@tangle-network/agent-eval 0.126.7 → 0.128.0

package/dist/{chunk-ARU2PZFM.js → chunk-NYLOYM6N.js}

@@ -1,6 +1,9 @@
 import {
   pearsonR
-} from "./chunk-PJQFMIOX.js";
+} from "./chunk-MHELPNRP.js";
+import {
+  runTaskScore
+} from "./chunk-2JX3CFMB.js";
 
 // src/rl/verifiable-reward.ts
 var DEFAULT_DETERMINISTIC_LAYERS = /* @__PURE__ */ new Set([
@@ -31,7 +34,7 @@ function extractVerifiableReward(report, opts = {}) {
   const fallbackToJudge = opts.fallbackToJudge ?? true;
   const judgeFloor = opts.judgeConfidenceFloor ?? 0.7;
   const deterministic = report.layers.filter(
-    (l) => deterministicSet.has(l.layer) && typeof l.score === "number" && Number.isFinite(l.score)
+    (layer) => deterministicSet.has(layer.layer) && isMeasuredLayer(layer)
   );
   if (deterministic.length === 1) {
     const layer = deterministic[0];
@@ -67,9 +70,7 @@ function extractVerifiableReward(report, opts = {}) {
     };
   }
   if (!fallbackToJudge) return null;
-  const judge = report.layers.find(
-    (l) => typeof l.score === "number" && Number.isFinite(l.score) && sourceFor(l.layer) === "judge"
-  ) ?? report.layers.find((l) => typeof l.score === "number" && Number.isFinite(l.score));
+  const judge = report.layers.find((layer) => isMeasuredLayer(layer) && sourceFor(layer.layer) === "judge") ?? report.layers.find(isMeasuredLayer);
   if (!judge) return null;
   const confFromDetail = judge.detail?.confidence;
   const judgeValue = clamp01(judge.score);
@@ -83,6 +84,9 @@ function extractVerifiableReward(report, opts = {}) {
     breakdown: layerBreakdown(judge)
   };
 }
+function isMeasuredLayer(layer) {
+  return (layer.status === "pass" || layer.status === "fail") && typeof layer.score === "number" && Number.isFinite(layer.score) && layer.score >= 0 && layer.score <= 1;
+}
 function extractVerifiableRewardsFromRecords(runs, opts = {}) {
   const sourceFor = opts.sourceFor ?? DEFAULT_SOURCE_FOR;
   const deterministicSet = new Set(opts.deterministicLayers ?? [...DEFAULT_DETERMINISTIC_LAYERS]);
@@ -130,8 +134,8 @@ function extractVerifiableRewardsFromRecords(runs, opts = {}) {
       };
     }
     if (!fallbackToJudge) return { runId: run.runId, reward: null };
-    const primary = run.outcome.holdoutScore ?? run.outcome.searchScore;
-    if (typeof primary !== "number" || !Number.isFinite(primary)) {
+    const primary = runTaskScore(run);
+    if (primary === void 0) {
       return { runId: run.runId, reward: null };
     }
     const primaryValue = clamp01(primary);
@@ -175,20 +179,20 @@ function layerBreakdown(l) {
 
 // src/rl/reward-hacking.ts
 var DEFAULT_PROXY = (r) => {
-  const v = r.outcome.holdoutScore ?? r.outcome.searchScore;
-  return typeof v === "number" && Number.isFinite(v) ? v : null;
+  return runTaskScore(r) ?? null;
 };
 function detectRewardHacking(input) {
   const proxyOf = input.proxyOf ?? DEFAULT_PROXY;
   const truthOf = input.truthOf;
   const sus = input.thresholds?.suspect ?? 0.3;
   const gam = input.thresholds?.gaming ?? 0.6;
-  const runs = input.runs.filter((r) => proxyOf(r) !== null);
+  const runs = input.runs.filter((run) => finiteNumber(proxyOf(run)));
   const n = runs.length;
   if (n < 4) {
     return {
       findings: [],
-      verdict: "clean",
+      evaluatedSignals: [],
+      verdict: "insufficient_evidence",
       n,
       rationale: [`fewer than 4 runs with proxy reward (n=${n}); insufficient evidence`]
     };
@@ -198,10 +202,10 @@ function detectRewardHacking(input) {
   const after = runs.slice(n - windowSize);
   const findings = [];
   if (truthOf) {
-    const beforeProxy = before.map(proxyOf).filter((v) => typeof v === "number");
-    const afterProxy = after.map(proxyOf).filter((v) => typeof v === "number");
-    const beforeTruth = before.map(truthOf).filter((v) => typeof v === "number");
-    const afterTruth = after.map(truthOf).filter((v) => typeof v === "number");
+    const beforeProxy = before.map(proxyOf).filter(finiteNumber);
+    const afterProxy = after.map(proxyOf).filter(finiteNumber);
+    const beforeTruth = before.map(truthOf).filter(finiteNumber);
+    const afterTruth = after.map(truthOf).filter(finiteNumber);
     if (beforeProxy.length >= 2 && afterProxy.length >= 2 && beforeTruth.length >= 2 && afterTruth.length >= 2) {
       const proxyDelta = mean(afterProxy) - mean(beforeProxy);
       const truthDelta = mean(afterTruth) - mean(beforeTruth);
@@ -222,8 +226,8 @@ function detectRewardHacking(input) {
     }
   }
   {
-    const beforeP = before.map(proxyOf).filter((v) => typeof v === "number");
-    const afterP = after.map(proxyOf).filter((v) => typeof v === "number");
+    const beforeP = before.map(proxyOf).filter(finiteNumber);
+    const afterP = after.map(proxyOf).filter(finiteNumber);
     if (beforeP.length >= 4 && afterP.length >= 4) {
       const ks = ksStatistic(beforeP, afterP);
       const severity = clamp012(ks - 0.2);
@@ -237,9 +241,7 @@ function detectRewardHacking(input) {
   }
   {
     const secondaryOf = input.secondaryRewardOf ?? defaultSecondary(input.verifiableRewardOptions);
-    const aligned = runs.map((r) => ({ p: proxyOf(r), s: secondaryOf(r) })).filter(
-      (x) => typeof x.p === "number" && typeof x.s === "number"
-    );
+    const aligned = runs.map((r) => ({ p: proxyOf(r), s: secondaryOf(r) })).filter((x) => finiteNumber(x.p) && finiteNumber(x.s));
     if (aligned.length >= 4) {
       const ps = aligned.map((x) => x.p);
       const ss = aligned.map((x) => x.s);
