CAAC;CACtE,OAAO,UAAU,IAAI,IAAI,MAAM;AACjC;AAEA,SAAS,eAAe,WAAmB,eAA+B;CACxE,IAAI,aAAa,GAAG,OAAO;CAC3B,QAAQ,KAAK,YAAY,KAAK,KAAK,KAAK,gBAAgB,CAAC;AAC3D"}

package/dist/builder-eval/index.d.ts

@@ -1,7 +1,28 @@
-import { f as Run } from "../schema-BtVldJ3T.js";
+import { f as Run, o as FailureClass } from "../schema-BtVldJ3T.js";
 import { s as TraceStore } from "../store-CT9YIIve.js";
 import { i as TraceEmitter } from "../emitter-DGQGoLyj.js";
-import { l as SandboxHarnessResult, n as TestGradedRunResult, o as HarnessConfig, r as TestGradedScenario, s as SandboxDriver } from "../test-graded-scenario-D1TaI2va.js";
+import { n as SandboxDriver, r as SandboxHarnessResult, t as HarnessConfig } from "../sandbox-harness-BlSOu4LX.js";
+//#region src/test-graded-scenario.d.ts
+interface TestGradedScenario {
+  id: string;
+  description?: string;
+  harness: HarnessConfig;
+  /** Optional pass threshold in 0..1 (default 1.0 = all tests must pass). */
+  passThreshold?: number;
+  /** Provenance for dataset tracking. */
+  datasetVersion?: string;
+  /** Free-form tags (difficulty, category, etc.). */
+  tags?: Record<string, string>;
+}
+interface TestGradedRunResult {
+  runId: string;
+  scenario: TestGradedScenario;
+  harness: SandboxHarnessResult;
+  pass: boolean;
+  score: number;
+  failureClass?: FailureClass;
+}
+//#endregion
 //#region src/builder-eval/builder-session.d.ts
 interface BuilderSessionInit {
   projectId: string;

package/dist/builder-eval/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/builder-eval/builder-session.ts","../../src/builder-eval/three-layer-eval.ts","../../src/builder-eval/correlation.ts","../../src/builder-eval/project-registry.ts"],"mappings":";;;;;UA0BiB;EACf;EACA;;EAEA,OAAO;;UAGQ;EACf,SAAS;EACT,SAAS;;EAET;;UAGe;EACf,UAAU;;EAEV,SAAS;;cAGE;UACH;UACA;WACC;WACA;UACD;UACA;UACA;EAER,YAAY,OAAO,YAAY,MAAM,oBAAoB,SAAS;;EAS5D,UAAU,sBAAwC;;MAYpD,WAAW;;;;;EAST,KAAK,SAAS,cAAc;IAAU;IAAe,QAAQ;;;;;;;;EA2B7D,eAAe,SAAS,wBAAwB,QAAQ;;;EAqBxD,gBAAgB,eAAe,qBAAqB;;EAcpD,QAAQ;IAAW;IAAe;IAAgB;MAAmB;;;;;;;;EAerE,gBAAgB,qBAAqB,QAAQ;;;;;;EAsB7C,iBAAiB;IACrB;IACA;IACA;IACA;MACE;MAqBA;MAGA;;;;;;;iBAUgB,qBACpB,OAAO,YACP,oBACC;EACD;EACA,UAAU;EACV,iBAAiB;EACjB,eAAe;EACf,oBAAoB;;;;KC3MV;UAEK;EACf;;;;;;;EAOA,MAAM;EACN;;EAEA;EACA;;EAEA;EACA;;EAEA;EACA;;;;;;EAMA;;iBAGoB,aACpB,OAAO,YACP,oBACC,QAAQ;;iBAuCW,iBAAiB,OAAO,aAAa,QAAQ;;;UC/ElD;EACf;EACA;EACA;;UAGe;;EAEf,cAAc;EACd,gBAAgB;EAChB,iBAAiB;;EAEjB;;iBAGc,gBAAgB,SAAS,4BAA4B;;;UCpBpD;EACf;EACA;EACA;EACA;EACA;EACA;EACA;IAAkB;IAAe;;;UAGlB;EACf;EACA;EACA;EACA;EACA;EACA,QAAQ;EACR,UAAU;;EAEV;EACA;EACA;EACA;;UAGe;EACf,KAAK;EACL;;cAGW;UACS;EAApB,YAAoB,OAAO;EAErB,gBAAgB,QAAQ;EAgCxB,gBAAgB,oBAAoB,QAAQ;EAgB5C,aAAa,oBAAoB,QAAQ"}
+{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/test-graded-scenario.ts","../../src/builder-eval/builder-session.ts","../../src/builder-eval/three-layer-eval.ts","../../src/builder-eval/correlation.ts","../../src/builder-eval/project-registry.ts"],"mappings":";;;;;UAmBiB;EACf;EACA;EACA,SAAS;;EAET;;EAEA;;EAEA,OAAO;;UAUQ;EACf;EACA,UAAU;EACV,SAAS;EACT;EACA;EACA,eAAe;;;;UClBA;EACf;EACA;;EAEA,OAAO;;UAGQ;EACf,SAAS;EACT,SAAS;;EAET;;UAGe;EACf,UAAU;;EAEV,SAAS;;cAGE;UACH;UACA;WACC;WACA;UACD;UACA;UACA;EAER,YAAY,OAAO,YAAY,MAAM,oBAAoB,SAAS;;EAS5D,UAAU,sBAAwC;;MAYpD,WAAW;;;;;EAST,KAAK,SAAS,cAAc;IAAU;IAAe,QAAQ;;;;;;;;EA2B7D,eAAe,SAAS,wBAAwB,QAAQ;;;EAqBxD,gBAAgB,eAAe,qBAAqB;;EAcpD,QAAQ;IAAW;IAAe;IAAgB;MAAmB;;;;;;;;EAerE,gBAAgB,qBAAqB,QAAQ;;;;;;EAsB7C,iBAAiB;IACrB;IACA;IACA;IACA;MACE;MAqBA;MAGA;;;;;;;iBAUgB,qBACpB,OAAO,YACP,oBACC;EACD;EACA,UAAU;EACV,iBAAiB;EACjB,eAAe;EACf,oBAAoB;;;;KC3MV;UAEK;EACf;;;;;;;EAOA,MAAM;EACN;;EAEA;EACA;;EAEA;EACA;;EAEA;EACA;;;;;;EAMA;;iBAGoB,aACpB,OAAO,YACP,oBACC,QAAQ;;iBAuCW,iBAAiB,OAAO,aAAa,QAAQ;;;UC/ElD;EACf;EACA;EACA;;UAGe;;EAEf,cAAc;EACd,gBAAgB;EAChB,iBAAiB;;EAEjB;;iBAGc,gBAAgB,SAAS,4BAA4B;;;UCpBpD;EACf;EACA;EACA;EACA;EACA;EACA;EACA;IAAkB;IAAe;;;UAGlB;EACf;EACA;EACA;EACA;EACA;EACA,QAAQ;EACR,UAAU;;EAEV;EACA;EACA;EACA;;UAGe;EACf,KAAK;EACL;;cAGW;UACS;EAApB,YAAoB,OAAO;EAErB,gBAAgB,QAAQ;EAgCxB,gBAAgB,oBAAoB,QAAQ;EAgB5C,aAAa,oBAAoB,QAAQ"}

package/dist/builder-eval/index.js

@@ -1,7 +1,231 @@
-import { L as pearsonR, V as spearmanR } from "../statistics-ByxzSiOM.js";
-import { t as TraceEmitter } from "../emitter-CPBAhxum.js";
+import { a as spearmanR, r as pearsonR } from "../descriptive-B5MwKfbf.js";
 import { a as judgeSpans } from "../query-Di7eEQ79.js";
-import { r as SandboxHarness, t as runTestGradedScenario } from "../test-graded-scenario-JHcKQNpq.js";
+import { t as TraceEmitter } from "../emitter-CPBAhxum.js";
+//#region src/sandbox-harness.ts
+var SubprocessSandboxDriver = class {
+	id = "subprocess";
+	defaultCwd;
+	defaultEnv;
+	constructor(options = {}) {
+		this.defaultCwd = options.cwd;
+		this.defaultEnv = options.env;
+	}
+	async exec(phase, command, config) {
+		const { spawn } = await import("node:child_process");
+		const start = Date.now();
+		const effectiveCwd = config.cwd ?? this.defaultCwd;
+		const effectiveEnv = {
+			...process.env,
+			...this.defaultEnv ?? {},
+			...config.env ?? {}
+		};
+		const maxOutputBytes = config.maxOutputBytes ?? 16 * 1024 * 1024;
+		return await new Promise((resolve) => {
+			const child = spawn(command, {
+				shell: true,
+				cwd: effectiveCwd,
+				env: effectiveEnv,
+				detached: process.platform !== "win32"
+			});
+			let stdout = "";
+			let stderr = "";
+			let outputBytes = 0;
+			let outputTruncated = false;
+			let killedByTimeout = false;
+			let settled = false;
+			const onStdout = (d) => {
+				const chunk = capture(String(d));
+				if (chunk) stdout += chunk;
+			};
+			const onStderr = (d) => {
+				const chunk = capture(String(d));
+				if (chunk) stderr += chunk;
+			};
+			function capture(s) {
+				if (outputBytes >= maxOutputBytes) {
+					outputTruncated = true;
+					return "";
+				}
+				const room = maxOutputBytes - outputBytes;
+				if (s.length > room) {
+					outputBytes = maxOutputBytes;
+					outputTruncated = true;
+					return s.slice(0, room);
+				}
+				outputBytes += s.length;
+				return s;
+			}
+			child.stdout?.on("data", onStdout);
+			child.stderr?.on("data", onStderr);
+			const cleanup = () => {
+				clearTimeout(timeout);
+				if (forceResolve) clearTimeout(forceResolve);
+				child.stdout?.off("data", onStdout);
+				child.stderr?.off("data", onStderr);
+				child.removeAllListeners("close");
+				child.removeAllListeners("error");
+			};
+			const finish = (outcome) => {
+				if (settled) return;
+				settled = true;
+				cleanup();
+				const wallMs = Date.now() - start;
+				const exitCode = killedByTimeout ? outcome.code && outcome.code !== 0 ? outcome.code : 124 : outcome.code ?? 1;
+				const parsed = !killedByTimeout && phase === "test" && config.testParser ? config.testParser.parse(stdout, stderr, exitCode) : void 0;
+				resolve({
+					phase,
+					exitCode,
+					stdout,
+					stderr: outcome.runnerError ? stderr + outcome.runnerError : stderr,
+					wallMs,
+					testsTotal: parsed?.testsTotal,
+					testsPassed: parsed?.testsPassed,
+					killedByTimeout: killedByTimeout || void 0,
+					outputTruncated: outputTruncated || void 0
+				});
+			};
+			const killTree = () => {
+				try {
+					if (process.platform !== "win32" && typeof child.pid === "number") process.kill(-child.pid, "SIGKILL");
+					else child.kill("SIGKILL");
+				} catch (err) {
+					console.warn("[sandbox-harness] SIGKILL on timeout failed:", err);
+				}
+			};
+			let forceResolve;
+			const timeout = setTimeout(() => {
+				killedByTimeout = true;
+				killTree();
+				forceResolve = setTimeout(() => finish({ code: null }), 2e3);
+			}, config.timeoutMs ?? 10 * 6e4);
+			child.on("close", (code) => {
+				if (forceResolve) clearTimeout(forceResolve);
+				finish({ code });
+			});
+			child.on("error", (err) => {
+				if (forceResolve) clearTimeout(forceResolve);
+				finish({
+					code: 127,
+					runnerError: String(err)
+				});
+			});
+		});
+	}
+};
+var SandboxHarness = class {
+	driver;
+	constructor(driver = new SubprocessSandboxDriver()) {
+		this.driver = driver;
+	}
+	async run(config, emitter) {
+		const handle = await emitter.sandbox({
+			name: `sandbox(${this.driver.id})`,
+			image: config.image,
+			command: [
+				config.setupCommand,
+				config.runCommand,
+				config.testCommand
+			].filter(Boolean).join(" && ")
+		});
+		const result = {
+			passed: false,
+			totalWallMs: 0,
+			score: 0
+		};
+		try {
+			if (config.setupCommand) {
+				result.setup = await this.driver.exec("setup", config.setupCommand, config);
+				result.totalWallMs += result.setup.wallMs;
+				if (result.setup.killedByTimeout || result.setup.exitCode !== 0) {
+					const reason = result.setup.killedByTimeout ? `setup timed out after ${result.setup.wallMs}ms` : `setup failed (exit ${result.setup.exitCode})`;
+					await handle.fail(reason, {
+						exitCode: result.setup.exitCode,
+						wallMs: result.totalWallMs
+					});
+					return result;
+				}
+			}
+			if (config.runCommand) {
+				result.run = await this.driver.exec("run", config.runCommand, config);
+				result.totalWallMs += result.run.wallMs;
+				if (result.run.killedByTimeout || result.run.exitCode !== 0) {
+					const reason = result.run.killedByTimeout ? `run timed out after ${result.run.wallMs}ms` : `run failed (exit ${result.run.exitCode})`;
+					await handle.fail(reason, {
+						exitCode: result.run.exitCode,
+						wallMs: result.totalWallMs
+					});
+					return result;
+				}
+			}
+			if (config.testCommand) {
+				result.test = await this.driver.exec("test", config.testCommand, config);
+				result.totalWallMs += result.test.wallMs;
+				const passed = !result.test.killedByTimeout && result.test.exitCode === 0;
+				result.passed = passed;
+				if (result.test.testsTotal !== void 0 && result.test.testsTotal > 0) result.score = (result.test.testsPassed ?? 0) / result.test.testsTotal;
+				else result.score = passed ? 1 : 0;
+				await handle.end({
+					exitCode: result.test.exitCode,
+					testsTotal: result.test.testsTotal,
+					testsPassed: result.test.testsPassed,
+					wallMs: result.totalWallMs,
+					status: passed ? "ok" : "error"
+				});
+			} else {
+				result.passed = true;
+				result.score = 1;
+				await handle.end({ wallMs: result.totalWallMs });
+			}
+		} catch (err) {
+			await handle.fail(err instanceof Error ? err : String(err));
+			throw err;
+		}
+		return result;
+	}
+};
+//#endregion
+//#region src/test-graded-scenario.ts
+async function runTestGradedScenario(scenario, store, options = {}) {
+	const emitter = new TraceEmitter(store);
+	await emitter.startRun({
+		scenarioId: scenario.id,
+		variantId: options.variantId,
+		datasetVersion: scenario.datasetVersion,
+		tags: scenario.tags,
+		...options.provenance
+	});
+	const result = await new SandboxHarness(options.driver).run(scenario.harness, emitter);
+	const threshold = scenario.passThreshold ?? 1;
+	const pass = result.passed && result.score >= threshold;
+	const setupFailed = result.setup !== void 0 && result.setup.exitCode !== 0;
+	const runFailed = result.run !== void 0 && result.run.exitCode !== 0;
+	const testFailed = result.test !== void 0 && result.test.exitCode !== 0;
+	const failureClass = pass ? "success" : setupFailed || runFailed ? "sandbox_failure" : testFailed ? "format_drift" : "unknown";
+	await emitter.endRun({
+		pass,
+		score: result.score,
+		failureClass,
+		notes: pass ? void 0 : reasonForFailure(result)
+	});
+	return {
+		runId: emitter.runId,
+		scenario,
+		harness: result,
+		pass,
+		score: result.score,
+		failureClass
+	};
+}