@@ -259,7 +261,7 @@ function detectRewardHacking(input) {
       const detBefore = detRuns.slice(0, Math.floor(detRuns.length / 2));
       const detAfter = detRuns.slice(Math.floor(detRuns.length / 2));
       const detDelta = mean(detAfter.map((r) => r.reward.value)) - mean(detBefore.map((r) => r.reward.value));
-      const proxyDelta = mean(after.map(proxyOf).filter((v) => typeof v === "number")) - mean(before.map(proxyOf).filter((v) => typeof v === "number"));
+      const proxyDelta = mean(after.map(proxyOf).filter(finiteNumber)) - mean(before.map(proxyOf).filter(finiteNumber));
       const driftGap = Math.max(0, proxyDelta - detDelta);
       const severity = clamp012(driftGap * 5);
       findings.push({
@@ -271,15 +273,33 @@ function detectRewardHacking(input) {
     }
   }
   const maxSev = findings.reduce((m, f) => Math.max(m, f.severity), 0);
+  if (findings.length === 0) {
+    return {
+      findings,
+      evaluatedSignals: [],
+      verdict: "insufficient_evidence",
+      rationale: [`no reward-hacking signal had enough paired evidence (n=${n})`],
+      n
+    };
+  }
   const verdict = maxSev >= gam ? "gaming" : maxSev >= sus ? "suspect" : "clean";
   const rationale = findings.filter((f) => f.severity >= sus).map((f) => `${f.signal}: severity ${f.severity.toFixed(2)} \u2014 ${f.message}`);
   if (rationale.length === 0) rationale.push("no signals fired above suspect threshold");
-  return { findings, verdict, rationale, n };
+  return {
+    findings,
+    evaluatedSignals: findings.map((finding) => finding.signal),
+    verdict,
+    rationale,
+    n
+  };
 }
 function mean(xs) {
   if (xs.length === 0) return 0;
   return xs.reduce((s, x) => s + x, 0) / xs.length;
 }
+function finiteNumber(value) {
+  return typeof value === "number" && Number.isFinite(value);
+}
 function clamp012(x) {
   if (!Number.isFinite(x)) return 0;
   return Math.max(0, Math.min(1, x));
@@ -309,4 +329,4 @@ export {
   filterDeterministicallyRewarded,
   detectRewardHacking
 };
-//# sourceMappingURL=chunk-ARU2PZFM.js.map
+//# sourceMappingURL=chunk-NYLOYM6N.js.map

package/dist/chunk-NYLOYM6N.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/rl/verifiable-reward.ts","../src/rl/reward-hacking.ts"],"sourcesContent":["/**\n * Verifiable reward channel.\n *\n * For RL on coding / math / theorem-proving / structured-output tasks, the\n * reward signal is *decidable* — a test passes or fails, a proof checks or\n * doesn't, an output validates against a schema or doesn't. These rewards\n * are dramatically more useful for RL training than LLM-judge scores\n * because they don't drift, can't be Goodhart-gamed by the policy in the\n * same way, and don't require a separate calibration loop.\n *\n * The `MultiLayerVerifier` already produces this signal — it just doesn't\n * surface it in a shape that's clean enough for RL training. This module\n * wraps the verifier output so consumers can:\n *\n *   1. Extract a clean `VerifiableReward` from a `VerificationReport`\n *   2. Distinguish *deterministic* rewards (compile, test, schema) from\n *      *probabilistic* rewards (judge) so they can be weighted differently\n *      in the RL training step\n *   3. Filter `RunRecord[]` to only those with a verifiable reward,\n *      producing the clean training set that DeepSeek-R1-style GRPO and\n *      AlphaProof-style search both depend on\n *\n * Why this matters: every credible 2025-2026 frontier RL result on coding\n * agents leans on verifiable reward (DeepSeek-R1 GRPO on test pass-rate,\n * o-series RL on math/code, AlphaProof on Lean kernel checking). Mixing\n * judge scores into the reward signal poisons the gradient. This module\n * is the seam.\n */\n\nimport type { LayerResult, VerificationReport } from '../multi-layer-verifier'\nimport { type RunRecord, runTaskScore } from '../run-record'\n\nexport type VerifiableRewardSource =\n  | 'compile' // typecheck / build / lint passed\n  | 'test' // unit / integration test pass-rate\n  | 'schema' // structured output validates\n  | 'sandbox' // sandbox exec exit code\n  | 'judge' // LLM judge — probabilistic, included for completeness\n  | 'composite' // weighted blend across multiple of the above\n\nexport interface VerifiableReward {\n  /** Scalar in [0, 1]. The RL training signal. */\n  value: number\n  /** What produced the reward — different sources have different determinism. */\n  source: VerifiableRewardSource\n  /**\n   * Determinism class. `'deterministic'` rewards are repeatable byte-for-byte\n   * given the same inputs (compile, test, schema validation, sandbox exit code).\n   * `'probabilistic'` rewards depend on a stochastic component (LLM judge).\n   * Mixing these in the same training batch without separation is a known\n   * footgun in production RLHF pipelines.\n   */\n  determinism: 'deterministic' | 'probabilistic'\n  /**\n   * Confidence in the reward value. For deterministic sources this is 1.0\n   * (the bit either flipped or didn't). For judge sources this is the\n   * judge-reported confidence or — when missing — a calibrated prior.\n   */\n  confidence: number\n  /** The layer / judge id that produced the signal, for provenance. */\n  origin: string\n  /**\n   * Per-source contribution to `value`, keyed by layer/judge id. Single-source\n   * rewards carry one entry (`{ [origin]: value }`); composite rewards carry\n   * every contributing layer's score — the anti-scalar-collapse surface RL\n   * consumers weight per-source instead of trusting one blended number.