+function reasonForFailure(result) {
+	if (result.setup && result.setup.exitCode !== 0) return `setup failed: exit ${result.setup.exitCode}`;
+	if (result.run && result.run.exitCode !== 0) return `run failed: exit ${result.run.exitCode}`;
+	if (result.test) {
+		if (result.test.testsTotal !== void 0) return `tests: ${result.test.testsPassed ?? 0}/${result.test.testsTotal}`;
+		return `test exit ${result.test.exitCode}`;
+	}
+	return "no test command";
+}
+//#endregion
 //#region src/builder-eval/builder-session.ts
 var BuilderSession = class {
 	store;

package/dist/builder-eval/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","names":[],"sources":["../../src/builder-eval/builder-session.ts","../../src/builder-eval/correlation.ts","../../src/builder-eval/project-registry.ts","../../src/builder-eval/three-layer-eval.ts"],"sourcesContent":["/**\n * BuilderSession — ties a builder-of-builders workflow together.\n *\n * Models agent-builder's shape: Project → Chat → Edit → Ship → App →\n * AppAgent. Each layer is a Run (linked via parentRunId). The\n * framework-enforced invariants:\n *\n *   - One Project → many Chats; chatId scopes runs within a project.\n *   - One Chat = one builder Run with `layer='builder'`.\n *   - One Ship = one child Run with `layer='app-build'` + SandboxHarness.\n *   - One AppScenario = one grandchild Run with `layer='app-runtime'`.\n *\n * Consumers obtain a BuilderSession, call `startChat`, drive the\n * builder agent (emitting spans), and call `ship` / `runAppScenario`\n * as the workflow progresses. The session reconstructs itself from\n * trace data via `resume(store, projectId)`.\n */\n\nimport type { HarnessConfig, SandboxDriver, SandboxHarnessResult } from '../sandbox-harness'\nimport { SandboxHarness } from '../sandbox-harness'\nimport type { TestGradedRunResult, TestGradedScenario } from '../test-graded-scenario'\nimport { runTestGradedScenario } from '../test-graded-scenario'\nimport { TraceEmitter } from '../trace/emitter'\nimport type { Run } from '../trace/schema'\nimport type { TraceStore } from '../trace/store'\n\nexport interface BuilderSessionInit {\n  projectId: string\n  chatId?: string\n  /** Free-form: user's task description, project name, etc. Stored on the builder Run. */\n  tags?: Record<string, string>\n}\n\nexport interface ShipOptions {\n  harness: HarnessConfig\n  driver?: SandboxDriver\n  /** scenarioId of this app-build run. Defaults to `${projectId}/build`. */\n  scenarioId?: string\n}\n\nexport interface RunAppScenarioOptions {\n  scenario: TestGradedScenario\n  /** Harness driver override; defaults to the one the session was created with. */\n  driver?: SandboxDriver\n}\n\nexport class BuilderSession {\n  private store: TraceStore\n  private builderEmitter: TraceEmitter\n  readonly projectId: string\n  readonly chatId: string\n  private builderRunId?: string\n  private lastBuildRunId?: string\n  private defaultDriver?: SandboxDriver\n\n  constructor(store: TraceStore, init: BuilderSessionInit, driver?: SandboxDriver) {\n    this.store = store\n    this.projectId = init.projectId\n    this.chatId = init.chatId ?? cryptoId()\n    this.defaultDriver = driver\n    this.builderEmitter = new TraceEmitter(store)\n  }\n\n  /** Start the builder (L0) run for this chat. Returns the runId. */\n  async startChat(scenarioId = `${this.projectId}/chat`): Promise<string> {\n    await this.builderEmitter.startRun({\n      scenarioId,\n      projectId: this.projectId,\n      chatId: this.chatId,\n      layer: 'builder',\n    })\n    this.builderRunId = this.builderEmitter.runId\n    return this.builderRunId\n  }\n\n  /** The emitter for builder-level spans (edits, LLM calls, tool invocations). */\n  get emitter(): TraceEmitter {\n    if (!this.builderRunId) throw new Error('BuilderSession.emitter: call startChat() first')\n    return this.builderEmitter\n  }\n\n  /**\n   * Ship the project's generated app: run the sandbox harness as a child\n   * Run (`layer='app-build'`). Returns the build result + runId.\n   */\n  async ship(options: ShipOptions): Promise<{ runId: string; result: SandboxHarnessResult }> {\n    if (!this.builderRunId) throw new Error('BuilderSession.ship: call startChat() first')\n    const buildEmitter = new TraceEmitter(this.store)\n    await buildEmitter.startRun({\n      scenarioId: options.scenarioId ?? `${this.projectId}/build`,\n      projectId: this.projectId,\n      chatId: this.chatId,\n      parentRunId: this.builderRunId,\n      layer: 'app-build',\n    })\n    const harness = new SandboxHarness(options.driver ?? this.defaultDriver)\n    const result = await harness.run(options.harness, buildEmitter)\n    await buildEmitter.endRun({\n      pass: result.passed,\n      score: result.score,\n      failureClass: result.passed ? 'success' : 'sandbox_failure',\n    })\n    this.lastBuildRunId = buildEmitter.runId\n    return { runId: buildEmitter.runId, result }\n  }\n\n  /**\n   * Run a domain scenario against the just-built app as a grandchild Run\n   * (`layer='app-runtime'`). The `ship` call must precede this so the\n   * parent is set correctly; if no build exists yet the session attaches\n   * directly to the builder run (useful for prototypes).\n   */\n  async runAppScenario(options: RunAppScenarioOptions): Promise<TestGradedRunResult> {\n    const parentRunId = this.lastBuildRunId ?? this.builderRunId\n    if (!parentRunId)\n      throw new Error('BuilderSession.runAppScenario: call startChat() + ship() first')\n    const { scenario, driver } = options\n    const result = await runTestGradedScenario(scenario, this.store, {\n      driver: driver ?? this.defaultDriver,\n      provenance: { codeSha: undefined, promptSha: undefined, modelFingerprint: undefined },\n    })\n    // Attach to the parent chain by updating the stored Run in place.\n    await this.store.updateRun(result.runId, {\n      parentRunId,\n      projectId: this.projectId,\n      chatId: this.chatId,\n      layer: 'app-runtime',\n    })\n    return result\n  }\n\n  /** Record an end-of-chat meta score (judge verdict on whether the builder\n   *  served the user's intent). Accepts a numeric score + optional rationale. */\n  async recordMetaScore(score: number, rationale?: string): Promise<void> {\n    if (!this.builderRunId)\n      throw new Error('BuilderSession.recordMetaScore: call startChat() first')\n    await this.builderEmitter.recordJudge({\n      judgeId: 'builder-meta',\n      targetSpanId: this.builderRunId, // attach to the builder run itself\n      dimension: 'user_intent_satisfaction',\n      score,\n      rationale,\n      name: 'builder-meta',\n    })\n  }\n\n  /** Close the builder Run with a final outcome. */\n  async endChat(outcome: { pass: boolean; score?: number; notes?: string }): Promise<void> {\n    await this.builderEmitter.endRun({\n      pass: outcome.pass,\n      score: outcome.score,\n      notes: outcome.notes,\n    })\n  }\n\n  /**\n   * Inline app-runtime run — for cases where the \"scenario\" isn't a\n   * SWE-bench-style test suite but a live agent interaction (LLM chat,\n   * domain flow). Returns an emitter bound to a fresh Run in the\n   * `app-runtime` layer; caller emits spans inside and calls\n   * `.endRun()` with the final verdict.\n   */\n  async startAppRuntime(scenarioId: string): Promise<TraceEmitter> {\n    const parentRunId = this.lastBuildRunId ?? this.builderRunId\n    if (!parentRunId)\n      throw new Error(\n        'BuilderSession.startAppRuntime: call startChat() + (optionally) ship() first',\n      )\n    const emitter = new TraceEmitter(this.store)\n    await emitter.startRun({\n      scenarioId,\n      projectId: this.projectId,\n      chatId: this.chatId,\n      parentRunId,\n      layer: 'app-runtime',\n    })\n    return emitter\n  }\n\n  /**\n   * Lightweight \"ship marker\" — record an app-build Run with a caller-\n   * provided verdict. Use when there isn't a sandbox harness to run but\n   * you still want to mark the build state at publish time.\n   */\n  async recordShipMarker(args: {\n    pass: boolean\n    score: number\n    scenarioId?: string\n    notes?: string\n  }): Promise<string> {\n    if (!this.builderRunId)\n      throw new Error('BuilderSession.recordShipMarker: call startChat() first')\n    const emitter = new TraceEmitter(this.store)\n    await emitter.startRun({\n      scenarioId: args.scenarioId ?? `${this.projectId}/ship`,\n      projectId: this.projectId,\n      chatId: this.chatId,\n      parentRunId: this.builderRunId,\n      layer: 'app-build',\n    })\n    await emitter.endRun({\n      pass: args.pass,\n      score: args.score,\n      failureClass: args.pass ? 'success' : 'sandbox_failure',\n      notes: args.notes,\n    })\n    this.lastBuildRunId = emitter.runId\n    return emitter.runId\n  }\n\n  get lastBuildRunIdValue(): string | undefined {\n    return this.lastBuildRunId\n  }\n  get builderRunIdValue(): string | undefined {\n    return this.builderRunId\n  }\n}\n\n/**\n * Reconstruct the most recent BuilderSession state for a given project —\n * returns { builderRunId, lastBuildRunId, chatRuns }. For chat-first UIs\n * this is how a resumed session finds its place in the edit history.\n */\nexport async function resumeBuilderSession(\n  store: TraceStore,\n  projectId: string,\n): Promise<{\n  projectId: string\n  chatRuns: Run[]\n  lastBuilderRun?: Run\n  lastBuildRun?: Run\n  lastAppRuntimeRuns: Run[]\n}> {\n  const runs = await store.listRuns({ projectId })\n  const chatRuns = runs\n    .filter((r) => r.layer === 'builder')\n    .sort((a, b) => b.startedAt - a.startedAt)\n  const buildRuns = runs\n    .filter((r) => r.layer === 'app-build')\n    .sort((a, b) => b.startedAt - a.startedAt)\n  const appRuntimeRuns = runs\n    .filter((r) => r.layer === 'app-runtime')\n    .sort((a, b) => b.startedAt - a.startedAt)\n  return {\n    projectId,\n    chatRuns,\n    lastBuilderRun: chatRuns[0],\n    lastBuildRun: buildRuns[0],\n    lastAppRuntimeRuns: appRuntimeRuns,\n  }\n}\n\nfunction cryptoId(): string {\n  if (typeof globalThis.crypto?.randomUUID === 'function') return globalThis.crypto.randomUUID()\n  return `${Date.now().toString(36)}-${Math.random().toString(36).slice(2, 10)}`\n}\n","/**\n * Meta-eval correlation — the highest-leverage signal in the framework.\n *\n * Given a corpus of three-layer project reports, compute how well each\n * pair of layers correlates. The question we care about most:\n *\n *   Does `metaScore` (what the builder thinks it did) predict\n *   `runtimeScore` (what the user actually gets)?\n *\n * If r < ~0.4, the builder's self-scoring is broken — it's optimizing\n * for something other than real-world success. If r > 0.7, meta_score\n * is a usable proxy and can drive CI gates cheaply.\n *\n * Non-parametric rank correlation (Spearman) is also reported because\n * meta scores are often ordinal-ish.\n */\n\nimport { pearsonR, spearmanR } from '../statistics'\nimport type { ThreeLayerProjectReport } from './three-layer-eval'\n\nexport interface LayerCorrelation {\n  n: number\n  pearson: number\n  spearman: number\n}\n\nexport interface CorrelationReport {\n  /** Pairs present in the corpus (layers with ≥ 2 matched data points). */\n  metaVsBuild?: LayerCorrelation\n  metaVsRuntime?: LayerCorrelation\n  buildVsRuntime?: LayerCorrelation\n  /** Number of complete projects (all 3 scores present). */\n  completeProjects: number\n}\n\nexport function correlateLayers(reports: ThreeLayerProjectReport[]): CorrelationReport {\n  const completeProjects = reports.filter((r) => r.complete).length\n  return {\n    metaVsBuild: pairwise(\n      reports,\n      (r) => r.metaScore,\n      (r) => r.buildScore,\n    ),\n    metaVsRuntime: pairwise(\n      reports,\n      (r) => r.metaScore,\n      (r) => r.runtimeScore,\n    ),\n    buildVsRuntime: pairwise(\n      reports,\n      (r) => r.buildScore,\n      (r) => r.runtimeScore,\n    ),\n    completeProjects,\n  }\n}\n\nfunction pairwise(\n  reports: ThreeLayerProjectReport[],\n  a: (r: ThreeLayerProjectReport) => number | null,\n  b: (r: ThreeLayerProjectReport) => number | null,\n): LayerCorrelation | undefined {\n  const xs: number[] = []\n  const ys: number[] = []\n  for (const r of reports) {\n    const x = a(r)\n    const y = b(r)\n    if (x !== null && y !== null && Number.isFinite(x) && Number.isFinite(y)) {\n      xs.push(x)\n      ys.push(y)\n    }\n  }\n  if (xs.length < 2) return undefined\n  return {\n    n: xs.length,\n    pearson: pearsonR(xs, ys),\n    spearman: spearmanR(xs, ys),\n  }\n}\n","/**\n * ProjectRegistry — project-level aggregation over the trace corpus.\n *\n * Thin reader over TraceStore that answers the questions a chat-first,\n * resumable UI needs:\n *   - listProjects() → project IDs with latest activity\n *   - projectTimeline(id) → chats + builds + runtime runs, chronological\n *   - projectChats(id) → chat-level summaries (turn count, outcome)\n *\n * All queries are pure reads; no state duplication.\n */\n\nimport type { Run } from '../trace/schema'\nimport type { TraceStore } from '../trace/store'\n\nexport interface ProjectSummary {\n  projectId: string\n  chatCount: number\n  buildCount: number\n  appRuntimeCount: number\n  lastActivityAt: number\n  latestChatId?: string\n  latestOutcome?: { pass: boolean; score?: number }\n}\n\nexport interface ChatSummary {\n  chatId: string\n  projectId: string\n  builderRunId: string\n  startedAt: number\n  endedAt?: number\n  status: Run['status']\n  outcome?: Run['outcome']\n  /** Counts of spans emitted during the chat. */\n  llmTurns?: number\n  toolCalls?: number\n  buildRunId?: string\n  appRuntimeRunIds: string[]\n}\n\nexport interface ProjectTimelineEntry {\n  run: Run\n  layerBucket: 'chat' | 'build' | 'runtime' | 'other'\n}\n\nexport class ProjectRegistry {\n  constructor(private store: TraceStore) {}\n\n  async listProjects(): Promise<ProjectSummary[]> {\n    const runs = await this.store.listRuns()\n    const byProject = new Map<string, Run[]>()\n    for (const r of runs) {\n      if (!r.projectId) continue\n      const arr = byProject.get(r.projectId) ?? []\n      arr.push(r)\n      byProject.set(r.projectId, arr)\n    }\n    const summaries: ProjectSummary[] = []\n    for (const [projectId, projectRuns] of byProject) {\n      const sorted = projectRuns.slice().sort((a, b) => b.startedAt - a.startedAt)\n      const chats = projectRuns.filter((r) => r.layer === 'builder')\n      const builds = projectRuns.filter((r) => r.layer === 'app-build')\n      const runtimes = projectRuns.filter((r) => r.layer === 'app-runtime')\n      const latest = sorted[0]\n      if (!latest) continue\n      summaries.push({\n        projectId,\n        chatCount: chats.length,\n        buildCount: builds.length,\n        appRuntimeCount: runtimes.length,\n        lastActivityAt: latest.startedAt,\n        latestChatId: chats[0]?.chatId,\n        latestOutcome: latest.outcome\n          ? { pass: latest.outcome.pass ?? false, score: latest.outcome.score }\n          : undefined,\n      })\n    }\n    return summaries.sort((a, b) => b.lastActivityAt - a.lastActivityAt)\n  }\n\n  async projectTimeline(projectId: string): Promise<ProjectTimelineEntry[]> {\n    const runs = await this.store.listRuns({ projectId })\n    const ordered = runs.slice().sort((a, b) => a.startedAt - b.startedAt)\n    return ordered.map((run) => ({\n      run,\n      layerBucket:\n        run.layer === 'builder'\n          ? 'chat'\n          : run.layer === 'app-build'\n            ? 'build'\n            : run.layer === 'app-runtime'\n              ? 'runtime'\n              : 'other',\n    }))\n  }\n\n  async projectChats(projectId: string): Promise<ChatSummary[]> {\n    const builderRuns = (await this.store.listRuns({ projectId, layer: 'builder' })).sort(\n      (a, b) => b.startedAt - a.startedAt,\n    )\n    const childrenFor = async (runId: string) => this.store.listRuns({ parentRunId: runId })\n    const out: ChatSummary[] = []\n    for (const run of builderRuns) {\n      const spans = await this.store.spans({ runId: run.runId })\n      const children = await childrenFor(run.runId)\n      const build = children.find((c) => c.layer === 'app-build')\n      const runtime: string[] = []\n      // Runtime runs can be grandchildren (attached to the build) or siblings\n      // when shipped skipped.\n      if (build) {\n        const grands = await childrenFor(build.runId)\n        for (const g of grands) if (g.layer === 'app-runtime') runtime.push(g.runId)\n      }\n      for (const c of children) if (c.layer === 'app-runtime') runtime.push(c.runId)\n      out.push({\n        chatId: run.chatId ?? run.runId,\n        projectId,\n        builderRunId: run.runId,\n        startedAt: run.startedAt,\n        endedAt: run.endedAt,\n        status: run.status,\n        outcome: run.outcome,\n        llmTurns: spans.filter((s) => s.kind === 'llm').length,\n        toolCalls: spans.filter((s) => s.kind === 'tool').length,\n        buildRunId: build?.runId,\n        appRuntimeRunIds: runtime,\n      })\n    }\n    return out\n  }\n}\n","/**\n * Three-layer evaluation — the canonical scoring breakdown for\n * builder-of-builders workflows.\n *\n *   meta_score:    did the builder understand + satisfy user intent?\n *                  (judge verdict attached to the builder run)\n *   build_score:   did the generated scaffold build + pass its own tests?\n *                  (outcome.score on the app-build child run)\n *   runtime_score: did the generated agent pass its domain scenarios?\n *                  (mean outcome.score over app-runtime grandchild runs)\n *\n * Returns a structured report per project. The cross-layer correlation\n * is the highest-leverage signal the framework computes — if\n * meta_score doesn't predict runtime_score, the builder's self-scoring\n * is broken.\n *\n * Scaffold-only mode: when a project has no `app-runtime` runs (e.g. a\n * scaffold-builder eval that grades compose + build without driving a\n * runtime scenario), `kind` is `'scaffold-only'` and `complete` measures\n * meta + build only. Consumers can tell the two apart without having to\n * interpret null-runtime as either \"not yet computed\" or \"N/A for this\n * project shape\".\n */\n\nimport { judgeSpans } from '../trace/query'\nimport type { Run } from '../trace/schema'\nimport type { TraceStore } from '../trace/store'\n\nexport type ProjectKind = 'full' | 'scaffold-only'\n\nexport interface ThreeLayerProjectReport {\n  projectId: string\n  /**\n   * `'full'` when the project has at least one `app-runtime` run;\n   * `'scaffold-only'` when it only has meta + build layers. Lets\n   * downstream consumers treat a null runtime score as expected\n   * (scaffold-only) vs. missing (full, pipeline broke).\n   */\n  kind: ProjectKind\n  builderRunId?: string\n  /** Judge-verdict score on the builder run (0..1 after normalization). */\n  metaScore: number | null\n  buildRunId?: string\n  /** 0..1 from the sandbox harness (testsPassed / testsTotal). */\n  buildScore: number | null\n  appRuntimeRunIds: string[]\n  /** Mean of outcome.score over app-runtime runs, 0..1. Always null in scaffold-only mode. */\n  runtimeScore: number | null\n  runtimePassRate: number | null\n  /**\n   * Layer-aware completeness:\n   *   - `kind='full'`: all three layers scored\n   *   - `kind='scaffold-only'`: meta + build scored (runtime not applicable)\n   */\n  complete: boolean\n}\n\nexport async function scoreProject(\n  store: TraceStore,\n  projectId: string,\n): Promise<ThreeLayerProjectReport> {\n  const allRuns = await store.listRuns({ projectId })\n  const builder = latestByLayer(allRuns, 'builder')\n  const build = latestByLayer(allRuns, 'app-build')\n  const runtime = allRuns.filter((r) => r.layer === 'app-runtime')\n\n  const metaScore = builder ? await extractMetaScore(store, builder.runId) : null\n  const buildScore = build?.outcome?.score ?? null\n  const runtimeScores = runtime\n    .map((r) => r.outcome?.score)\n    .filter((s): s is number => typeof s === 'number')\n  const runtimeScore =\n    runtimeScores.length > 0\n      ? runtimeScores.reduce((a, b) => a + b, 0) / runtimeScores.length\n      : null\n  const runtimePassed = runtime.filter((r) => r.outcome?.pass === true).length\n  const runtimePassRate = runtime.length > 0 ? runtimePassed / runtime.length : null\n\n  const kind: ProjectKind = runtime.length === 0 ? 'scaffold-only' : 'full'\n  const complete =\n    kind === 'scaffold-only'\n      ? metaScore !== null && buildScore !== null\n      : metaScore !== null && buildScore !== null && runtimeScore !== null\n\n  return {\n    projectId,\n    kind,\n    builderRunId: builder?.runId,\n    metaScore,\n    buildRunId: build?.runId,\n    buildScore,\n    appRuntimeRunIds: runtime.map((r) => r.runId),\n    runtimeScore,\n    runtimePassRate,\n    complete,\n  }\n}\n\n/** Aggregate scoring across every project in a corpus. */\nexport async function scoreAllProjects(store: TraceStore): Promise<ThreeLayerProjectReport[]> {\n  const runs = await store.listRuns()\n  const projectIds = [...new Set(runs.map((r) => r.projectId).filter((p): p is string => !!p))]\n  return Promise.all(projectIds.map((p) => scoreProject(store, p)))\n}\n\nfunction latestByLayer(runs: Run[], layer: Run['layer']): Run | undefined {\n  const filtered = runs.filter((r) => r.layer === layer).sort((a, b) => b.startedAt - a.startedAt)\n  return filtered[0]\n}\n\nasync function extractMetaScore(store: TraceStore, builderRunId: string): Promise<number | null> {\n  const js = await judgeSpans(store, builderRunId)\n  const meta = js.find(\n    (s) => s.judgeId === 'builder-meta' && s.dimension === 'user_intent_satisfaction',\n  )\n  if (!meta) return null\n  // Normalize score to 0..1. Accept 0-1 natively; 0-10 scale is also common.