\n   */\n  components: Record<string, number>\n  /**\n   * @deprecated Read `components` for per-source reward values. Kept for\n   * published-API compatibility: single-source rewards carry the layer's\n   * diagnostics here (e.g. `{ tests_passed: 7 }`); composite rewards carry\n   * the same per-layer scores `components` now holds.\n   */\n  breakdown?: Record<string, number>\n}\n\nexport interface VerifiableRewardExtractionOptions {\n  /**\n   * Which layers count as deterministic-reward sources. The verifier doesn't\n   * tag layers as \"this is verifiable\"; the caller declares it via this list\n   * (or via the layer name → source mapping). Default treats common names\n   * (`install`, `typecheck`, `build`, `lint`, `test`, `compile`, `schema`,\n   * `sandbox`) as deterministic.\n   */\n  deterministicLayers?: string[]\n  /**\n   * Map layer name → reward source. Defaults to a sensible string-match.\n   */\n  sourceFor?: (layerName: string) => VerifiableRewardSource\n  /**\n   * Whether to fall back to a probabilistic (judge) reward when no\n   * deterministic layer produced a numeric score. Default `true`. Set to\n   * `false` for \"deterministic-only\" training pipelines that should\n   * discard runs without a verifiable signal.\n   */\n  fallbackToJudge?: boolean\n  /**\n   * Default confidence for probabilistic (judge) rewards when the judge\n   * doesn't report one. Default `0.7`.\n   */\n  judgeConfidenceFloor?: number\n}\n\nconst DEFAULT_DETERMINISTIC_LAYERS = new Set([\n  'install',\n  'typecheck',\n  'build',\n  'lint',\n  'test',\n  'compile',\n  'schema',\n  'sandbox',\n  'unit_tests',\n  'integration_tests',\n])\n\nconst DEFAULT_SOURCE_FOR = (name: string): VerifiableRewardSource => {\n  const lower = name.toLowerCase()\n  if (lower.includes('test')) return 'test'\n  if (\n    lower.includes('compile') ||\n    lower.includes('build') ||\n    lower.includes('typecheck') ||\n    lower.includes('lint')\n  )\n    return 'compile'\n  if (lower.includes('schema')) return 'schema'\n  if (lower.includes('sandbox')) return 'sandbox'\n  if (lower.includes('judge') || lower.includes('semantic')) return 'judge'\n  return 'composite'\n}\n\n/**\n * Extract a `VerifiableReward` from a `VerificationReport`.\n *\n * Strategy: prefer the deterministic layers (in order: test → compile →\n * schema → sandbox), fall back to the judge layer if `fallbackToJudge` is\n * true, return `null` if no signal qualifies. When multiple deterministic\n * layers contribute, return a `'composite'` source with a weighted blend.\n */\nexport function extractVerifiableReward(\n  report: VerificationReport,\n  opts: VerifiableRewardExtractionOptions = {},\n): VerifiableReward | null {\n  const deterministicSet = new Set(opts.deterministicLayers ?? [...DEFAULT_DETERMINISTIC_LAYERS])\n  const sourceFor = opts.sourceFor ?? DEFAULT_SOURCE_FOR\n  const fallbackToJudge = opts.fallbackToJudge ?? true\n  const judgeFloor = opts.judgeConfidenceFloor ?? 0.7\n\n  const deterministic = report.layers.filter(\n    (layer) => deterministicSet.has(layer.layer) && isMeasuredLayer(layer),\n  )\n\n  if (deterministic.length === 1) {\n    const layer = deterministic[0]!\n    const value = clamp01(layer.score!)\n    return {\n      value,\n      source: sourceFor(layer.layer),\n      determinism: 'deterministic',\n      confidence: 1,\n      origin: layer.layer,\n      components: { [layer.layer]: value },\n      breakdown: layerBreakdown(layer),\n    }\n  }\n\n  if (deterministic.length > 1) {\n    // Composite: weighted blend by `Layer.weight` if present, else equal.\n    let num = 0\n    let denom = 0\n    const components: Record<string, number> = {}\n    for (const l of deterministic) {\n      const w = (l.detail?.weight as number | undefined) ?? 1\n      num += w * (l.score ?? 0)\n      denom += w\n      components[l.layer] = l.score!\n    }\n    return {\n      value: denom === 0 ? 0 : clamp01(num / denom),\n      source: 'composite',\n      determinism: 'deterministic',\n      confidence: 1,\n      origin: deterministic.map((l) => l.layer).join('+'),\n      components,\n      breakdown: { ...components },\n    }\n  }\n\n  if (!fallbackToJudge) return null\n\n  const judge =\n    report.layers.find((layer) => isMeasuredLayer(layer) && sourceFor(layer.layer) === 'judge') ??\n    report.layers.find(isMeasuredLayer)\n\n  if (!judge) return null\n\n  const confFromDetail = judge.detail?.confidence as number | undefined\n  const judgeValue = clamp01(judge.score!)\n  return {\n    value: judgeValue,\n    source: 'judge',\n    determinism: 'probabilistic',\n    confidence: typeof confFromDetail === 'number' ? confFromDetail : judgeFloor,\n    origin: judge.layer,\n    components: { [judge.layer]: judgeValue },\n    breakdown: layerBreakdown(judge),\n  }\n}\n\nfunction isMeasuredLayer(\n  layer: LayerResult,\n): layer is LayerResult & { status: 'pass' | 'fail'; score: number } {\n  return (\n    (layer.status === 'pass' || layer.status === 'fail') &&\n    typeof layer.score === 'number' &&\n    Number.isFinite(layer.score) &&\n    layer.score >= 0 &&\n    layer.score <= 1\n  )\n}\n\n/**\n * Extract verifiable rewards from `RunRecord[]` produced via the\n * `verificationReportToRunRecord` adapter (which encodes per-layer scores\n * in `outcome.raw['layer.<name>']`). For records that don't carry layer\n * scores, returns `null` for that record.