\n  if (meta.score >= 0 && meta.score <= 1) return meta.score\n  if (meta.score >= 0 && meta.score <= 10) return meta.score / 10\n  return null\n}\n"],"mappings":";;;;;AA8CA,IAAa,iBAAb,MAA4B;CAC1B;CACA;CACA;CACA;CACA;CACA;CACA;CAEA,YAAY,OAAmB,MAA0B,QAAwB;EAC/E,KAAK,QAAQ;EACb,KAAK,YAAY,KAAK;EACtB,KAAK,SAAS,KAAK,UAAU,SAAS;EACtC,KAAK,gBAAgB;EACrB,KAAK,iBAAiB,IAAI,aAAa,KAAK;CAC9C;;CAGA,MAAM,UAAU,aAAa,GAAG,KAAK,UAAU,QAAyB;EACtE,MAAM,KAAK,eAAe,SAAS;GACjC;GACA,WAAW,KAAK;GAChB,QAAQ,KAAK;GACb,OAAO;EACT,CAAC;EACD,KAAK,eAAe,KAAK,eAAe;EACxC,OAAO,KAAK;CACd;;CAGA,IAAI,UAAwB;EAC1B,IAAI,CAAC,KAAK,cAAc,MAAM,IAAI,MAAM,gDAAgD;EACxF,OAAO,KAAK;CACd;;;;;CAMA,MAAM,KAAK,SAAgF;EACzF,IAAI,CAAC,KAAK,cAAc,MAAM,IAAI,MAAM,6CAA6C;EACrF,MAAM,eAAe,IAAI,aAAa,KAAK,KAAK;EAChD,MAAM,aAAa,SAAS;GAC1B,YAAY,QAAQ,cAAc,GAAG,KAAK,UAAU;GACpD,WAAW,KAAK;GAChB,QAAQ,KAAK;GACb,aAAa,KAAK;GAClB,OAAO;EACT,CAAC;EAED,MAAM,SAAS,MAAM,IADD,eAAe,QAAQ,UAAU,KAAK,aAC/B,CAAC,CAAC,IAAI,QAAQ,SAAS,YAAY;EAC9D,MAAM,aAAa,OAAO;GACxB,MAAM,OAAO;GACb,OAAO,OAAO;GACd,cAAc,OAAO,SAAS,YAAY;EAC5C,CAAC;EACD,KAAK,iBAAiB,aAAa;EACnC,OAAO;GAAE,OAAO,aAAa;GAAO;EAAO;CAC7C;;;;;;;CAQA,MAAM,eAAe,SAA8D;EACjF,MAAM,cAAc,KAAK,kBAAkB,KAAK;EAChD,IAAI,CAAC,aACH,MAAM,IAAI,MAAM,gEAAgE;EAClF,MAAM,EAAE,UAAU,WAAW;EAC7B,MAAM,SAAS,MAAM,sBAAsB,UAAU,KAAK,OAAO;GAC/D,QAAQ,UAAU,KAAK;GACvB,YAAY;IAAE,SAAS,KAAA;IAAW,WAAW,KAAA;IAAW,kBAAkB,KAAA;GAAU;EACtF,CAAC;EAED,MAAM,KAAK,MAAM,UAAU,OAAO,OAAO;GACvC;GACA,WAAW,KAAK;GAChB,QAAQ,KAAK;GACb,OAAO;EACT,CAAC;EACD,OAAO;CACT;;;CAIA,MAAM,gBAAgB,OAAe,WAAmC;EACtE,IAAI,CAAC,KAAK,cACR,MAAM,IAAI,MAAM,wDAAwD;EAC1E,MAAM,KAAK,eAAe,YAAY;GACpC,SAAS;GACT,cAAc,KAAK;GACnB,WAAW;GACX;GACA;GACA,MAAM;EACR,CAAC;CACH;;CAGA,MAAM,QAAQ,SAA2E;EACvF,MAAM,KAAK,eAAe,OAAO;GAC/B,MAAM,QAAQ;GACd,OAAO,QAAQ;GACf,OAAO,QAAQ;EACjB,CAAC;CACH;;;;;;;;CASA,MAAM,gBAAgB,YAA2C;EAC/D,MAAM,cAAc,KAAK,kBAAkB,KAAK;EAChD,IAAI,CAAC,aACH,MAAM,IAAI,MACR,8EACF;EACF,MAAM,UAAU,IAAI,aAAa,KAAK,KAAK;EAC3C,MAAM,QAAQ,SAAS;GACrB;GACA,WAAW,KAAK;GAChB,QAAQ,KAAK;GACb;GACA,OAAO;EACT,CAAC;EACD,OAAO;CACT;;;;;;CAOA,MAAM,iBAAiB,MAKH;EAClB,IAAI,CAAC,KAAK,cACR,MAAM,IAAI,MAAM,yDAAyD;EAC3E,MAAM,UAAU,IAAI,aAAa,KAAK,KAAK;EAC3C,MAAM,QAAQ,SAAS;GACrB,YAAY,KAAK,cAAc,GAAG,KAAK,UAAU;GACjD,WAAW,KAAK;GAChB,QAAQ,KAAK;GACb,aAAa,KAAK;GAClB,OAAO;EACT,CAAC;EACD,MAAM,QAAQ,OAAO;GACnB,MAAM,KAAK;GACX,OAAO,KAAK;GACZ,cAAc,KAAK,OAAO,YAAY;GACtC,OAAO,KAAK;EACd,CAAC;EACD,KAAK,iBAAiB,QAAQ;EAC9B,OAAO,QAAQ;CACjB;CAEA,IAAI,sBAA0C;EAC5C,OAAO,KAAK;CACd;CACA,IAAI,oBAAwC;EAC1C,OAAO,KAAK;CACd;AACF;;;;;;AAOA,eAAsB,qBACpB,OACA,WAOC;CACD,MAAM,OAAO,MAAM,MAAM,SAAS,EAAE,UAAU,CAAC;CAC/C,MAAM,WAAW,KACd,QAAQ,MAAM,EAAE,UAAU,SAAS,CAAC,CACpC,MAAM,GAAG,MAAM,EAAE,YAAY,EAAE,SAAS;CAC3C,MAAM,YAAY,KACf,QAAQ,MAAM,EAAE,UAAU,WAAW,CAAC,CACtC,MAAM,GAAG,MAAM,EAAE,YAAY,EAAE,SAAS;CAC3C,MAAM,iBAAiB,KACpB,QAAQ,MAAM,EAAE,UAAU,aAAa,CAAC,CACxC,MAAM,GAAG,MAAM,EAAE,YAAY,EAAE,SAAS;CAC3C,OAAO;EACL;EACA;EACA,gBAAgB,SAAS;EACzB,cAAc,UAAU;EACxB,oBAAoB;CACtB;AACF;AAEA,SAAS,WAAmB;CAC1B,IAAI,OAAO,WAAW,QAAQ,eAAe,YAAY,OAAO,WAAW,OAAO,WAAW;CAC7F,OAAO,GAAG,KAAK,IAAI,CAAC,CAAC,SAAS,EAAE,EAAE,GAAG,KAAK,OAAO,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC,MAAM,GAAG,EAAE;AAC7E;;;;;;;;;;;;;;;;;;;AC5NA,SAAgB,gBAAgB,SAAuD;CACrF,MAAM,mBAAmB,QAAQ,QAAQ,MAAM,EAAE,QAAQ,CAAC,CAAC;CAC3D,OAAO;EACL,aAAa,SACX,UACC,MAAM,EAAE,YACR,MAAM,EAAE,UACX;EACA,eAAe,SACb,UACC,MAAM,EAAE,YACR,MAAM,EAAE,YACX;EACA,gBAAgB,SACd,UACC,MAAM,EAAE,aACR,MAAM,EAAE,YACX;EACA;CACF;AACF;AAEA,SAAS,SACP,SACA,GACA,GAC8B;CAC9B,MAAM,KAAe,CAAC;CACtB,MAAM,KAAe,CAAC;CACtB,KAAK,MAAM,KAAK,SAAS;EACvB,MAAM,IAAI,EAAE,CAAC;EACb,MAAM,IAAI,EAAE,CAAC;EACb,IAAI,MAAM,QAAQ,MAAM,QAAQ,OAAO,SAAS,CAAC,KAAK,OAAO,SAAS,CAAC,GAAG;GACxE,GAAG,KAAK,CAAC;GACT,GAAG,KAAK,CAAC;EACX;CACF;CACA,IAAI,GAAG,SAAS,GAAG,OAAO,KAAA;CAC1B,OAAO;EACL,GAAG,GAAG;EACN,SAAS,SAAS,IAAI,EAAE;EACxB,UAAU,UAAU,IAAI,EAAE;CAC5B;AACF;;;ACjCA,IAAa,kBAAb,MAA6B;CACP;CAApB,YAAY,OAA2B;EAAnB,KAAA,QAAA;CAAoB;CAExC,MAAM,eAA0C;EAC9C,MAAM,OAAO,MAAM,KAAK,MAAM,SAAS;EACvC,MAAM,4BAAY,IAAI,IAAmB;EACzC,KAAK,MAAM,KAAK,MAAM;GACpB,IAAI,CAAC,EAAE,WAAW;GAClB,MAAM,MAAM,UAAU,IAAI,EAAE,SAAS,KAAK,CAAC;GAC3C,IAAI,KAAK,CAAC;GACV,UAAU,IAAI,EAAE,WAAW,GAAG;EAChC;EACA,MAAM,YAA8B,CAAC;EACrC,KAAK,MAAM,CAAC,WAAW,gBAAgB,WAAW;GAChD,MAAM,SAAS,YAAY,MAAM,CAAC,CAAC,MAAM,GAAG,MAAM,EAAE,YAAY,EAAE,SAAS;GAC3E,MAAM,QAAQ,YAAY,QAAQ,MAAM,EAAE,UAAU,SAAS;GAC7D,MAAM,SAAS,YAAY,QAAQ,MAAM,EAAE,UAAU,WAAW;GAChE,MAAM,WAAW,YAAY,QAAQ,MAAM,EAAE,UAAU,aAAa;GACpE,MAAM,SAAS,OAAO;GACtB,IAAI,CAAC,QAAQ;GACb,UAAU,KAAK;IACb;IACA,WAAW,MAAM;IACjB,YAAY,OAAO;IACnB,iBAAiB,SAAS;IAC1B,gBAAgB,OAAO;IACvB,cAAc,MAAM,EAAE,EAAE;IACxB,eAAe,OAAO,UAClB;KAAE,MAAM,OAAO,QAAQ,QAAQ;KAAO,OAAO,OAAO,QAAQ;IAAM,IAClE,KAAA;GACN,CAAC;EACH;EACA,OAAO,UAAU,MAAM,GAAG,MAAM,EAAE,iBAAiB,EAAE,cAAc;CACrE;CAEA,MAAM,gBAAgB,WAAoD;EAGxE,QADgB,MADG,KAAK,MAAM,SAAS,EAAE,UAAU,CAAC,EAAA,CAC/B,MAAM,CAAC,CAAC,MAAM,GAAG,MAAM,EAAE,YAAY,EAAE,SAC/C,CAAC,CAAC,KAAK,SAAS;GAC3B;GACA,aACE,IAAI,UAAU,YACV,SACA,IAAI,UAAU,cACZ,UACA,IAAI,UAAU,gBACZ,YACA;EACZ,EAAE;CACJ;CAEA,MAAM,aAAa,WAA2C;EAC5D,MAAM,eAAe,MAAM,KAAK,MAAM,SAAS;GAAE;GAAW,OAAO;EAAU,CAAC,EAAA,CAAG,MAC9E,GAAG,MAAM,EAAE,YAAY,EAAE,SAC5B;EACA,MAAM,cAAc,OAAO,UAAkB,KAAK,MAAM,SAAS,EAAE,aAAa,MAAM,CAAC;EACvF,MAAM,MAAqB,CAAC;EAC5B,KAAK,MAAM,OAAO,aAAa;GAC7B,MAAM,QAAQ,MAAM,KAAK,MAAM,MAAM,EAAE,OAAO,IAAI,MAAM,CAAC;GACzD,MAAM,WAAW,MAAM,YAAY,IAAI,KAAK;GAC5C,MAAM,QAAQ,SAAS,MAAM,MAAM,EAAE,UAAU,WAAW;GAC1D,MAAM,UAAoB,CAAC;GAG3B,IAAI,OAAO;IACT,MAAM,SAAS,MAAM,YAAY,MAAM,KAAK;IAC5C,KAAK,MAAM,KAAK,QAAQ,IAAI,EAAE,UAAU,eAAe,QAAQ,KAAK,EAAE,KAAK;GAC7E;GACA,KAAK,MAAM,KAAK,UAAU,IAAI,EAAE,UAAU,eAAe,QAAQ,KAAK,EAAE,KAAK;GAC7E,IAAI,KAAK;IACP,QAAQ,IAAI,UAAU,IAAI;IAC1B;IACA,cAAc,IAAI;IAClB,WAAW,IAAI;IACf,SAAS,IAAI;IACb,QAAQ,IAAI;IACZ,SAAS,IAAI;IACb,UAAU,MAAM,QAAQ,MAAM,EAAE,SAAS,KAAK,CAAC,CAAC;IAChD,WAAW,MAAM,QAAQ,MAAM,EAAE,SAAS,MAAM,CAAC,CAAC;IAClD,YAAY,OAAO;IACnB,kBAAkB;GACpB,CAAC;EACH;EACA,OAAO;CACT;AACF;;;;;;;;;;;;;;;;;;;;;;;;;;ACzEA,eAAsB,aACpB,OACA,WACkC;CAClC,MAAM,UAAU,MAAM,MAAM,SAAS,EAAE,UAAU,CAAC;CAClD,MAAM,UAAU,cAAc,SAAS,SAAS;CAChD,MAAM,QAAQ,cAAc,SAAS,WAAW;CAChD,MAAM,UAAU,QAAQ,QAAQ,MAAM,EAAE,UAAU,aAAa;CAE/D,MAAM,YAAY,UAAU,MAAM,iBAAiB,OAAO,QAAQ,KAAK,IAAI;CAC3E,MAAM,aAAa,OAAO,SAAS,SAAS;CAC5C,MAAM,gBAAgB,QACnB,KAAK,MAAM,EAAE,SAAS,KAAK,CAAC,CAC5B,QAAQ,MAAmB,OAAO,MAAM,QAAQ;CACnD,MAAM,eACJ,cAAc,SAAS,IACnB,cAAc,QAAQ,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,cAAc,SACzD;CACN,MAAM,gBAAgB,QAAQ,QAAQ,MAAM,EAAE,SAAS,SAAS,IAAI,CAAC,CAAC;CACtE,MAAM,kBAAkB,QAAQ,SAAS,IAAI,gBAAgB,QAAQ,SAAS;CAE9E,MAAM,OAAoB,QAAQ,WAAW,IAAI,kBAAkB;CACnE,MAAM,WACJ,SAAS,kBACL,cAAc,QAAQ,eAAe,OACrC,cAAc,QAAQ,eAAe,QAAQ,iBAAiB;CAEpE,OAAO;EACL;EACA;EACA,cAAc,SAAS;EACvB;EACA,YAAY,OAAO;EACnB;EACA,kBAAkB,QAAQ,KAAK,MAAM,EAAE,KAAK;EAC5C;EACA;EACA;CACF;AACF;;AAGA,eAAsB,iBAAiB,OAAuD;CAC5F,MAAM,OAAO,MAAM,MAAM,SAAS;CAClC,MAAM,aAAa,CAAC,GAAG,IAAI,IAAI,KAAK,KAAK,MAAM,EAAE,SAAS,CAAC,CAAC,QAAQ,MAAmB,CAAC,CAAC,CAAC,CAAC,CAAC;CAC5F,OAAO,QAAQ,IAAI,WAAW,KAAK,MAAM,aAAa,OAAO,CAAC,CAAC,CAAC;AAClE;AAEA,SAAS,cAAc,MAAa,OAAsC;CAExE,OADiB,KAAK,QAAQ,MAAM,EAAE,UAAU,KAAK,CAAC,CAAC,MAAM,GAAG,MAAM,EAAE,YAAY,EAAE,SACxE,CAAC,CAAC;AAClB;AAEA,eAAe,iBAAiB,OAAmB,cAA8C;CAE/F,MAAM,QAAO,MADI,WAAW,OAAO,YAAY,EAAA,CAC/B,MACb,MAAM,EAAE,YAAY,kBAAkB,EAAE,cAAc,0BACzD;CACA,IAAI,CAAC,MAAM,OAAO;CAElB,IAAI,KAAK,SAAS,KAAK,KAAK,SAAS,GAAG,OAAO,KAAK;CACpD,IAAI,KAAK,SAAS,KAAK,KAAK,SAAS,IAAI,OAAO,KAAK,QAAQ;CAC7D,OAAO;AACT"}
+{"version":3,"file":"index.js","names":[],"sources":["../../src/sandbox-harness.ts","../../src/test-graded-scenario.ts","../../src/builder-eval/builder-session.ts","../../src/builder-eval/correlation.ts","../../src/builder-eval/project-registry.ts","../../src/builder-eval/three-layer-eval.ts"],"sourcesContent":["/**\n * SandboxHarness — executes a scenario in an isolated environment and\n * emits a rich SandboxSpan into the trace.\n *\n * Two built-in drivers:\n *   - `SubprocessSandboxDriver` — spawn in a local cwd with env vars.\n *     Fast, no dependencies, fine for unit tests and most CI gates.\n *   - `DockerSandboxDriver` — lifted from tangle-router's sandbox path;\n *     shells out to `docker run`. Stronger isolation, slower startup.\n *\n * Consumers implement `SandboxDriver` for custom backends (Firecracker,\n * Cloudflare sandbox product, etc.). The harness doesn't care which.\n */\n\nimport { ConfigError } from './errors'\nimport type { TraceEmitter } from './trace/emitter'\nimport type { SandboxSpan } from './trace/schema'\n\nexport interface HarnessConfig {\n  /** Setup command (e.g. \"pnpm install\"). Non-zero exit fails the run. */\n  setupCommand?: string\n  /** Run command (e.g. \"pnpm build\"). */\n  runCommand?: string\n  /** Test command (e.g. \"pnpm test --run\"). Drives the test count + pass count. */\n  testCommand?: string\n  /** Absolute cwd for the subprocess driver. Ignored by docker driver. */\n  cwd?: string\n  /** Max wall-clock per phase in ms. Default 10 minutes. */\n  timeoutMs?: number\n  /**\n   * Cap on captured stdout+stderr bytes per phase. A runaway process can\n   * otherwise grow the in-memory buffer without bound. Once hit, further\n   * output is dropped and `outputTruncated` is set. Default 16 MiB.\n   */\n  maxOutputBytes?: number\n  /** Image for the docker driver. */\n  image?: string\n  /** Extra env vars (validated; shell-escaped). */\n  env?: Record<string, string>\n  /** Parser for the test output — maps stdout/stderr/exit code → pass count. */\n  testParser?: TestOutputParser\n}\n\nexport interface TestOutputParser {\n  id: string\n  parse(\n    stdout: string,\n    stderr: string,\n    exitCode: number,\n  ): { testsTotal: number; testsPassed: number } | undefined\n}\n\nexport interface SandboxResult {\n  phase: 'setup' | 'run' | 'test'\n  exitCode: number\n  stdout: string\n  stderr: string\n  wallMs: number\n  testsTotal?: number\n  testsPassed?: number\n  /**\n   * True when the process was killed because it exceeded `timeoutMs`. A\n   * SIGKILLed child can still close with exit code 0; callers MUST treat\n   * a timed-out phase as a hard failure regardless of `exitCode`, never\n   * as a pass. `undefined`/`false` means the process completed on its own.\n   */\n  killedByTimeout?: boolean\n  /**\n   * True when captured stdout/stderr hit `maxOutputBytes` and further\n   * output was dropped. The result is still returned (the process was\n   * not killed for this), but downstream parsers see truncated text.\n   */\n  outputTruncated?: boolean\n}\n\nexport interface SandboxDriver {\n  id: string\n  exec(\n    phase: SandboxResult['phase'],\n    command: string,\n    config: HarnessConfig,\n  ): Promise<SandboxResult>\n}\n\n// ── Parsers ──────────────────────────────────────────────────────────\n\n/** Vitest default summary line: \"Tests  X passed | Y failed\". */\nexport const vitestTestParser: TestOutputParser = {\n  id: 'vitest',\n  parse(stdout) {\n    const m = stdout.match(/Tests\\s+(\\d+)\\s+(passed|failed)(?:\\s*\\|\\s*(\\d+)\\s+(passed|failed))?/i)\n    if (!m) return undefined\n    let passed = 0\n    let failed = 0\n    const a = parseInt(m[1]!, 10)\n    const aLabel = m[2]!.toLowerCase()\n    if (aLabel === 'passed') passed += a\n    else failed += a\n    if (m[3] && m[4]) {\n      const b = parseInt(m[3], 10)\n      if (m[4].toLowerCase() === 'passed') passed += b\n      else failed += b\n    }\n    return { testsTotal: passed + failed, testsPassed: passed }\n  },\n}\n\n/** Pytest default: \"collected N items\" + \" X passed, Y failed\". */\nexport const pytestTestParser: TestOutputParser = {\n  id: 'pytest',\n  parse(stdout) {\n    const total = stdout.match(/collected\\s+(\\d+)\\s+items?/i)\n    const passed = stdout.match(/(\\d+)\\s+passed/)\n    if (!total || !passed) return undefined\n    return { testsTotal: parseInt(total[1]!, 10), testsPassed: parseInt(passed[1]!, 10) }\n  },\n}\n\n/** Jest: \"Tests: X passed, Y total\" (and optional failed). */\nexport const jestTestParser: TestOutputParser = {\n  id: 'jest',\n  parse(stdout) {\n    const m = stdout.match(/Tests:\\s+(?:(\\d+)\\s+failed[^,]*,\\s*)?(\\d+)\\s+passed,\\s+(\\d+)\\s+total/i)\n    if (!m) return undefined\n    return { testsTotal: parseInt(m[3]!, 10), testsPassed: parseInt(m[2]!, 10) }\n  },\n}\n\n/** Composite parser — tries a list of parsers in order. */\nexport function composeParsers(...parsers: TestOutputParser[]): TestOutputParser {\n  return {\n    id: parsers.map((p) => p.id).join('|'),\n    parse(stdout, stderr, exitCode) {\n      for (const p of parsers) {\n        const res = p.parse(stdout, stderr, exitCode)\n        if (res) return res\n      }\n      return undefined\n    },\n  }\n}\n\n// ── Drivers ──────────────────────────────────────────────────────────\n\nexport interface SubprocessSandboxDriverOptions {\n  /**\n   * Default cwd for all `exec` calls. Used when the per-call `HarnessConfig`\n   * does not set its own `cwd`. Lets callers bind the driver to a working\n   * directory once instead of spreading cwd into every harness config —\n   * useful when the harness config is constructed far from the call site\n   * (e.g. starter-foundry's promoter passes a static HarnessConfig per\n   * family taxonomy but needs a per-run composed-scaffold cwd).