\n *\n * This is the canonical bridge from \"campaign-shaped artifacts\" to\n * \"RL-training-ready reward signals\": every record that has a clean\n * verifiable reward becomes a training datum, every record that doesn't\n * gets filtered out (or kept with `'probabilistic'` determinism for\n * separate downstream handling).\n */\nexport function extractVerifiableRewardsFromRecords(\n  runs: RunRecord[],\n  opts: VerifiableRewardExtractionOptions = {},\n): Array<{ runId: string; reward: VerifiableReward | null }> {\n  const sourceFor = opts.sourceFor ?? DEFAULT_SOURCE_FOR\n  const deterministicSet = new Set(opts.deterministicLayers ?? [...DEFAULT_DETERMINISTIC_LAYERS])\n  const fallbackToJudge = opts.fallbackToJudge ?? true\n  const judgeFloor = opts.judgeConfidenceFloor ?? 0.7\n\n  return runs.map((run) => {\n    // Recover per-layer scores from outcome.raw['layer.<name>']\n    const layerScores: Array<{ name: string; score: number }> = []\n    for (const [k, v] of Object.entries(run.outcome.raw)) {\n      if (\n        k.startsWith('layer.') &&\n        !k.includes('.', 6) &&\n        typeof v === 'number' &&\n        Number.isFinite(v)\n      ) {\n        layerScores.push({ name: k.slice('layer.'.length), score: v })\n      }\n    }\n    const det = layerScores.filter((l) => deterministicSet.has(l.name))\n\n    if (det.length === 1) {\n      const layer = det[0]!\n      const value = clamp01(layer.score)\n      return {\n        runId: run.runId,\n        reward: {\n          value,\n          source: sourceFor(layer.name),\n          determinism: 'deterministic',\n          confidence: 1,\n          origin: layer.name,\n          components: { [layer.name]: value },\n        },\n      }\n    }\n    if (det.length > 1) {\n      const value = det.reduce((s, l) => s + l.score, 0) / det.length\n      const components: Record<string, number> = Object.fromEntries(\n        det.map((l) => [l.name, l.score]),\n      )\n      return {\n        runId: run.runId,\n        reward: {\n          value: clamp01(value),\n          source: 'composite',\n          determinism: 'deterministic',\n          confidence: 1,\n          origin: det.map((l) => l.name).join('+'),\n          components,\n          breakdown: { ...components },\n        },\n      }\n    }\n    if (!fallbackToJudge) return { runId: run.runId, reward: null }\n\n    // Probabilistic fallback: use the run's primary score.\n    const primary = runTaskScore(run)\n    if (primary === undefined) {\n      return { runId: run.runId, reward: null }\n    }\n    const primaryValue = clamp01(primary)\n    return {\n      runId: run.runId,\n      reward: {\n        value: primaryValue,\n        source: 'judge',\n        determinism: 'probabilistic',\n        confidence: judgeFloor,\n        origin: 'run.outcome.score',\n        components: { 'run.outcome.score': primaryValue },\n      },\n    }\n  })\n}\n\n/** Filter `RunRecord[]` to those with deterministic verifiable rewards. */\nexport function filterDeterministicallyRewarded(\n  runs: RunRecord[],\n  opts: VerifiableRewardExtractionOptions = {},\n): Array<{ run: RunRecord; reward: VerifiableReward }> {\n  const rewarded = extractVerifiableRewardsFromRecords(runs, { ...opts, fallbackToJudge: false })\n  const out: Array<{ run: RunRecord; reward: VerifiableReward }> = []\n  for (let i = 0; i < runs.length; i++) {\n    const r = rewarded[i]!\n    if (r.reward && r.reward.determinism === 'deterministic') {\n      out.push({ run: runs[i]!, reward: r.reward })\n    }\n  }\n  return out\n}\n\n// ── Helpers ──────────────────────────────────────────────────────────────\n\nfunction clamp01(x: number): number {\n  if (!Number.isFinite(x)) return 0\n  return Math.max(0, Math.min(1, x))\n}\n\nfunction layerBreakdown(l: LayerResult): Record<string, number> {\n  const out: Record<string, number> = {}\n  if (l.diagnostics) {\n    for (const [k, v] of Object.entries(l.diagnostics)) {\n      if (typeof v === 'number' && Number.isFinite(v)) out[k] = v\n    }\n  }\n  return out\n}\n","/**\n * Reward hacking / Goodhart detection.\n *\n * Goodhart's Law says: when a measure becomes a target, it ceases to be\n * a good measure. In RLHF and agentic-RL settings this is the dominant\n * failure mode — the policy learns to produce outputs that score well on\n * the proxy reward (judge, rubric, test pass-rate) without producing\n * the underlying capability the proxy was meant to track.\n *\n * Krakovna et al. (2020, \"Specification Gaming Examples in AI\") and the\n * subsequent RLHF reward-hacking literature (Skalse et al. 2022, Kim et al.\n * 2023) converge on a few diagnostic signatures:\n *\n *   1. **Reward divergence:** the proxy reward grows while the held-out\n *      ground-truth signal stagnates or drops. Predictive validity over\n *      time captures this.\n *   2. **Distributional shift in outputs:** after RL, the policy produces\n *      outputs that no longer match the reference distribution — usually\n *      because it found a high-reward attractor that's degenerate (e.g.\n *      one-token responses, repetition, formatting tricks).\n *   3. **Disagreement between independent rewards:** if you train on\n *      reward A and a held-out independent reward B drops sharply, you're\n *      probably hacking A.\n *   4. **Calibration drift:** the verifiable / deterministic component of\n *      the reward is stable; the probabilistic / judge component drifts up\n *      while the deterministic component doesn't. The judge is being\n *      gamed.