\n   */\n  cwd?: string\n  /**\n   * Default env merged into every `exec` call's env (per-call `HarnessConfig.env`\n   * still wins on key collision). Same ergonomic rationale as `cwd` above.\n   */\n  env?: Record<string, string>\n}\n\nexport class SubprocessSandboxDriver implements SandboxDriver {\n  id = 'subprocess'\n  private defaultCwd?: string\n  private defaultEnv?: Record<string, string>\n\n  constructor(options: SubprocessSandboxDriverOptions = {}) {\n    this.defaultCwd = options.cwd\n    this.defaultEnv = options.env\n  }\n\n  async exec(\n    phase: SandboxResult['phase'],\n    command: string,\n    config: HarnessConfig,\n  ): Promise<SandboxResult> {\n    const { spawn } = await import('node:child_process')\n    const start = Date.now()\n    // Per-call config wins; fall back to constructor defaults. Honoring\n    // the constructor `cwd` keeps the subprocess from inheriting Node's\n    // cwd when only the constructor arg is supplied.\n    const effectiveCwd = config.cwd ?? this.defaultCwd\n    const effectiveEnv = { ...process.env, ...(this.defaultEnv ?? {}), ...(config.env ?? {}) }\n    const maxOutputBytes = config.maxOutputBytes ?? 16 * 1024 * 1024\n    return await new Promise<SandboxResult>((resolve) => {\n      const child = spawn(command, {\n        shell: true,\n        cwd: effectiveCwd,\n        env: effectiveEnv,\n        // Own process group so a timeout can SIGKILL the whole tree, not\n        // just the shell — otherwise a runaway grandchild keeps the pipes\n        // open and `close` never fires (the timeout silently does nothing).\n        detached: process.platform !== 'win32',\n      })\n      let stdout = ''\n      let stderr = ''\n      let outputBytes = 0\n      let outputTruncated = false\n      let killedByTimeout = false\n      let settled = false\n\n      const onStdout = (d: unknown) => {\n        const chunk = capture(String(d))\n        if (chunk) stdout += chunk\n      }\n      const onStderr = (d: unknown) => {\n        const chunk = capture(String(d))\n        if (chunk) stderr += chunk\n      }\n      // Bound the in-memory buffer so a runaway process can't OOM the\n      // harness. Once the cap is hit we keep draining the streams (to let\n      // the child reach `close`) but discard the bytes.\n      function capture(s: string): string {\n        if (outputBytes >= maxOutputBytes) {\n          outputTruncated = true\n          return ''\n        }\n        const room = maxOutputBytes - outputBytes\n        if (s.length > room) {\n          outputBytes = maxOutputBytes\n          outputTruncated = true\n          return s.slice(0, room)\n        }\n        outputBytes += s.length\n        return s\n      }\n\n      child.stdout?.on('data', onStdout)\n      child.stderr?.on('data', onStderr)\n\n      const cleanup = () => {\n        clearTimeout(timeout)\n        if (forceResolve) clearTimeout(forceResolve)\n        child.stdout?.off('data', onStdout)\n        child.stderr?.off('data', onStderr)\n        child.removeAllListeners('close')\n        child.removeAllListeners('error')\n      }\n\n      // Resolve exactly once. `code`/`signal` come from `close`; on the\n      // timeout path we force a non-zero exit so a SIGKILLed child that\n      // reports exit 0 can never read as a clean pass downstream.\n      const finish = (outcome: { code: number | null; runnerError?: string }) => {\n        if (settled) return\n        settled = true\n        cleanup()\n        const wallMs = Date.now() - start\n        const exitCode = killedByTimeout\n          ? outcome.code && outcome.code !== 0\n            ? outcome.code\n            : 124\n          : (outcome.code ?? 1)\n        const parsed =\n          !killedByTimeout && phase === 'test' && config.testParser\n            ? config.testParser.parse(stdout, stderr, exitCode)\n            : undefined\n        resolve({\n          phase,\n          exitCode,\n          stdout,\n          stderr: outcome.runnerError ? stderr + outcome.runnerError : stderr,\n          wallMs,\n          testsTotal: parsed?.testsTotal,\n          testsPassed: parsed?.testsPassed,\n          killedByTimeout: killedByTimeout || undefined,\n          outputTruncated: outputTruncated || undefined,\n        })\n      }\n\n      // Kill the whole process group on win32-less platforms (we spawned\n      // detached). On a clean close this is a no-op.\n      const killTree = () => {\n        try {\n          if (process.platform !== 'win32' && typeof child.pid === 'number') {\n            process.kill(-child.pid, 'SIGKILL')\n          } else {\n            child.kill('SIGKILL')\n          }\n        } catch (err) {\n          console.warn('[sandbox-harness] SIGKILL on timeout failed:', err)\n        }\n      }\n\n      let forceResolve: ReturnType<typeof setTimeout> | undefined\n      const timeout = setTimeout(\n        () => {\n          killedByTimeout = true\n          killTree()\n          // Give `close` a brief window to fire (flushing any final output)\n          // after the group dies; if an orphaned grandchild keeps the pipes\n          // open, force-resolve so the harness can't hang on a runaway.\n          forceResolve = setTimeout(() => finish({ code: null }), 2000)\n        },\n        config.timeoutMs ?? 10 * 60_000,\n      )\n\n      child.on('close', (code) => {\n        if (forceResolve) clearTimeout(forceResolve)\n        finish({ code })\n      })\n      child.on('error', (err) => {\n        if (forceResolve) clearTimeout(forceResolve)\n        finish({ code: 127, runnerError: String(err) })\n      })\n    })\n  }\n}\n\nexport class DockerSandboxDriver implements SandboxDriver {\n  id = 'docker'\n\n  async exec(\n    phase: SandboxResult['phase'],\n    command: string,\n    config: HarnessConfig,\n  ): Promise<SandboxResult> {\n    if (!config.image) throw new ConfigError('DockerSandboxDriver requires config.image')\n    const sub = new SubprocessSandboxDriver()\n    const envArgs = Object.entries(config.env ?? {})\n      .map(([k, v]) => `-e ${shellQuote(k)}=${shellQuote(v)}`)\n      .join(' ')\n    const wrapped = `docker run --rm ${envArgs} ${shellQuote(config.image)} sh -c ${shellQuote(command)}`\n    return sub.exec(phase, wrapped, { ...config, env: undefined })\n  }\n}\n\nfunction shellQuote(v: string): string {\n  if (/^[A-Za-z0-9_\\-/.@:=]+$/.test(v)) return v\n  return `'${v.replace(/'/g, `'\\\\''`)}'`\n}\n\n// ── Harness orchestration ────────────────────────────────────────────\n\nexport interface SandboxHarnessResult {\n  passed: boolean\n  setup?: SandboxResult\n  run?: SandboxResult\n  test?: SandboxResult\n  totalWallMs: number\n  /** Final score — 0 when no tests; otherwise testsPassed/testsTotal. */\n  score: number\n}\n\nexport class SandboxHarness {\n  private driver: SandboxDriver\n  constructor(driver: SandboxDriver = new SubprocessSandboxDriver()) {\n    this.driver = driver\n  }\n\n  async run(config: HarnessConfig, emitter: TraceEmitter): Promise<SandboxHarnessResult> {\n    const handle = await emitter.sandbox({\n      name: `sandbox(${this.driver.id})`,\n      image: config.image,\n      command: [config.setupCommand, config.runCommand, config.testCommand]\n        .filter(Boolean)\n        .join(' && '),\n    })\n    const result: SandboxHarnessResult = { passed: false, totalWallMs: 0, score: 0 }\n    try {\n      if (config.setupCommand) {\n        result.setup = await this.driver.exec('setup', config.setupCommand, config)\n        result.totalWallMs += result.setup.wallMs\n        if (result.setup.killedByTimeout || result.setup.exitCode !== 0) {\n          const reason = result.setup.killedByTimeout\n            ? `setup timed out after ${result.setup.wallMs}ms`\n            : `setup failed (exit ${result.setup.exitCode})`\n          await handle.fail(reason, {\n            exitCode: result.setup.exitCode,\n            wallMs: result.totalWallMs,\n          } as Partial<SandboxSpan>)\n          return result\n        }\n      }\n      if (config.runCommand) {\n        result.run = await this.driver.exec('run', config.runCommand, config)\n        result.totalWallMs += result.run.wallMs\n        if (result.run.killedByTimeout || result.run.exitCode !== 0) {\n          const reason = result.run.killedByTimeout\n            ? `run timed out after ${result.run.wallMs}ms`\n            : `run failed (exit ${result.run.exitCode})`\n          await handle.fail(reason, {\n            exitCode: result.run.exitCode,\n            wallMs: result.totalWallMs,\n          } as Partial<SandboxSpan>)\n          return result\n        }\n      }\n      if (config.testCommand) {\n        result.test = await this.driver.exec('test', config.testCommand, config)\n        result.totalWallMs += result.test.wallMs\n        // A timed-out test phase is a hard fail regardless of the exit code\n        // a SIGKILLed child happens to report — never let it score as a pass.\n        const passed = !result.test.killedByTimeout && result.test.exitCode === 0\n        result.passed = passed\n        if (result.test.testsTotal !== undefined && result.test.testsTotal > 0) {\n          result.score = (result.test.testsPassed ?? 0) / result.test.testsTotal\n        } else {\n          result.score = passed ? 1 : 0\n        }\n        await handle.end({\n          exitCode: result.test.exitCode,\n          testsTotal: result.test.testsTotal,\n          testsPassed: result.test.testsPassed,\n          wallMs: result.totalWallMs,\n          status: passed ? 'ok' : 'error',\n        } as Partial<SandboxSpan>)\n      } else {\n        result.passed = true\n        result.score = 1\n        await handle.end({ wallMs: result.totalWallMs } as Partial<SandboxSpan>)\n      }\n    } catch (err) {\n      await handle.fail(err instanceof Error ? err : String(err))\n      throw err\n    }\n    return result\n  }\n}\n","/**\n * TestGradedScenario — a scenario whose score comes from a test suite.\n *\n * This is the SWE-bench pattern generalized. The scenario ships:\n *   - fixture data (setup instructions)\n *   - a test command the harness runs\n *   - optional assertion overrides\n *\n * The runner emits a run, delegates to SandboxHarness, records the\n * outcome, and returns a structured verdict. Consumers bind their own\n * agent execution to this contract.\n */\n\nimport type { HarnessConfig, SandboxDriver, SandboxHarnessResult } from './sandbox-harness'\nimport { SandboxHarness } from './sandbox-harness'\nimport { TraceEmitter } from './trace/emitter'\nimport type { FailureClass, Run } from './trace/schema'\nimport type { TraceStore } from './trace/store'\n\nexport interface TestGradedScenario {\n  id: string\n  description?: string\n  harness: HarnessConfig\n  /** Optional pass threshold in 0..1 (default 1.0 = all tests must pass). */\n  passThreshold?: number\n  /** Provenance for dataset tracking. */\n  datasetVersion?: string\n  /** Free-form tags (difficulty, category, etc.). */\n  tags?: Record<string, string>\n}\n\nexport interface TestGradedRunOptions {\n  variantId?: string\n  driver?: SandboxDriver\n  /** Metadata recorded on the Run (codeSha, promptSha, modelFingerprint, seed). */\n  provenance?: Pick<Run, 'codeSha' | 'promptSha' | 'modelFingerprint' | 'seed' | 'envFingerprint'>\n}\n\nexport interface TestGradedRunResult {\n  runId: string\n  scenario: TestGradedScenario\n  harness: SandboxHarnessResult\n  pass: boolean\n  score: number\n  failureClass?: FailureClass\n}\n\nexport async function runTestGradedScenario(\n  scenario: TestGradedScenario,\n  store: TraceStore,\n  options: TestGradedRunOptions = {},\n): Promise<TestGradedRunResult> {\n  const emitter = new TraceEmitter(store)\n  await emitter.startRun({\n    scenarioId: scenario.id,\n    variantId: options.variantId,\n    datasetVersion: scenario.datasetVersion,\n    tags: scenario.tags,\n    ...options.provenance,\n  })\n  const harness = new SandboxHarness(options.driver)\n  const result = await harness.run(scenario.harness, emitter)\n  const threshold = scenario.passThreshold ?? 1.0\n  const pass = result.passed && result.score >= threshold\n  const setupFailed = result.setup !== undefined && result.setup.exitCode !== 0\n  const runFailed = result.run !== undefined && result.run.exitCode !== 0\n  const testFailed = result.test !== undefined && result.test.exitCode !== 0\n  const failureClass: FailureClass | undefined = pass\n    ? 'success'\n    : setupFailed || runFailed\n      ? 'sandbox_failure'\n      : testFailed\n        ? 'format_drift'\n        : 'unknown'\n  await emitter.endRun({\n    pass,\n    score: result.score,\n    failureClass,\n    notes: pass ? undefined : reasonForFailure(result),\n  })\n  return {\n    runId: emitter.runId,\n    scenario,\n    harness: result,\n    pass,\n    score: result.score,\n    failureClass,\n  }\n}\n\nfunction reasonForFailure(result: SandboxHarnessResult): string {\n  if (result.setup && result.setup.exitCode !== 0)\n    return `setup failed: exit ${result.setup.exitCode}`\n  if (result.run && result.run.exitCode !== 0) return `run failed: exit ${result.run.exitCode}`\n  if (result.test) {\n    if (result.test.testsTotal !== undefined) {\n      return `tests: ${result.test.testsPassed ?? 0}/${result.test.testsTotal}`\n    }\n    return `test exit ${result.test.exitCode}`\n  }\n  return 'no test command'\n}\n","/**\n * BuilderSession — ties a builder-of-builders workflow together.\n *\n * Models agent-builder's shape: Project → Chat → Edit → Ship → App →\n * AppAgent. Each layer is a Run (linked via parentRunId). The\n * framework-enforced invariants:\n *\n *   - One Project → many Chats; chatId scopes runs within a project.\n *   - One Chat = one builder Run with `layer='builder'`.\n *   - One Ship = one child Run with `layer='app-build'` + SandboxHarness.\n *   - One AppScenario = one grandchild Run with `layer='app-runtime'`.\n *\n * Consumers obtain a BuilderSession, call `startChat`, drive the\n * builder agent (emitting spans), and call `ship` / `runAppScenario`\n * as the workflow progresses. The session reconstructs itself from\n * trace data via `resume(store, projectId)`.\n */\n\nimport type { HarnessConfig, SandboxDriver, SandboxHarnessResult } from '../sandbox-harness'\nimport { SandboxHarness } from '../sandbox-harness'\nimport type { TestGradedRunResult, TestGradedScenario } from '../test-graded-scenario'\nimport { runTestGradedScenario } from '../test-graded-scenario'\nimport { TraceEmitter } from '../trace/emitter'\nimport type { Run } from '../trace/schema'\nimport type { TraceStore } from '../trace/store'\n\nexport interface BuilderSessionInit {\n  projectId: string\n  chatId?: string\n  /** Free-form: user's task description, project name, etc. Stored on the builder Run. */\n  tags?: Record<string, string>\n}\n\nexport interface ShipOptions {\n  harness: HarnessConfig\n  driver?: SandboxDriver\n  /** scenarioId of this app-build run. Defaults to `${projectId}/build`. */\n  scenarioId?: string\n}\n\nexport interface RunAppScenarioOptions {\n  scenario: TestGradedScenario\n  /** Harness driver override; defaults to the one the session was created with. */\n  driver?: SandboxDriver\n}\n\nexport class BuilderSession {\n  private store: TraceStore\n  private builderEmitter: TraceEmitter\n  readonly projectId: string\n  readonly chatId: string\n  private builderRunId?: string\n  private lastBuildRunId?: string\n  private defaultDriver?: SandboxDriver\n\n  constructor(store: TraceStore, init: BuilderSessionInit, driver?: SandboxDriver) {\n    this.store = store\n    this.projectId = init.projectId\n    this.chatId = init.chatId ?? cryptoId()\n    this.defaultDriver = driver\n    this.builderEmitter = new TraceEmitter(store)\n  }\n\n  /** Start the builder (L0) run for this chat. Returns the runId. */\n  async startChat(scenarioId = `${this.projectId}/chat`): Promise<string> {\n    await this.builderEmitter.startRun({\n      scenarioId,\n      projectId: this.projectId,\n      chatId: this.chatId,\n      layer: 'builder',\n    })\n    this.builderRunId = this.builderEmitter.runId\n    return this.builderRunId\n  }\n\n  /** The emitter for builder-level spans (edits, LLM calls, tool invocations). */\n  get emitter(): TraceEmitter {\n    if (!this.builderRunId) throw new Error('BuilderSession.emitter: call startChat() first')\n    return this.builderEmitter\n  }\n\n  /**\n   * Ship the project's generated app: run the sandbox harness as a child\n   * Run (`layer='app-build'`). Returns the build result + runId.\n   */\n  async ship(options: ShipOptions): Promise<{ runId: string; result: SandboxHarnessResult }> {\n    if (!this.builderRunId) throw new Error('BuilderSession.ship: call startChat() first')\n    const buildEmitter = new TraceEmitter(this.store)\n    await buildEmitter.startRun({\n      scenarioId: options.scenarioId ?? `${this.projectId}/build`,\n      projectId: this.projectId,\n      chatId: this.chatId,\n      parentRunId: this.builderRunId,\n      layer: 'app-build',\n    })\n    const harness = new SandboxHarness(options.driver ?? this.defaultDriver)\n    const result = await harness.run(options.harness, buildEmitter)\n    await buildEmitter.endRun({\n      pass: result.passed,\n      score: result.score,\n      failureClass: result.passed ? 'success' : 'sandbox_failure',\n    })\n    this.lastBuildRunId = buildEmitter.runId\n    return { runId: buildEmitter.runId, result }\n  }\n\n  /**\n   * Run a domain scenario against the just-built app as a grandchild Run\n   * (`layer='app-runtime'`). The `ship` call must precede this so the\n   * parent is set correctly; if no build exists yet the session attaches\n   * directly to the builder run (useful for prototypes).\n   */\n  async runAppScenario(options: RunAppScenarioOptions): Promise<TestGradedRunResult> {\n    const parentRunId = this.lastBuildRunId ?? this.builderRunId\n    if (!parentRunId)\n      throw new Error('BuilderSession.runAppScenario: call startChat() + ship() first')\n    const { scenario, driver } = options\n    const result = await runTestGradedScenario(scenario, this.store, {\n      driver: driver ?? this.defaultDriver,\n      provenance: { codeSha: undefined, promptSha: undefined, modelFingerprint: undefined },\n    })\n    // Attach to the parent chain by updating the stored Run in place.\n    await this.store.updateRun(result.runId, {\n      parentRunId,\n      projectId: this.projectId,\n      chatId: this.chatId,\n      layer: 'app-runtime',\n    })\n    return result\n  }\n\n  /** Record an end-of-chat meta score (judge verdict on whether the builder\n   *  served the user's intent). Accepts a numeric score + optional rationale. */\n  async recordMetaScore(score: number, rationale?: string): Promise<void> {\n    if (!this.builderRunId)\n      throw new Error('BuilderSession.recordMetaScore: call startChat() first')\n    await this.builderEmitter.recordJudge({\n      judgeId: 'builder-meta',\n      targetSpanId: this.builderRunId, // attach to the builder run itself\n      dimension: 'user_intent_satisfaction',\n      score,\n      rationale,\n      name: 'builder-meta',\n    })\n  }\n\n  /** Close the builder Run with a final outcome. */\n  async endChat(outcome: { pass: boolean; score?: number; notes?: string }): Promise<void> {\n    await this.builderEmitter.endRun({\n      pass: outcome.pass,\n      score: outcome.score,\n      notes: outcome.notes,\n    })\n  }\n\n  /**\n   * Inline app-runtime run — for cases where the \"scenario\" isn't a\n   * SWE-bench-style test suite but a live agent interaction (LLM chat,\n   * domain flow). Returns an emitter bound to a fresh Run in the\n   * `app-runtime` layer; caller emits spans inside and calls\n   * `.endRun()` with the final verdict.\n   */\n  async startAppRuntime(scenarioId: string): Promise<TraceEmitter> {\n    const parentRunId = this.lastBuildRunId ?? this.builderRunId\n    if (!parentRunId)\n      throw new Error(\n        'BuilderSession.startAppRuntime: call startChat() + (optionally) ship() first',\n      )\n    const emitter = new TraceEmitter(this.store)\n    await emitter.startRun({\n      scenarioId,\n      projectId: this.projectId,\n      chatId: this.chatId,\n      parentRunId,\n      layer: 'app-runtime',\n    })\n    return emitter\n  }\n\n  /**\n   * Lightweight \"ship marker\" — record an app-build Run with a caller-\n   * provided verdict. Use when there isn't a sandbox harness to run but\n   * you still want to mark the build state at publish time.\n   */\n  async recordShipMarker(args: {\n    pass: boolean\n    score: number\n    scenarioId?: string\n    notes?: string\n  }): Promise<string> {\n    if (!this.builderRunId)\n      throw new Error('BuilderSession.recordShipMarker: call startChat() first')\n    const emitter = new TraceEmitter(this.store)\n    await emitter.startRun({\n      scenarioId: args.scenarioId ?? `${this.projectId}/ship`,\n      projectId: this.projectId,\n      chatId: this.chatId,\n      parentRunId: this.builderRunId,\n      layer: 'app-build',\n    })\n    await emitter.endRun({\n      pass: args.pass,\n      score: args.score,\n      failureClass: args.pass ? 'success' : 'sandbox_failure',\n      notes: args.notes,\n    })\n    this.lastBuildRunId = emitter.runId\n    return emitter.runId\n  }\n\n  get lastBuildRunIdValue(): string | undefined {\n    return this.lastBuildRunId\n  }\n  get builderRunIdValue(): string | undefined {\n    return this.builderRunId\n  }\n}\n\n/**\n * Reconstruct the most recent BuilderSession state for a given project —\n * returns { builderRunId, lastBuildRunId, chatRuns }. For chat-first UIs\n * this is how a resumed session finds its place in the edit history.\n */\nexport async function resumeBuilderSession(\n  store: TraceStore,\n  projectId: string,\n): Promise<{\n  projectId: string\n  chatRuns: Run[]\n  lastBuilderRun?: Run\n  lastBuildRun?: Run\n  lastAppRuntimeRuns: Run[]\n}> {\n  const runs = await store.listRuns({ projectId })\n  const chatRuns = runs\n    .filter((r) => r.layer === 'builder')\n    .sort((a, b) => b.startedAt - a.startedAt)\n  const buildRuns = runs\n    .filter((r) => r.layer === 'app-build')\n    .sort((a, b) => b.startedAt - a.startedAt)\n  const appRuntimeRuns = runs\n    .filter((r) => r.layer === 'app-runtime')\n    .sort((a, b) => b.startedAt - a.startedAt)\n  return {\n    projectId,\n    chatRuns,\n    lastBuilderRun: chatRuns[0],\n    lastBuildRun: buildRuns[0],\n    lastAppRuntimeRuns: appRuntimeRuns,\n  }\n}\n\nfunction cryptoId(): string {\n  if (typeof globalThis.crypto?.randomUUID === 'function') return globalThis.crypto.randomUUID()\n  return `${Date.now().toString(36)}-${Math.random().toString(36).slice(2, 10)}`\n}\n","/**\n * Meta-eval correlation — the highest-leverage signal in the framework.\n *\n * Given a corpus of three-layer project reports, compute how well each\n * pair of layers correlates. The question we care about most:\n *\n *   Does `metaScore` (what the builder thinks it did) predict\n *   `runtimeScore` (what the user actually gets)?