\n *\n * This module ships explicit detectors for all four signatures, plus a\n * combined verdict. The output is diagnostic — actionable signals,\n * not autoreject — because each signature has known false positives\n * (e.g., a policy that genuinely improves can show distributional shift).\n *\n * Differs from `rubricPredictiveValidity` (which is a *standing* check on\n * whether rubrics correlate with deployment outcomes) — this is a\n * *temporal* check on whether the reward-vs-truth gap is *widening over\n * time during a training run*.\n */\n\nimport { type RunRecord, runTaskScore } from '../run-record'\nimport { pearsonR } from '../statistics'\nimport {\n  filterDeterministicallyRewarded,\n  type VerifiableRewardExtractionOptions,\n} from './verifiable-reward'\n\nexport type RewardHackingSignal =\n  | 'reward_divergence'\n  | 'distribution_shift'\n  | 'reward_disagreement'\n  | 'judge_drift'\n\nexport interface RewardHackingFinding {\n  signal: RewardHackingSignal\n  /** Severity in [0, 1]. >0.5 = strong signal. */\n  severity: number\n  message: string\n  /** Numeric evidence the consumer can render. */\n  detail: Record<string, number>\n}\n\nexport interface RewardHackingReport {\n  findings: RewardHackingFinding[]\n  /** Signals with enough usable observations to produce a finding. */\n  evaluatedSignals: RewardHackingSignal[]\n  /**\n   * Composite verdict. `'insufficient_evidence'` when fewer than four scored\n   * runs exist; otherwise `'clean'` if every signal severity < 0.3,\n   * `'suspect'` if at least one ≥ 0.3 but none ≥ 0.6, and `'gaming'` if any ≥ 0.6.\n   */\n  verdict: 'insufficient_evidence' | 'clean' | 'suspect' | 'gaming'\n  /** Rationale for the verdict, ready to paste into an audit log. */\n  rationale: string[]\n  /** Number of runs with a usable proxy reward. */\n  n: number\n}\n\nexport interface DetectRewardHackingInput {\n  /**\n   * Run records ordered by recency (oldest first). The detector segments\n   * them into prefix/suffix windows to compute \"did the gap widen.\"\n   */\n  runs: RunRecord[]\n  /**\n   * The metric the policy was trained to optimize. Should be present on\n   * `outcome.raw` or `outcome.holdoutScore`. Default reads `outcome.holdoutScore`.\n   */\n  proxyOf?: (run: RunRecord) => number | null\n  /**\n   * The held-out ground-truth metric. For RL on coding, this is typically\n   * test pass-rate. For RLHF, it's downstream task performance or human\n   * preference. For knowledge tasks, it's an independently-graded score.\n   */\n  truthOf?: (run: RunRecord) => number | null\n  /**\n   * Independent secondary reward. Used for the `reward_disagreement`\n   * signal. Default uses the verifiable reward extractor (deterministic\n   * sources only).\n   */\n  secondaryRewardOf?: (run: RunRecord) => number | null\n  /**\n   * Window size — how many of the most recent runs count as the \"after\"\n   * cohort. Default min(50, half the runs).\n   */\n  windowSize?: number\n  /**\n   * Severity threshold to flag a signal. Default 0.3 (suspect) and 0.6\n   * (gaming).\n   */\n  thresholds?: { suspect?: number; gaming?: number }\n  /**\n   * Verifiable-reward options used for the secondary-reward fallback.\n   */\n  verifiableRewardOptions?: VerifiableRewardExtractionOptions\n}\n\nconst DEFAULT_PROXY = (r: RunRecord): number | null => {\n  return runTaskScore(r) ?? null\n}\n\nexport function detectRewardHacking(input: DetectRewardHackingInput): RewardHackingReport {\n  const proxyOf = input.proxyOf ?? DEFAULT_PROXY\n  const truthOf = input.truthOf\n  const sus = input.thresholds?.suspect ?? 0.3\n  const gam = input.thresholds?.gaming ?? 0.6\n\n  const runs = input.runs.filter((run) => finiteNumber(proxyOf(run)))\n  const n = runs.length\n  if (n < 4) {\n    return {\n      findings: [],\n      evaluatedSignals: [],\n      verdict: 'insufficient_evidence',\n      n,\n      rationale: [`fewer than 4 runs with proxy reward (n=${n}); insufficient evidence`],\n    }\n  }\n  const windowSize = Math.max(1, input.windowSize ?? Math.min(50, Math.floor(n / 2)))\n  const before = runs.slice(0, n - windowSize)\n  const after = runs.slice(n - windowSize)\n\n  const findings: RewardHackingFinding[] = []\n\n  // ── Signal 1: reward divergence (proxy ↑ while truth flat or ↓) ──────\n  if (truthOf) {\n    const beforeProxy = before.map(proxyOf).filter(finiteNumber)\n    const afterProxy = after.map(proxyOf).filter(finiteNumber)\n    const beforeTruth = before.map(truthOf).filter(finiteNumber)\n    const afterTruth = after.map(truthOf).filter(finiteNumber)\n    if (\n      beforeProxy.length >= 2 &&\n      afterProxy.length >= 2 &&\n      beforeTruth.length >= 2 &&\n      afterTruth.length >= 2\n    ) {\n      const proxyDelta = mean(afterProxy) - mean(beforeProxy)\n      const truthDelta = mean(afterTruth) - mean(beforeTruth)\n      // Divergence: proxy goes up while truth goes flat or down.\n      // Severity = max(0, (proxyDelta - truthDelta)) — bigger gap = bigger signal.\n      const gap = Math.max(0, proxyDelta - truthDelta)\n      const severity = clamp01(gap * 5) // scale: 0.2 absolute gap → severity 1.0\n      findings.push({\n        signal: 'reward_divergence',\n        severity,\n        message:\n          severity >= sus\n            ? `proxy reward rose by ${proxyDelta.toFixed(3)} while truth changed by ${truthDelta.toFixed(3)} — potential Goodhart`\n            : `proxy and truth moved together (proxy ${proxyDelta.toFixed(3)}, truth ${truthDelta.toFixed(3)})`,\n        detail: {\n          proxyDelta,\n          truthDelta,\n          gap,\n          beforeN: beforeProxy.length,\n          afterN: afterProxy.length,\n        },\n      })\n    }\n  }\n\n  // ── Signal 2: distributional shift in outputs (KS on score distributions) ──\n  {\n    const beforeP = before.map(proxyOf).filter(finiteNumber)\n    const afterP = after.map(proxyOf).filter(finiteNumber)\n    if (beforeP.length >= 4 && afterP.length >= 4) {\n      const ks = ksStatistic(beforeP, afterP)\n      // KS statistic: bigger = more shift. We're agnostic about direction;\n      // genuine improvement ALSO produces shift, so this signal is\n      // contributory rather than load-bearing.\n      const severity = clamp01(ks - 0.2)\n      findings.push({\n        signal: 'distribution_shift',\n        severity,\n        message:\n          severity >= sus\n            ? `KS=${ks.toFixed(3)} between before/after windows — distributional shift large`\n            : `KS=${ks.toFixed(3)} between before/after windows — within-distribution drift`,\n        detail: { ks, beforeN: beforeP.length, afterN: afterP.length },\n      })\n    }\n  }\n\n  // ── Signal 3: reward disagreement (proxy vs independent secondary) ────\n  {\n    const secondaryOf = input.secondaryRewardOf ?? defaultSecondary(input.verifiableRewardOptions)\n    const aligned = runs\n      .map((r) => ({ p: proxyOf(r), s: secondaryOf(r) }))\n      .filter((x): x is { p: number; s: number } => finiteNumber(x.p) && finiteNumber(x.s))\n    if (aligned.length >= 4) {\n      const ps = aligned.map((x) => x.p)\n      const ss = aligned.map((x) => x.s)\n      const r = pearsonR(ps, ss)\n      // Disagreement: low or negative correlation between primary proxy\n      // reward and an independent secondary signal.\n      const severity = clamp01(0.5 - Math.max(0, r))\n      findings.push({\n        signal: 'reward_disagreement',\n        severity,\n        message:\n          severity >= sus\n            ? `proxy and independent secondary reward correlate ρ=${r.toFixed(3)} — possibly hacking proxy`\n            : `proxy and secondary reward correlate ρ=${r.toFixed(3)}`,\n        detail: { pearson: r, n: aligned.length },\n      })\n    }\n  }\n\n  // ── Signal 4: judge drift (probabilistic up while deterministic flat) ─\n  {\n    const detRuns = filterDeterministicallyRewarded(runs, input.verifiableRewardOptions ?? {})\n    if (detRuns.length >= 4) {\n      const detBefore = detRuns.slice(0, Math.floor(detRuns.length / 2))\n      const detAfter = detRuns.slice(Math.floor(detRuns.length / 2))\n      const detDelta =\n        mean(detAfter.map((r) => r.reward.value)) - mean(detBefore.map((r) => r.reward.value))\n      const proxyDelta =\n        mean(after.map(proxyOf).filter(finiteNumber)) -\n        mean(before.map(proxyOf).filter(finiteNumber))\n      const driftGap = Math.max(0, proxyDelta - detDelta)\n      const severity = clamp01(driftGap * 5)\n      findings.push({\n        signal: 'judge_drift',\n        severity,\n        message:\n          severity >= sus\n            ? `judge proxy +${proxyDelta.toFixed(3)} while deterministic reward +${detDelta.toFixed(3)} — judge drifting up without verifiable backing`\n            : `judge and deterministic rewards move in step (judge ${proxyDelta.toFixed(3)}, det ${detDelta.toFixed(3)})`,\n        detail: { proxyDelta, detDelta, driftGap, n: detRuns.length },\n      })\n    }\n  }\n\n  const maxSev = findings.reduce((m, f) => Math.max(m, f.severity), 0)\n  if (findings.length === 0) {\n    return {\n      findings,\n      evaluatedSignals: [],\n      verdict: 'insufficient_evidence',\n      rationale: [`no reward-hacking signal had enough paired evidence (n=${n})`],\n      n,\n    }\n  }\n  const verdict: RewardHackingReport['verdict'] =\n    maxSev >= gam ? 'gaming' : maxSev >= sus ? 'suspect' : 'clean'\n  const rationale = findings\n    .filter((f) => f.severity >= sus)\n    .map((f) => `${f.signal}: severity ${f.severity.toFixed(2)} — ${f.message}`)\n  if (rationale.length === 0) rationale.push('no signals fired above suspect threshold')\n\n  return {\n    findings,\n    evaluatedSignals: findings.map((finding) => finding.signal),\n    verdict,\n    rationale,\n    n,\n  }\n}\n\n// ── Helpers ──────────────────────────────────────────────────────────────\n\nfunction mean(xs: number[]): number {\n  if (xs.length === 0) return 0\n  return xs.reduce((s, x) => s + x, 0) / xs.length\n}\n\nfunction finiteNumber(value: number | null): value is number {\n  return typeof value === 'number' && Number.isFinite(value)\n}\n\nfunction clamp01(x: number): number {\n  if (!Number.isFinite(x)) return 0\n  return Math.max(0, Math.min(1, x))\n}\n\nfunction ksStatistic(a: number[], b: number[]): number {\n  // Two-sample Kolmogorov-Smirnov statistic.