\n *\n * If r < ~0.4, the builder's self-scoring is broken — it's optimizing\n * for something other than real-world success. If r > 0.7, meta_score\n * is a usable proxy and can drive CI gates cheaply.\n *\n * Non-parametric rank correlation (Spearman) is also reported because\n * meta scores are often ordinal-ish.\n */\n\nimport { pearsonR, spearmanR } from '../statistics'\nimport type { ThreeLayerProjectReport } from './three-layer-eval'\n\nexport interface LayerCorrelation {\n  n: number\n  pearson: number\n  spearman: number\n}\n\nexport interface CorrelationReport {\n  /** Pairs present in the corpus (layers with ≥ 2 matched data points). */\n  metaVsBuild?: LayerCorrelation\n  metaVsRuntime?: LayerCorrelation\n  buildVsRuntime?: LayerCorrelation\n  /** Number of complete projects (all 3 scores present). */\n  completeProjects: number\n}\n\nexport function correlateLayers(reports: ThreeLayerProjectReport[]): CorrelationReport {\n  const completeProjects = reports.filter((r) => r.complete).length\n  return {\n    metaVsBuild: pairwise(\n      reports,\n      (r) => r.metaScore,\n      (r) => r.buildScore,\n    ),\n    metaVsRuntime: pairwise(\n      reports,\n      (r) => r.metaScore,\n      (r) => r.runtimeScore,\n    ),\n    buildVsRuntime: pairwise(\n      reports,\n      (r) => r.buildScore,\n      (r) => r.runtimeScore,\n    ),\n    completeProjects,\n  }\n}\n\nfunction pairwise(\n  reports: ThreeLayerProjectReport[],\n  a: (r: ThreeLayerProjectReport) => number | null,\n  b: (r: ThreeLayerProjectReport) => number | null,\n): LayerCorrelation | undefined {\n  const xs: number[] = []\n  const ys: number[] = []\n  for (const r of reports) {\n    const x = a(r)\n    const y = b(r)\n    if (x !== null && y !== null && Number.isFinite(x) && Number.isFinite(y)) {\n      xs.push(x)\n      ys.push(y)\n    }\n  }\n  if (xs.length < 2) return undefined\n  return {\n    n: xs.length,\n    pearson: pearsonR(xs, ys),\n    spearman: spearmanR(xs, ys),\n  }\n}\n","/**\n * ProjectRegistry — project-level aggregation over the trace corpus.\n *\n * Thin reader over TraceStore that answers the questions a chat-first,\n * resumable UI needs:\n *   - listProjects() → project IDs with latest activity\n *   - projectTimeline(id) → chats + builds + runtime runs, chronological\n *   - projectChats(id) → chat-level summaries (turn count, outcome)\n *\n * All queries are pure reads; no state duplication.\n */\n\nimport type { Run } from '../trace/schema'\nimport type { TraceStore } from '../trace/store'\n\nexport interface ProjectSummary {\n  projectId: string\n  chatCount: number\n  buildCount: number\n  appRuntimeCount: number\n  lastActivityAt: number\n  latestChatId?: string\n  latestOutcome?: { pass: boolean; score?: number }\n}\n\nexport interface ChatSummary {\n  chatId: string\n  projectId: string\n  builderRunId: string\n  startedAt: number\n  endedAt?: number\n  status: Run['status']\n  outcome?: Run['outcome']\n  /** Counts of spans emitted during the chat. */\n  llmTurns?: number\n  toolCalls?: number\n  buildRunId?: string\n  appRuntimeRunIds: string[]\n}\n\nexport interface ProjectTimelineEntry {\n  run: Run\n  layerBucket: 'chat' | 'build' | 'runtime' | 'other'\n}\n\nexport class ProjectRegistry {\n  constructor(private store: TraceStore) {}\n\n  async listProjects(): Promise<ProjectSummary[]> {\n    const runs = await this.store.listRuns()\n    const byProject = new Map<string, Run[]>()\n    for (const r of runs) {\n      if (!r.projectId) continue\n      const arr = byProject.get(r.projectId) ?? []\n      arr.push(r)\n      byProject.set(r.projectId, arr)\n    }\n    const summaries: ProjectSummary[] = []\n    for (const [projectId, projectRuns] of byProject) {\n      const sorted = projectRuns.slice().sort((a, b) => b.startedAt - a.startedAt)\n      const chats = projectRuns.filter((r) => r.layer === 'builder')\n      const builds = projectRuns.filter((r) => r.layer === 'app-build')\n      const runtimes = projectRuns.filter((r) => r.layer === 'app-runtime')\n      const latest = sorted[0]\n      if (!latest) continue\n      summaries.push({\n        projectId,\n        chatCount: chats.length,\n        buildCount: builds.length,\n        appRuntimeCount: runtimes.length,\n        lastActivityAt: latest.startedAt,\n        latestChatId: chats[0]?.chatId,\n        latestOutcome: latest.outcome\n          ? { pass: latest.outcome.pass ?? false, score: latest.outcome.score }\n          : undefined,\n      })\n    }\n    return summaries.sort((a, b) => b.lastActivityAt - a.lastActivityAt)\n  }\n\n  async projectTimeline(projectId: string): Promise<ProjectTimelineEntry[]> {\n    const runs = await this.store.listRuns({ projectId })\n    const ordered = runs.slice().sort((a, b) => a.startedAt - b.startedAt)\n    return ordered.map((run) => ({\n      run,\n      layerBucket:\n        run.layer === 'builder'\n          ? 'chat'\n          : run.layer === 'app-build'\n            ? 'build'\n            : run.layer === 'app-runtime'\n              ? 'runtime'\n              : 'other',\n    }))\n  }\n\n  async projectChats(projectId: string): Promise<ChatSummary[]> {\n    const builderRuns = (await this.store.listRuns({ projectId, layer: 'builder' })).sort(\n      (a, b) => b.startedAt - a.startedAt,\n    )\n    const childrenFor = async (runId: string) => this.store.listRuns({ parentRunId: runId })\n    const out: ChatSummary[] = []\n    for (const run of builderRuns) {\n      const spans = await this.store.spans({ runId: run.runId })\n      const children = await childrenFor(run.runId)\n      const build = children.find((c) => c.layer === 'app-build')\n      const runtime: string[] = []\n      // Runtime runs can be grandchildren (attached to the build) or siblings\n      // when shipped skipped.\n      if (build) {\n        const grands = await childrenFor(build.runId)\n        for (const g of grands) if (g.layer === 'app-runtime') runtime.push(g.runId)\n      }\n      for (const c of children) if (c.layer === 'app-runtime') runtime.push(c.runId)\n      out.push({\n        chatId: run.chatId ?? run.runId,\n        projectId,\n        builderRunId: run.runId,\n        startedAt: run.startedAt,\n        endedAt: run.endedAt,\n        status: run.status,\n        outcome: run.outcome,\n        llmTurns: spans.filter((s) => s.kind === 'llm').length,\n        toolCalls: spans.filter((s) => s.kind === 'tool').length,\n        buildRunId: build?.runId,\n        appRuntimeRunIds: runtime,\n      })\n    }\n    return out\n  }\n}\n","/**\n * Three-layer evaluation — the canonical scoring breakdown for\n * builder-of-builders workflows.\n *\n *   meta_score:    did the builder understand + satisfy user intent?\n *                  (judge verdict attached to the builder run)\n *   build_score:   did the generated scaffold build + pass its own tests?\n *                  (outcome.score on the app-build child run)\n *   runtime_score: did the generated agent pass its domain scenarios?\n *                  (mean outcome.score over app-runtime grandchild runs)\n *\n * Returns a structured report per project. The cross-layer correlation\n * is the highest-leverage signal the framework computes — if\n * meta_score doesn't predict runtime_score, the builder's self-scoring\n * is broken.\n *\n * Scaffold-only mode: when a project has no `app-runtime` runs (e.g. a\n * scaffold-builder eval that grades compose + build without driving a\n * runtime scenario), `kind` is `'scaffold-only'` and `complete` measures\n * meta + build only. Consumers can tell the two apart without having to\n * interpret null-runtime as either \"not yet computed\" or \"N/A for this\n * project shape\".\n */\n\nimport { judgeSpans } from '../trace/query'\nimport type { Run } from '../trace/schema'\nimport type { TraceStore } from '../trace/store'\n\nexport type ProjectKind = 'full' | 'scaffold-only'\n\nexport interface ThreeLayerProjectReport {\n  projectId: string\n  /**\n   * `'full'` when the project has at least one `app-runtime` run;\n   * `'scaffold-only'` when it only has meta + build layers. Lets\n   * downstream consumers treat a null runtime score as expected\n   * (scaffold-only) vs. missing (full, pipeline broke).\n   */\n  kind: ProjectKind\n  builderRunId?: string\n  /** Judge-verdict score on the builder run (0..1 after normalization). */\n  metaScore: number | null\n  buildRunId?: string\n  /** 0..1 from the sandbox harness (testsPassed / testsTotal). */\n  buildScore: number | null\n  appRuntimeRunIds: string[]\n  /** Mean of outcome.score over app-runtime runs, 0..1. Always null in scaffold-only mode. */\n  runtimeScore: number | null\n  runtimePassRate: number | null\n  /**\n   * Layer-aware completeness:\n   *   - `kind='full'`: all three layers scored\n   *   - `kind='scaffold-only'`: meta + build scored (runtime not applicable)\n   */\n  complete: boolean\n}\n\nexport async function scoreProject(\n  store: TraceStore,\n  projectId: string,\n): Promise<ThreeLayerProjectReport> {\n  const allRuns = await store.listRuns({ projectId })\n  const builder = latestByLayer(allRuns, 'builder')\n  const build = latestByLayer(allRuns, 'app-build')\n  const runtime = allRuns.filter((r) => r.layer === 'app-runtime')\n\n  const metaScore = builder ? await extractMetaScore(store, builder.runId) : null\n  const buildScore = build?.outcome?.score ?? null\n  const runtimeScores = runtime\n    .map((r) => r.outcome?.score)\n    .filter((s): s is number => typeof s === 'number')\n  const runtimeScore =\n    runtimeScores.length > 0\n      ? runtimeScores.reduce((a, b) => a + b, 0) / runtimeScores.length\n      : null\n  const runtimePassed = runtime.filter((r) => r.outcome?.pass === true).length\n  const runtimePassRate = runtime.length > 0 ? runtimePassed / runtime.length : null\n\n  const kind: ProjectKind = runtime.length === 0 ? 'scaffold-only' : 'full'\n  const complete =\n    kind === 'scaffold-only'\n      ? metaScore !== null && buildScore !== null\n      : metaScore !== null && buildScore !== null && runtimeScore !== null\n\n  return {\n    projectId,\n    kind,\n    builderRunId: builder?.runId,\n    metaScore,\n    buildRunId: build?.runId,\n    buildScore,\n    appRuntimeRunIds: runtime.map((r) => r.runId),\n    runtimeScore,\n    runtimePassRate,\n    complete,\n  }\n}\n\n/** Aggregate scoring across every project in a corpus. */\nexport async function scoreAllProjects(store: TraceStore): Promise<ThreeLayerProjectReport[]> {\n  const runs = await store.listRuns()\n  const projectIds = [...new Set(runs.map((r) => r.projectId).filter((p): p is string => !!p))]\n  return Promise.all(projectIds.map((p) => scoreProject(store, p)))\n}\n\nfunction latestByLayer(runs: Run[], layer: Run['layer']): Run | undefined {\n  const filtered = runs.filter((r) => r.layer === layer).sort((a, b) => b.startedAt - a.startedAt)\n  return filtered[0]\n}\n\nasync function extractMetaScore(store: TraceStore, builderRunId: string): Promise<number | null> {\n  const js = await judgeSpans(store, builderRunId)\n  const meta = js.find(\n    (s) => s.judgeId === 'builder-meta' && s.dimension === 'user_intent_satisfaction',\n  )\n  if (!meta) return null\n  // Normalize score to 0..1. Accept 0-1 natively; 0-10 scale is also common.\n  if (meta.score >= 0 && meta.score <= 1) return meta.score\n  if (meta.score >= 0 && meta.score <= 10) return meta.score / 10\n  return null\n}\n"],"mappings":";;;;AAiKA,IAAa,0BAAb,MAA8D;CAC5D,KAAK;CACL;CACA;CAEA,YAAY,UAA0C,CAAC,GAAG;EACxD,KAAK,aAAa,QAAQ;EAC1B,KAAK,aAAa,QAAQ;CAC5B;CAEA,MAAM,KACJ,OACA,SACA,QACwB;EACxB,MAAM,EAAE,UAAU,MAAM,OAAO;EAC/B,MAAM,QAAQ,KAAK,IAAI;EAIvB,MAAM,eAAe,OAAO,OAAO,KAAK;EACxC,MAAM,eAAe;GAAE,GAAG,QAAQ;GAAK,GAAI,KAAK,cAAc,CAAC;GAAI,GAAI,OAAO,OAAO,CAAC;EAAG;EACzF,MAAM,iBAAiB,OAAO,kBAAkB,KAAK,OAAO;EAC5D,OAAO,MAAM,IAAI,SAAwB,YAAY;GACnD,MAAM,QAAQ,MAAM,SAAS;IAC3B,OAAO;IACP,KAAK;IACL,KAAK;IAIL,UAAU,QAAQ,aAAa;GACjC,CAAC;GACD,IAAI,SAAS;GACb,IAAI,SAAS;GACb,IAAI,cAAc;GAClB,IAAI,kBAAkB;GACtB,IAAI,kBAAkB;GACtB,IAAI,UAAU;GAEd,MAAM,YAAY,MAAe;IAC/B,MAAM,QAAQ,QAAQ,OAAO,CAAC,CAAC;IAC/B,IAAI,OAAO,UAAU;GACvB;GACA,MAAM,YAAY,MAAe;IAC/B,MAAM,QAAQ,QAAQ,OAAO,CAAC,CAAC;IAC/B,IAAI,OAAO,UAAU;GACvB;GAIA,SAAS,QAAQ,GAAmB;IAClC,IAAI,eAAe,gBAAgB;KACjC,kBAAkB;KAClB,OAAO;IACT;IACA,MAAM,OAAO,iBAAiB;IAC9B,IAAI,EAAE,SAAS,MAAM;KACnB,cAAc;KACd,kBAAkB;KAClB,OAAO,EAAE,MAAM,GAAG,IAAI;IACxB;IACA,eAAe,EAAE;IACjB,OAAO;GACT;GAEA,MAAM,QAAQ,GAAG,QAAQ,QAAQ;GACjC,MAAM,QAAQ,GAAG,QAAQ,QAAQ;GAEjC,MAAM,gBAAgB;IACpB,aAAa,OAAO;IACpB,IAAI,cAAc,aAAa,YAAY;IAC3C,MAAM,QAAQ,IAAI,QAAQ,QAAQ;IAClC,MAAM,QAAQ,IAAI,QAAQ,QAAQ;IAClC,MAAM,mBAAmB,OAAO;IAChC,MAAM,mBAAmB,OAAO;GAClC;GAKA,MAAM,UAAU,YAA2D;IACzE,IAAI,SAAS;IACb,UAAU;IACV,QAAQ;IACR,MAAM,SAAS,KAAK,IAAI,IAAI;IAC5B,MAAM,WAAW,kBACb,QAAQ,QAAQ,QAAQ,SAAS,IAC/B,QAAQ,OACR,MACD,QAAQ,QAAQ;IACrB,MAAM,SACJ,CAAC,mBAAmB,UAAU,UAAU,OAAO,aAC3C,OAAO,WAAW,MAAM,QAAQ,QAAQ,QAAQ,IAChD,KAAA;IACN,QAAQ;KACN;KACA;KACA;KACA,QAAQ,QAAQ,cAAc,SAAS,QAAQ,cAAc;KAC7D;KACA,YAAY,QAAQ;KACpB,aAAa,QAAQ;KACrB,iBAAiB,mBAAmB,KAAA;KACpC,iBAAiB,mBAAmB,KAAA;IACtC,CAAC;GACH;GAIA,MAAM,iBAAiB;IACrB,IAAI;KACF,IAAI,QAAQ,aAAa,WAAW,OAAO,MAAM,QAAQ,UACvD,QAAQ,KAAK,CAAC,MAAM,KAAK,SAAS;UAElC,MAAM,KAAK,SAAS;IAExB,SAAS,KAAK;KACZ,QAAQ,KAAK,gDAAgD,GAAG;IAClE;GACF;GAEA,IAAI;GACJ,MAAM,UAAU,iBACR;IACJ,kBAAkB;IAClB,SAAS;IAIT,eAAe,iBAAiB,OAAO,EAAE,MAAM,KAAK,CAAC,GAAG,GAAI;GAC9D,GACA,OAAO,aAAa,KAAK,GAC3B;GAEA,MAAM,GAAG,UAAU,SAAS;IAC1B,IAAI,cAAc,aAAa,YAAY;IAC3C,OAAO,EAAE,KAAK,CAAC;GACjB,CAAC;GACD,MAAM,GAAG,UAAU,QAAQ;IACzB,IAAI,cAAc,aAAa,YAAY;IAC3C,OAAO;KAAE,MAAM;KAAK,aAAa,OAAO,GAAG;IAAE,CAAC;GAChD,CAAC;EACH,CAAC;CACH;AACF;AAqCA,IAAa,iBAAb,MAA4B;CAC1B;CACA,YAAY,SAAwB,IAAI,wBAAwB,GAAG;EACjE,KAAK,SAAS;CAChB;CAEA,MAAM,IAAI,QAAuB,SAAsD;EACrF,MAAM,SAAS,MAAM,QAAQ,QAAQ;GACnC,MAAM,WAAW,KAAK,OAAO,GAAG;GAChC,OAAO,OAAO;GACd,SAAS;IAAC,OAAO;IAAc,OAAO;IAAY,OAAO;GAAW,CAAC,CAClE,OAAO,OAAO,CAAC,CACf,KAAK,MAAM;EAChB,CAAC;EACD,MAAM,SAA+B;GAAE,QAAQ;GAAO,aAAa;GAAG,OAAO;EAAE;EAC/E,IAAI;GACF,IAAI,OAAO,cAAc;IACvB,OAAO,QAAQ,MAAM,KAAK,OAAO,KAAK,SAAS,OAAO,cAAc,MAAM;IAC1E,OAAO,eAAe,OAAO,MAAM;IACnC,IAAI,OAAO,MAAM,mBAAmB,OAAO,MAAM,aAAa,GAAG;KAC/D,MAAM,SAAS,OAAO,MAAM,kBACxB,yBAAyB,OAAO,MAAM,OAAO,MAC7C,sBAAsB,OAAO,MAAM,SAAS;KAChD,MAAM,OAAO,KAAK,QAAQ;MACxB,UAAU,OAAO,MAAM;MACvB,QAAQ,OAAO;KACjB,CAAyB;KACzB,OAAO;IACT;GACF;GACA,IAAI,OAAO,YAAY;IACrB,OAAO,MAAM,MAAM,KAAK,OAAO,KAAK,OAAO,OAAO,YAAY,MAAM;IACpE,OAAO,eAAe,OAAO,IAAI;IACjC,IAAI,OAAO,IAAI,mBAAmB,OAAO,IAAI,aAAa,GAAG;KAC3D,MAAM,SAAS,OAAO,IAAI,kBACtB,uBAAuB,OAAO,IAAI,OAAO,MACzC,oBAAoB,OAAO,IAAI,SAAS;KAC5C,MAAM,OAAO,KAAK,QAAQ;MACxB,UAAU,OAAO,IAAI;MACrB,QAAQ,OAAO;KACjB,CAAyB;KACzB,OAAO;IACT;GACF;GACA,IAAI,OAAO,aAAa;IACtB,OAAO,OAAO,MAAM,KAAK,OAAO,KAAK,QAAQ,OAAO,aAAa,MAAM;IACvE,OAAO,eAAe,OAAO,KAAK;IAGlC,MAAM,SAAS,CAAC,OAAO,KAAK,mBAAmB,OAAO,KAAK,aAAa;IACxE,OAAO,SAAS;IAChB,IAAI,OAAO,KAAK,eAAe,KAAA,KAAa,OAAO,KAAK,aAAa,GACnE,OAAO,SAAS,OAAO,KAAK,eAAe,KAAK,OAAO,KAAK;SAE5D,OAAO,QAAQ,SAAS,IAAI;IAE9B,MAAM,OAAO,IAAI;KACf,UAAU,OAAO,KAAK;KACtB,YAAY,OAAO,KAAK;KACxB,aAAa,OAAO,KAAK;KACzB,QAAQ,OAAO;KACf,QAAQ,SAAS,OAAO;IAC1B,CAAyB;GAC3B,OAAO;IACL,OAAO,SAAS;IAChB,OAAO,QAAQ;IACf,MAAM,OAAO,IAAI,EAAE,QAAQ,OAAO,YAAY,CAAyB;GACzE;EACF,SAAS,KAAK;GACZ,MAAM,OAAO,KAAK,eAAe,QAAQ,MAAM,OAAO,GAAG,CAAC;GAC1D,MAAM;EACR;EACA,OAAO;CACT;AACF;;;AClXA,eAAsB,sBACpB,UACA,OACA,UAAgC,CAAC,GACH;CAC9B,MAAM,UAAU,IAAI,aAAa,KAAK;CACtC,MAAM,QAAQ,SAAS;EACrB,YAAY,SAAS;EACrB,WAAW,QAAQ;EACnB,gBAAgB,SAAS;EACzB,MAAM,SAAS;EACf,GAAG,QAAQ;CACb,CAAC;CAED,MAAM,SAAS,MAAM,IADD,eAAe,QAAQ,MAChB,CAAC,CAAC,IAAI,SAAS,SAAS,OAAO;CAC1D,MAAM,YAAY,SAAS,iBAAiB;CAC5C,MAAM,OAAO,OAAO,UAAU,OAAO,SAAS;CAC9C,MAAM,cAAc,OAAO,UAAU,KAAA,KAAa,OAAO,MAAM,aAAa;CAC5E,MAAM,YAAY,OAAO,QAAQ,KAAA,KAAa,OAAO,IAAI,aAAa;CACtE,MAAM,aAAa,OAAO,SAAS,KAAA,KAAa,OAAO,KAAK,aAAa;CACzE,MAAM,eAAyC,OAC3C,YACA,eAAe,YACb,oBACA,aACE,iBACA;CACR,MAAM,QAAQ,OAAO;EACnB;EACA,OAAO,OAAO;EACd;EACA,OAAO,OAAO,KAAA,IAAY,iBAAiB,MAAM;CACnD,CAAC;CACD,OAAO;EACL,OAAO,QAAQ;EACf;EACA,SAAS;EACT;EACA,OAAO,OAAO;EACd;CACF;AACF;AAEA,SAAS,iBAAiB,QAAsC;CAC9D,IAAI,OAAO,SAAS,OAAO,MAAM,aAAa,GAC5C,OAAO,sBAAsB,OAAO,MAAM;CAC5C,IAAI,OAAO,OAAO,OAAO,IAAI,aAAa,GAAG,OAAO,oBAAoB,OAAO,IAAI;CACnF,IAAI,OAAO,MAAM;EACf,IAAI,OAAO,KAAK,eAAe,KAAA,GAC7B,OAAO,UAAU,OAAO,KAAK,eAAe,EAAE,GAAG,OAAO,KAAK;EAE/D,OAAO,aAAa,OAAO,KAAK;CAClC;CACA,OAAO;AACT;;;ACvDA,IAAa,iBAAb,MAA4B;CAC1B;CACA;CACA;CACA;CACA;CACA;CACA;CAEA,YAAY,OAAmB,MAA0B,QAAwB;EAC/E,KAAK,QAAQ;EACb,KAAK,YAAY,KAAK;EACtB,KAAK,SAAS,KAAK,UAAU,SAAS;EACtC,KAAK,gBAAgB;EACrB,KAAK,iBAAiB,IAAI,aAAa,KAAK;CAC9C;;CAGA,MAAM,UAAU,aAAa,GAAG,KAAK,UAAU,QAAyB;EACtE,MAAM,KAAK,eAAe,SAAS;GACjC;GACA,WAAW,KAAK;GAChB,QAAQ,KAAK;GACb,OAAO;EACT,CAAC;EACD,KAAK,eAAe,KAAK,eAAe;EACxC,OAAO,KAAK;CACd;;CAGA,IAAI,UAAwB;EAC1B,IAAI,CAAC,KAAK,cAAc,MAAM,IAAI,MAAM,gDAAgD;EACxF,OAAO,KAAK;CACd;;;;;CAMA,MAAM,KAAK,SAAgF;EACzF,IAAI,CAAC,KAAK,cAAc,MAAM,IAAI,MAAM,6CAA6C;EACrF,MAAM,eAAe,IAAI,aAAa,KAAK,KAAK;EAChD,MAAM,aAAa,SAAS;GAC1B,YAAY,QAAQ,cAAc,GAAG,KAAK,UAAU;GACpD,WAAW,KAAK;GAChB,QAAQ,KAAK;GACb,aAAa,KAAK;GAClB,OAAO;EACT,CAAC;EAED,MAAM,SAAS,MAAM,IADD,eAAe,QAAQ,UAAU,KAAK,aAC/B,CAAC,CAAC,IAAI,QAAQ,SAAS,YAAY;EAC9D,MAAM,aAAa,OAAO;GACxB,MAAM,OAAO;GACb,OAAO,OAAO;GACd,cAAc,OAAO,SAAS,YAAY;EAC5C,CAAC;EACD,KAAK,iBAAiB,aAAa;EACnC,OAAO;GAAE,OAAO,aAAa;GAAO;EAAO;CAC7C;;;;;;;CAQA,MAAM,eAAe,SAA8D;EACjF,MAAM,cAAc,KAAK,kBAAkB,KAAK;EAChD,IAAI,CAAC,aACH,MAAM,IAAI,MAAM,gEAAgE;EAClF,MAAM,EAAE,UAAU,WAAW;EAC7B,MAAM,SAAS,MAAM,sBAAsB,UAAU,KAAK,OAAO;GAC/D,QAAQ,UAAU,KAAK;GACvB,YAAY;IAAE,SAAS,KAAA;IAAW,WAAW,KAAA;IAAW,kBAAkB,KAAA;GAAU;EACtF,CAAC;EAED,MAAM,KAAK,MAAM,UAAU,OAAO,OAAO;GACvC;GACA,WAAW,KAAK;GAChB,QAAQ,KAAK;GACb,OAAO;EACT,CAAC;EACD,OAAO;CACT;;;CAIA,MAAM,gBAAgB,OAAe,WAAmC;EACtE,IAAI,CAAC,KAAK,cACR,MAAM,IAAI,MAAM,wDAAwD;EAC1E,MAAM,KAAK,eAAe,YAAY;GACpC,SAAS;GACT,cAAc,KAAK;GACnB,WAAW;GACX;GACA;GACA,MAAM;EACR,CAAC;CACH;;CAGA,MAAM,QAAQ,SAA2E;EACvF,MAAM,KAAK,eAAe,OAAO;GAC/B,MAAM,QAAQ;GACd,OAAO,QAAQ;GACf,OAAO,QAAQ;EACjB,CAAC;CACH;;;;;;;;CASA,MAAM,gBAAgB,YAA2C;EAC/D,MAAM,cAAc,KAAK,kBAAkB,KAAK;EAChD,IAAI,CAAC,aACH,MAAM,IAAI,MACR,8EACF;EACF,MAAM,UAAU,IAAI,aAAa,KAAK,KAAK;EAC3C,MAAM,QAAQ,SAAS;GACrB;GACA,WAAW,KAAK;GAChB,QAAQ,KAAK;GACb;GACA,OAAO;EACT,CAAC;EACD,OAAO;CACT;;;;;;CAOA,MAAM,iBAAiB,MAKH;EAClB,IAAI,CAAC,KAAK,cACR,MAAM,IAAI,MAAM,yDAAyD;EAC3E,MAAM,UAAU,IAAI,aAAa,KAAK,KAAK;EAC3C,MAAM,QAAQ,SAAS;GACrB,YAAY,KAAK,cAAc,GAAG,KAAK,UAAU;GACjD,WAAW,KAAK;GAChB,QAAQ,KAAK;GACb,aAAa,KAAK;GAClB,OAAO;EACT,CAAC;EACD,MAAM,QAAQ,OAAO;GACnB,MAAM,KAAK;GACX,OAAO,KAAK;GACZ,cAAc,KAAK,OAAO,YAAY;GACtC,OAAO,KAAK;EACd,CAAC;EACD,KAAK,iBAAiB,QAAQ;EAC9B,OAAO,QAAQ;CACjB;CAEA,IAAI,sBAA0C;EAC5C,OAAO,KAAK;CACd;CACA,IAAI,oBAAwC;EAC1C,OAAO,KAAK;CACd;AACF;;;;;;AAOA,eAAsB,qBACpB,OACA,WAOC;CACD,MAAM,OAAO,MAAM,MAAM,SAAS,EAAE,UAAU,CAAC;CAC/C,MAAM,WAAW,KACd,QAAQ,MAAM,EAAE,UAAU,SAAS,CAAC,CACpC,MAAM,GAAG,MAAM,EAAE,YAAY,EAAE,SAAS;CAC3C,MAAM,YAAY,KACf,QAAQ,MAAM,EAAE,UAAU,WAAW,CAAC,CACtC,MAAM,GAAG,MAAM,EAAE,YAAY,EAAE,SAAS;CAC3C,MAAM,iBAAiB,KACpB,QAAQ,MAAM,EAAE,UAAU,aAAa,CAAC,CACxC,MAAM,GAAG,MAAM,EAAE,YAAY,EAAE,SAAS;CAC3C,OAAO;EACL;EACA;EACA,gBAAgB,SAAS;EACzB,cAAc,UAAU;EACxB,oBAAoB;CACtB;AACF;AAEA,SAAS,WAAmB;CAC1B,IAAI,OAAO,WAAW,QAAQ,eAAe,YAAY,OAAO,WAAW,OAAO,WAAW;CAC7F,OAAO,GAAG,KAAK,IAAI,CAAC,CAAC,SAAS,EAAE,EAAE,GAAG,KAAK,OAAO,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC,MAAM,GAAG,EAAE;AAC7E;;;;;;;;;;;;;;;;;;;AC5NA,SAAgB,gBAAgB,SAAuD;CACrF,MAAM,mBAAmB,QAAQ,QAAQ,MAAM,EAAE,QAAQ,CAAC,CAAC;CAC3D,OAAO;EACL,aAAa,SACX,UACC,MAAM,EAAE,YACR,MAAM,EAAE,UACX;EACA,eAAe,SACb,UACC,MAAM,EAAE,YACR,MAAM,EAAE,YACX;EACA,gBAAgB,SACd,UACC,MAAM,EAAE,aACR,MAAM,EAAE,YACX;EACA;CACF;AACF;AAEA,SAAS,SACP,SACA,GACA,GAC8B;CAC9B,MAAM,KAAe,CAAC;CACtB,MAAM,KAAe,CAAC;CACtB,KAAK,MAAM,KAAK,SAAS;EACvB,MAAM,IAAI,EAAE,CAAC;EACb,MAAM,IAAI,EAAE,CAAC;EACb,IAAI,MAAM,QAAQ,MAAM,QAAQ,OAAO,SAAS,CAAC,KAAK,OAAO,SAAS,CAAC,GAAG;GACxE,GAAG,KAAK,CAAC;GACT,GAAG,KAAK,CAAC;EACX;CACF;CACA,IAAI,GAAG,SAAS,GAAG,OAAO,KAAA;CAC1B,OAAO;EACL,GAAG,GAAG;EACN,SAAS,SAAS,IAAI,EAAE;EACxB,UAAU,UAAU,IAAI,EAAE;CAC5B;AACF;;;ACjCA,IAAa,kBAAb,MAA6B;CACP;CAApB,YAAY,OAA2B;EAAnB,KAAA,QAAA;CAAoB;CAExC,MAAM,eAA0C;EAC9C,MAAM,OAAO,MAAM,KAAK,MAAM,SAAS;EACvC,MAAM,4BAAY,IAAI,IAAmB;EACzC,KAAK,MAAM,KAAK,MAAM;GACpB,IAAI,CAAC,EAAE,WAAW;GAClB,MAAM,MAAM,UAAU,IAAI,EAAE,SAAS,KAAK,CAAC;GAC3C,IAAI,KAAK,CAAC;GACV,UAAU,IAAI,EAAE,WAAW,GAAG;EAChC;EACA,MAAM,YAA8B,CAAC;EACrC,KAAK,MAAM,CAAC,WAAW,gBAAgB,WAAW;GAChD,MAAM,SAAS,YAAY,MAAM,CAAC,CAAC,MAAM,GAAG,MAAM,EAAE,YAAY,EAAE,SAAS;GAC3E,MAAM,QAAQ,YAAY,QAAQ,MAAM,EAAE,UAAU,SAAS;GAC7D,MAAM,SAAS,YAAY,QAAQ,MAAM,EAAE,UAAU,WAAW;GAChE,MAAM,WAAW,YAAY,QAAQ,MAAM,EAAE,UAAU,aAAa;GACpE,MAAM,SAAS,OAAO;GACtB,IAAI,CAAC,QAAQ;GACb,UAAU,KAAK;IACb;IACA,WAAW,MAAM;IACjB,YAAY,OAAO;IACnB,iBAAiB,SAAS;IAC1B,gBAAgB,OAAO;IACvB,cAAc,MAAM,EAAE,EAAE;IACxB,eAAe,OAAO,UAClB;KAAE,MAAM,OAAO,QAAQ,QAAQ;KAAO,OAAO,OAAO,QAAQ;IAAM,IAClE,KAAA;GACN,CAAC;EACH;EACA,OAAO,UAAU,MAAM,GAAG,MAAM,EAAE,iBAAiB,EAAE,cAAc;CACrE;CAEA,MAAM,gBAAgB,WAAoD;EAGxE,QADgB,MADG,KAAK,MAAM,SAAS,EAAE,UAAU,CAAC,EAAA,CAC/B,MAAM,CAAC,CAAC,MAAM,GAAG,MAAM,EAAE,YAAY,EAAE,SAC/C,CAAC,CAAC,KAAK,SAAS;GAC3B;GACA,aACE,IAAI,UAAU,YACV,SACA,IAAI,UAAU,cACZ,UACA,IAAI,UAAU,gBACZ,YACA;EACZ,EAAE;CACJ;CAEA,MAAM,aAAa,WAA2C;EAC5D,MAAM,eAAe,MAAM,KAAK,MAAM,SAAS;GAAE;GAAW,OAAO;EAAU,CAAC,EAAA,CAAG,MAC9E,GAAG,MAAM,EAAE,YAAY,EAAE,SAC5B;EACA,MAAM,cAAc,OAAO,UAAkB,KAAK,MAAM,SAAS,EAAE,aAAa,MAAM,CAAC;EACvF,MAAM,MAAqB,CAAC;EAC5B,KAAK,MAAM,OAAO,aAAa;GAC7B,MAAM,QAAQ,MAAM,KAAK,MAAM,MAAM,EAAE,OAAO,IAAI,MAAM,CAAC;GACzD,MAAM,WAAW,MAAM,YAAY,IAAI,KAAK;GAC5C,MAAM,QAAQ,SAAS,MAAM,MAAM,EAAE,UAAU,WAAW;GAC1D,MAAM,UAAoB,CAAC;GAG3B,IAAI,OAAO;IACT,MAAM,SAAS,MAAM,YAAY,MAAM,KAAK;IAC5C,KAAK,MAAM,KAAK,QAAQ,IAAI,EAAE,UAAU,eAAe,QAAQ,KAAK,EAAE,KAAK;GAC7E;GACA,KAAK,MAAM,KAAK,UAAU,IAAI,EAAE,UAAU,eAAe,QAAQ,KAAK,EAAE,KAAK;GAC7E,IAAI,KAAK;IACP,QAAQ,IAAI,UAAU,IAAI;IAC1B;IACA,cAAc,IAAI;IAClB,WAAW,IAAI;IACf,SAAS,IAAI;IACb,QAAQ,IAAI;IACZ,SAAS,IAAI;IACb,UAAU,MAAM,QAAQ,MAAM,EAAE,SAAS,KAAK,CAAC,CAAC;IAChD,WAAW,MAAM,QAAQ,MAAM,EAAE,SAAS,MAAM,CAAC,CAAC;IAClD,YAAY,OAAO;IACnB,kBAAkB;GACpB,CAAC;EACH;EACA,OAAO;CACT;AACF;;;;;;;;;;;;;;;;;;;;;;;;;;ACzEA,eAAsB,aACpB,OACA,WACkC;CAClC,MAAM,UAAU,MAAM,MAAM,SAAS,EAAE,UAAU,CAAC;CAClD,MAAM,UAAU,cAAc,SAAS,SAAS;CAChD,MAAM,QAAQ,cAAc,SAAS,WAAW;CAChD,MAAM,UAAU,QAAQ,QAAQ,MAAM,EAAE,UAAU,aAAa;CAE/D,MAAM,YAAY,UAAU,MAAM,iBAAiB,OAAO,QAAQ,KAAK,IAAI;CAC3E,MAAM,aAAa,OAAO,SAAS,SAAS;CAC5C,MAAM,gBAAgB,QACnB,KAAK,MAAM,EAAE,SAAS,KAAK,CAAC,CAC5B,QAAQ,MAAmB,OAAO,MAAM,QAAQ;CACnD,MAAM,eACJ,cAAc,SAAS,IACnB,cAAc,QAAQ,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,cAAc,SACzD;CACN,MAAM,gBAAgB,QAAQ,QAAQ,MAAM,EAAE,SAAS,SAAS,IAAI,CAAC,CAAC;CACtE,MAAM,kBAAkB,QAAQ,SAAS,IAAI,gBAAgB,QAAQ,SAAS;CAE9E,MAAM,OAAoB,QAAQ,WAAW,IAAI,kBAAkB;CACnE,MAAM,WACJ,SAAS,kBACL,cAAc,QAAQ,eAAe,OACrC,cAAc,QAAQ,eAAe,QAAQ,iBAAiB;CAEpE,OAAO;EACL;EACA;EACA,cAAc,SAAS;EACvB;EACA,YAAY,OAAO;EACnB;EACA,kBAAkB,QAAQ,KAAK,MAAM,EAAE,KAAK;EAC5C;EACA;EACA;CACF;AACF;;AAGA,eAAsB,iBAAiB,OAAuD;CAC5F,MAAM,OAAO,MAAM,MAAM,SAAS;CAClC,MAAM,aAAa,CAAC,GAAG,IAAI,IAAI,KAAK,KAAK,MAAM,EAAE,SAAS,CAAC,CAAC,QAAQ,MAAmB,CAAC,CAAC,CAAC,CAAC,CAAC;CAC5F,OAAO,QAAQ,IAAI,WAAW,KAAK,MAAM,aAAa,OAAO,CAAC,CAAC,CAAC;AAClE;AAEA,SAAS,cAAc,MAAa,OAAsC;CAExE,OADiB,KAAK,QAAQ,MAAM,EAAE,UAAU,KAAK,CAAC,CAAC,MAAM,GAAG,MAAM,EAAE,YAAY,EAAE,SACxE,CAAC,CAAC;AAClB;AAEA,eAAe,iBAAiB,OAAmB,cAA8C;CAE/F,MAAM,QAAO,MADI,WAAW,OAAO,YAAY,EAAA,CAC/B,MACb,MAAM,EAAE,YAAY,kBAAkB,EAAE,cAAc,0BACzD;CACA,IAAI,CAAC,MAAM,OAAO;CAElB,IAAI,KAAK,SAAS,KAAK,KAAK,SAAS,GAAG,OAAO,KAAK;CACpD,IAAI,KAAK,SAAS,KAAK,KAAK,SAAS,IAAI,OAAO,KAAK,QAAQ;CAC7D,OAAO;AACT"}