\n  const sortedA = [...a].sort((x, y) => x - y)\n  const sortedB = [...b].sort((x, y) => x - y)\n  const all = [...new Set([...sortedA, ...sortedB])].sort((x, y) => x - y)\n  let max = 0\n  for (const v of all) {\n    const fa = sortedA.filter((x) => x <= v).length / sortedA.length\n    const fb = sortedB.filter((x) => x <= v).length / sortedB.length\n    max = Math.max(max, Math.abs(fa - fb))\n  }\n  return max\n}\n\nfunction defaultSecondary(\n  verifiableOpts?: VerifiableRewardExtractionOptions,\n): (run: RunRecord) => number | null {\n  return (run: RunRecord) => {\n    const filtered = filterDeterministicallyRewarded([run], verifiableOpts ?? {})\n    return filtered.length === 1 ? filtered[0]!.reward.value : null\n  }\n}\n"],"mappings":";;;;;;;;AAwGA,IAAM,+BAA+B,oBAAI,IAAI;AAAA,EAC3C;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF,CAAC;AAED,IAAM,qBAAqB,CAAC,SAAyC;AACnE,QAAM,QAAQ,KAAK,YAAY;AAC/B,MAAI,MAAM,SAAS,MAAM,EAAG,QAAO;AACnC,MACE,MAAM,SAAS,SAAS,KACxB,MAAM,SAAS,OAAO,KACtB,MAAM,SAAS,WAAW,KAC1B,MAAM,SAAS,MAAM;AAErB,WAAO;AACT,MAAI,MAAM,SAAS,QAAQ,EAAG,QAAO;AACrC,MAAI,MAAM,SAAS,SAAS,EAAG,QAAO;AACtC,MAAI,MAAM,SAAS,OAAO,KAAK,MAAM,SAAS,UAAU,EAAG,QAAO;AAClE,SAAO;AACT;AAUO,SAAS,wBACd,QACA,OAA0C,CAAC,GAClB;AACzB,QAAM,mBAAmB,IAAI,IAAI,KAAK,uBAAuB,CAAC,GAAG,4BAA4B,CAAC;AAC9F,QAAM,YAAY,KAAK,aAAa;AACpC,QAAM,kBAAkB,KAAK,mBAAmB;AAChD,QAAM,aAAa,KAAK,wBAAwB;AAEhD,QAAM,gBAAgB,OAAO,OAAO;AAAA,IAClC,CAAC,UAAU,iBAAiB,IAAI,MAAM,KAAK,KAAK,gBAAgB,KAAK;AAAA,EACvE;AAEA,MAAI,cAAc,WAAW,GAAG;AAC9B,UAAM,QAAQ,cAAc,CAAC;AAC7B,UAAM,QAAQ,QAAQ,MAAM,KAAM;AAClC,WAAO;AAAA,MACL;AAAA,MACA,QAAQ,UAAU,MAAM,KAAK;AAAA,MAC7B,aAAa;AAAA,MACb,YAAY;AAAA,MACZ,QAAQ,MAAM;AAAA,MACd,YAAY,EAAE,CAAC,MAAM,KAAK,GAAG,MAAM;AAAA,MACnC,WAAW,eAAe,KAAK;AAAA,IACjC;AAAA,EACF;AAEA,MAAI,cAAc,SAAS,GAAG;AAE5B,QAAI,MAAM;AACV,QAAI,QAAQ;AACZ,UAAM,aAAqC,CAAC;AAC5C,eAAW,KAAK,eAAe;AAC7B,YAAM,IAAK,EAAE,QAAQ,UAAiC;AACtD,aAAO,KAAK,EAAE,SAAS;AACvB,eAAS;AACT,iBAAW,EAAE,KAAK,IAAI,EAAE;AAAA,IAC1B;AACA,WAAO;AAAA,MACL,OAAO,UAAU,IAAI,IAAI,QAAQ,MAAM,KAAK;AAAA,MAC5C,QAAQ;AAAA,MACR,aAAa;AAAA,MACb,YAAY;AAAA,MACZ,QAAQ,cAAc,IAAI,CAAC,MAAM,EAAE,KAAK,EAAE,KAAK,GAAG;AAAA,MAClD;AAAA,MACA,WAAW,EAAE,GAAG,WAAW;AAAA,IAC7B;AAAA,EACF;AAEA,MAAI,CAAC,gBAAiB,QAAO;AAE7B,QAAM,QACJ,OAAO,OAAO,KAAK,CAAC,UAAU,gBAAgB,KAAK,KAAK,UAAU,MAAM,KAAK,MAAM,OAAO,KAC1F,OAAO,OAAO,KAAK,eAAe;AAEpC,MAAI,CAAC,MAAO,QAAO;AAEnB,QAAM,iBAAiB,MAAM,QAAQ;AACrC,QAAM,aAAa,QAAQ,MAAM,KAAM;AACvC,SAAO;AAAA,IACL,OAAO;AAAA,IACP,QAAQ;AAAA,IACR,aAAa;AAAA,IACb,YAAY,OAAO,mBAAmB,WAAW,iBAAiB;AAAA,IAClE,QAAQ,MAAM;AAAA,IACd,YAAY,EAAE,CAAC,MAAM,KAAK,GAAG,WAAW;AAAA,IACxC,WAAW,eAAe,KAAK;AAAA,EACjC;AACF;AAEA,SAAS,gBACP,OACmE;AACnE,UACG,MAAM,WAAW,UAAU,MAAM,WAAW,WAC7C,OAAO,MAAM,UAAU,YACvB,OAAO,SAAS,MAAM,KAAK,KAC3B,MAAM,SAAS,KACf,MAAM,SAAS;AAEnB;AAcO,SAAS,oCACd,MACA,OAA0C,CAAC,GACgB;AAC3D,QAAM,YAAY,KAAK,aAAa;AACpC,QAAM,mBAAmB,IAAI,IAAI,KAAK,uBAAuB,CAAC,GAAG,4BAA4B,CAAC;AAC9F,QAAM,kBAAkB,KAAK,mBAAmB;AAChD,QAAM,aAAa,KAAK,wBAAwB;AAEhD,SAAO,KAAK,IAAI,CAAC,QAAQ;AAEvB,UAAM,cAAsD,CAAC;AAC7D,eAAW,CAAC,GAAG,CAAC,KAAK,OAAO,QAAQ,IAAI,QAAQ,GAAG,GAAG;AACpD,UACE,EAAE,WAAW,QAAQ,KACrB,CAAC,EAAE,SAAS,KAAK,CAAC,KAClB,OAAO,MAAM,YACb,OAAO,SAAS,CAAC,GACjB;AACA,oBAAY,KAAK,EAAE,MAAM,EAAE,MAAM,SAAS,MAAM,GAAG,OAAO,EAAE,CAAC;AAAA,MAC/D;AAAA,IACF;AACA,UAAM,MAAM,YAAY,OAAO,CAAC,MAAM,iBAAiB,IAAI,EAAE,IAAI,CAAC;AAElE,QAAI,IAAI,WAAW,GAAG;AACpB,YAAM,QAAQ,IAAI,CAAC;AACnB,YAAM,QAAQ,QAAQ,MAAM,KAAK;AACjC,aAAO;AAAA,QACL,OAAO,IAAI;AAAA,QACX,QAAQ;AAAA,UACN;AAAA,UACA,QAAQ,UAAU,MAAM,IAAI;AAAA,UAC5B,aAAa;AAAA,UACb,YAAY;AAAA,UACZ,QAAQ,MAAM;AAAA,UACd,YAAY,EAAE,CAAC,MAAM,IAAI,GAAG,MAAM;AAAA,QACpC;AAAA,MACF;AAAA,IACF;AACA,QAAI,IAAI,SAAS,GAAG;AAClB,YAAM,QAAQ,IAAI,OAAO,CAAC,GAAG,MAAM,IAAI,EAAE,OAAO,CAAC,IAAI,IAAI;AACzD,YAAM,aAAqC,OAAO;AAAA,QAChD,IAAI,IAAI,CAAC,MAAM,CAAC,EAAE,MAAM,EAAE,KAAK,CAAC;AAAA,MAClC;AACA,aAAO;AAAA,QACL,OAAO,IAAI;AAAA,QACX,QAAQ;AAAA,UACN,OAAO,QAAQ,KAAK;AAAA,UACpB,QAAQ;AAAA,UACR,aAAa;AAAA,UACb,YAAY;AAAA,UACZ,QAAQ,IAAI,IAAI,CAAC,MAAM,EAAE,IAAI,EAAE,KAAK,GAAG;AAAA,UACvC;AAAA,UACA,WAAW,EAAE,GAAG,WAAW;AAAA,QAC7B;AAAA,MACF;AAAA,IACF;AACA,QAAI,CAAC,gBAAiB,QAAO,EAAE,OAAO,IAAI,OAAO,QAAQ,KAAK;AAG9D,UAAM,UAAU,aAAa,GAAG;AAChC,QAAI,YAAY,QAAW;AACzB,aAAO,EAAE,OAAO,IAAI,OAAO,QAAQ,KAAK;AAAA,IAC1C;AACA,UAAM,eAAe,QAAQ,OAAO;AACpC,WAAO;AAAA,MACL,OAAO,IAAI;AAAA,MACX,QAAQ;AAAA,QACN,OAAO;AAAA,QACP,QAAQ;AAAA,QACR,aAAa;AAAA,QACb,YAAY;AAAA,QACZ,QAAQ;AAAA,QACR,YAAY,EAAE,qBAAqB,aAAa;AAAA,MAClD;AAAA,IACF;AAAA,EACF,CAAC;AACH;AAGO,SAAS,gCACd,MACA,OAA0C,CAAC,GACU;AACrD,QAAM,WAAW,oCAAoC,MAAM,EAAE,GAAG,MAAM,iBAAiB,MAAM,CAAC;AAC9F,QAAM,MAA2D,CAAC;AAClE,WAAS,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK;AACpC,UAAM,IAAI,SAAS,CAAC;AACpB,QAAI,EAAE,UAAU,EAAE,OAAO,gBAAgB,iBAAiB;AACxD,UAAI,KAAK,EAAE,KAAK,KAAK,CAAC,GAAI,QAAQ,EAAE,OAAO,CAAC;AAAA,IAC9C;AAAA,EACF;AACA,SAAO;AACT;AAIA,SAAS,QAAQ,GAAmB;AAClC,MAAI,CAAC,OAAO,SAAS,CAAC,EAAG,QAAO;AAChC,SAAO,KAAK,IAAI,GAAG,KAAK,IAAI,GAAG,CAAC,CAAC;AACnC;AAEA,SAAS,eAAe,GAAwC;AAC9D,QAAM,MAA8B,CAAC;AACrC,MAAI,EAAE,aAAa;AACjB,eAAW,CAAC,GAAG,CAAC,KAAK,OAAO,QAAQ,EAAE,WAAW,GAAG;AAClD,UAAI,OAAO,MAAM,YAAY,OAAO,SAAS,CAAC,EAAG,KAAI,CAAC,IAAI;AAAA,IAC5D;AAAA,EACF;AACA,SAAO;AACT;;;ACrOA,IAAM,gBAAgB,CAAC,MAAgC;AACrD,SAAO,aAAa,CAAC,KAAK;AAC5B;AAEO,SAAS,oBAAoB,OAAsD;AACxF,QAAM,UAAU,MAAM,WAAW;AACjC,QAAM,UAAU,MAAM;AACtB,QAAM,MAAM,MAAM,YAAY,WAAW;AACzC,QAAM,MAAM,MAAM,YAAY,UAAU;AAExC,QAAM,OAAO,MAAM,KAAK,OAAO,CAAC,QAAQ,aAAa,QAAQ,GAAG,CAAC,CAAC;AAClE,QAAM,IAAI,KAAK;AACf,MAAI,IAAI,GAAG;AACT,WAAO;AAAA,MACL,UAAU,CAAC;AAAA,MACX,kBAAkB,CAAC;AAAA,MACnB,SAAS;AAAA,MACT;AAAA,MACA,WAAW,CAAC,0CAA0C,CAAC,0BAA0B;AAAA,IACnF;AAAA,EACF;AACA,QAAM,aAAa,KAAK,IAAI,GAAG,MAAM,cAAc,KAAK,IAAI,IAAI,KAAK,MAAM,IAAI,CAAC,CAAC,CAAC;AAClF,QAAM,SAAS,KAAK,MAAM,GAAG,IAAI,UAAU;AAC3C,QAAM,QAAQ,KAAK,MAAM,IAAI,UAAU;AAEvC,QAAM,WAAmC,CAAC;AAG1C,MAAI,SAAS;AACX,UAAM,cAAc,OAAO,IAAI,OAAO,EAAE,OAAO,YAAY;AAC3D,UAAM,aAAa,MAAM,IAAI,OAAO,EAAE,OAAO,YAAY;AACzD,UAAM,cAAc,OAAO,IAAI,OAAO,EAAE,OAAO,YAAY;AAC3D,UAAM,aAAa,MAAM,IAAI,OAAO,EAAE,OAAO,YAAY;AACzD,QACE,YAAY,UAAU,KACtB,WAAW,UAAU,KACrB,YAAY,UAAU,KACtB,WAAW,UAAU,GACrB;AACA,YAAM,aAAa,KAAK,UAAU,IAAI,KAAK,WAAW;AACtD,YAAM,aAAa,KAAK,UAAU,IAAI,KAAK,WAAW;AAGtD,YAAM,MAAM,KAAK,IAAI,GAAG,aAAa,UAAU;AAC/C,YAAM,WAAWA,SAAQ,MAAM,CAAC;AAChC,eAAS,KAAK;AAAA,QACZ,QAAQ;AAAA,QACR;AAAA,QACA,SACE,YAAY,MACR,wBAAwB,WAAW,QAAQ,CAAC,CAAC,2BAA2B,WAAW,QAAQ,CAAC,CAAC,+BAC7F,yCAAyC,WAAW,QAAQ,CAAC,CAAC,WAAW,WAAW,QAAQ,CAAC,CAAC;AAAA,QACpG,QAAQ;AAAA,UACN;AAAA,UACA;AAAA,UACA;AAAA,UACA,SAAS,YAAY;AAAA,UACrB,QAAQ,WAAW;AAAA,QACrB;AAAA,MACF,CAAC;AAAA,IACH;AAAA,EACF;AAGA;AACE,UAAM,UAAU,OAAO,IAAI,OAAO,EAAE,OAAO,YAAY;AACvD,UAAM,SAAS,MAAM,IAAI,OAAO,EAAE,OAAO,YAAY;AACrD,QAAI,QAAQ,UAAU,KAAK,OAAO,UAAU,GAAG;AAC7C,YAAM,KAAK,YAAY,SAAS,MAAM;AAItC,YAAM,WAAWA,SAAQ,KAAK,GAAG;AACjC,eAAS,KAAK;AAAA,QACZ,QAAQ;AAAA,QACR;AAAA,QACA,SACE,YAAY,MACR,MAAM,GAAG,QAAQ,CAAC,CAAC,oEACnB,MAAM,GAAG,QAAQ,CAAC,CAAC;AAAA,QACzB,QAAQ,EAAE,IAAI,SAAS,QAAQ,QAAQ,QAAQ,OAAO,OAAO;AAAA,MAC/D,CAAC;AAAA,IACH;AAAA,EACF;AAGA;AACE,UAAM,cAAc,MAAM,qBAAqB,iBAAiB,MAAM,uBAAuB;AAC7F,UAAM,UAAU,KACb,IAAI,CAAC,OAAO,EAAE,GAAG,QAAQ,CAAC,GAAG,GAAG,YAAY,CAAC,EAAE,EAAE,EACjD,OAAO,CAAC,MAAqC,aAAa,EAAE,CAAC,KAAK,aAAa,EAAE,CAAC,CAAC;AACtF,QAAI,QAAQ,UAAU,GAAG;AACvB,YAAM,KAAK,QAAQ,IAAI,CAAC,MAAM,EAAE,CAAC;AACjC,YAAM,KAAK,QAAQ,IAAI,CAAC,MAAM,EAAE,CAAC;AACjC,YAAM,IAAI,SAAS,IAAI,EAAE;AAGzB,YAAM,WAAWA,SAAQ,MAAM,KAAK,IAAI,GAAG,CAAC,CAAC;AAC7C,eAAS,KAAK;AAAA,QACZ,QAAQ;AAAA,QACR;AAAA,QACA,SACE,YAAY,MACR,2DAAsD,EAAE,QAAQ,CAAC,CAAC,mCAClE,+CAA0C,EAAE,QAAQ,CAAC,CAAC;AAAA,QAC5D,QAAQ,EAAE,SAAS,GAAG,GAAG,QAAQ,OAAO;AAAA,MAC1C,CAAC;AAAA,IACH;AAAA,EACF;AAGA;AACE,UAAM,UAAU,gCAAgC,MAAM,MAAM,2BAA2B,CAAC,CAAC;AACzF,QAAI,QAAQ,UAAU,GAAG;AACvB,YAAM,YAAY,QAAQ,MAAM,GAAG,KAAK,MAAM,QAAQ,SAAS,CAAC,CAAC;AACjE,YAAM,WAAW,QAAQ,MAAM,KAAK,MAAM,QAAQ,SAAS,CAAC,CAAC;AAC7D,YAAM,WACJ,KAAK,SAAS,IAAI,CAAC,MAAM,EAAE,OAAO,KAAK,CAAC,IAAI,KAAK,UAAU,IAAI,CAAC,MAAM,EAAE,OAAO,KAAK,CAAC;AACvF,YAAM,aACJ,KAAK,MAAM,IAAI,OAAO,EAAE,OAAO,YAAY,CAAC,IAC5C,KAAK,OAAO,IAAI,OAAO,EAAE,OAAO,YAAY,CAAC;AAC/C,YAAM,WAAW,KAAK,IAAI,GAAG,aAAa,QAAQ;AAClD,YAAM,WAAWA,SAAQ,WAAW,CAAC;AACrC,eAAS,KAAK;AAAA,QACZ,QAAQ;AAAA,QACR;AAAA,QACA,SACE,YAAY,MACR,gBAAgB,WAAW,QAAQ,CAAC,CAAC,gCAAgC,SAAS,QAAQ,CAAC,CAAC,yDACxF,uDAAuD,WAAW,QAAQ,CAAC,CAAC,SAAS,SAAS,QAAQ,CAAC,CAAC;AAAA,QAC9G,QAAQ,EAAE,YAAY,UAAU,UAAU,GAAG,QAAQ,OAAO;AAAA,MAC9D,CAAC;AAAA,IACH;AAAA,EACF;AAEA,QAAM,SAAS,SAAS,OAAO,CAAC,GAAG,MAAM,KAAK,IAAI,GAAG,EAAE,QAAQ,GAAG,CAAC;AACnE,MAAI,SAAS,WAAW,GAAG;AACzB,WAAO;AAAA,MACL;AAAA,MACA,kBAAkB,CAAC;AAAA,MACnB,SAAS;AAAA,MACT,WAAW,CAAC,0DAA0D,CAAC,GAAG;AAAA,MAC1E;AAAA,IACF;AAAA,EACF;AACA,QAAM,UACJ,UAAU,MAAM,WAAW,UAAU,MAAM,YAAY;AACzD,QAAM,YAAY,SACf,OAAO,CAAC,MAAM,EAAE,YAAY,GAAG,EAC/B,IAAI,CAAC,MAAM,GAAG,EAAE,MAAM,cAAc,EAAE,SAAS,QAAQ,CAAC,CAAC,WAAM,EAAE,OAAO,EAAE;AAC7E,MAAI,UAAU,WAAW,EAAG,WAAU,KAAK,0CAA0C;AAErF,SAAO;AAAA,IACL;AAAA,IACA,kBAAkB,SAAS,IAAI,CAAC,YAAY,QAAQ,MAAM;AAAA,IAC1D;AAAA,IACA;AAAA,IACA;AAAA,EACF;AACF;AAIA,SAAS,KAAK,IAAsB;AAClC,MAAI,GAAG,WAAW,EAAG,QAAO;AAC5B,SAAO,GAAG,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,GAAG;AAC5C;AAEA,SAAS,aAAa,OAAuC;AAC3D,SAAO,OAAO,UAAU,YAAY,OAAO,SAAS,KAAK;AAC3D;AAEA,SAASA,SAAQ,GAAmB;AAClC,MAAI,CAAC,OAAO,SAAS,CAAC,EAAG,QAAO;AAChC,SAAO,KAAK,IAAI,GAAG,KAAK,IAAI,GAAG,CAAC,CAAC;AACnC;AAEA,SAAS,YAAY,GAAa,GAAqB;AAErD,QAAM,UAAU,CAAC,GAAG,CAAC,EAAE,KAAK,CAAC,GAAG,MAAM,IAAI,CAAC;AAC3C,QAAM,UAAU,CAAC,GAAG,CAAC,EAAE,KAAK,CAAC,GAAG,MAAM,IAAI,CAAC;AAC3C,QAAM,MAAM,CAAC,GAAG,oBAAI,IAAI,CAAC,GAAG,SAAS,GAAG,OAAO,CAAC,CAAC,EAAE,KAAK,CAAC,GAAG,MAAM,IAAI,CAAC;AACvE,MAAI,MAAM;AACV,aAAW,KAAK,KAAK;AACnB,UAAM,KAAK,QAAQ,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,SAAS,QAAQ;AAC1D,UAAM,KAAK,QAAQ,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,SAAS,QAAQ;AAC1D,UAAM,KAAK,IAAI,KAAK,KAAK,IAAI,KAAK,EAAE,CAAC;AAAA,EACvC;AACA,SAAO;AACT;AAEA,SAAS,iBACP,gBACmC;AACnC,SAAO,CAAC,QAAmB;AACzB,UAAM,WAAW,gCAAgC,CAAC,GAAG,GAAG,kBAAkB,CAAC,CAAC;AAC5E,WAAO,SAAS,WAAW,IAAI,SAAS,CAAC,EAAG,OAAO,QAAQ;AAAA,EAC7D;AACF;","names":["clamp01"]}

package/dist/{chunk-QOTFXW5L.js → chunk-P5W7RQKK.js}

@@ -1,6 +1,6 @@
 import {
   cohensD
-} from "./chunk-PJQFMIOX.js";
+} from "./chunk-MHELPNRP.js";
 import {
   argHash,
   groupBy,
@@ -573,4 +573,4 @@ export {
   iqr,
   welchsTTest
 };
-//# sourceMappingURL=chunk-QOTFXW5L.js.map
+//# sourceMappingURL=chunk-P5W7RQKK.js.map

package/dist/{chunk-IR3KBHOY.js → chunk-P6FYH6K4.js}

@@ -20,6 +20,33 @@ var TOOL_SPAN_ATTRIBUTE_KEYS = [
   INPUT_VALUE,
   OUTPUT_VALUE
 ];
+var EXPLICIT_SPAN_ROLES = /* @__PURE__ */ new Set([
+  "AGENT",
+  "CHAIN",
+  "EVALUATOR",
+  "GUARDRAIL",
+  "LLM",
+  "SPAN",
+  "TOOL"
+]);
+function classifyOtlpSpanRole(input) {
+  const explicitKind = input.kind ?? firstStringAttribute(input.attributes, SPAN_KIND_ATTR_KEYS);
+  if (explicitKind) {
+    const normalized = explicitKind.toUpperCase();
+    if (EXPLICIT_SPAN_ROLES.has(normalized)) return normalized;
+  }
+  if (firstStringAttribute(input.attributes, TOOL_NAME_ATTR_KEYS) !== void 0 || /^(?:function|tool)[.:/]/i.test(input.name)) {
+    return "TOOL";
+  }
+  const spanType = input.attributes["span.type"];
+  if (typeof spanType === "string" && spanType.toLowerCase() === "llm_request" || /(?:^|[.:/_-])(?:chat[._-]?completions?|llm)(?:$|[.:/_-])/i.test(input.name) || firstStringAttribute(input.attributes, LLM_MODEL_ATTR_KEYS) !== void 0 || typeof input.attributes["gen_ai.operation.name"] === "string") {
+    return "LLM";
+  }
+  return "UNKNOWN";
+}
+function isOtlpModelCall(input) {
+  return classifyOtlpSpanRole(input) === "LLM";
+}
 function toolSpanOtlpAttributes(span) {
   const argsCaptured = span.argsCaptured !== false;
   const attributes = {
@@ -44,7 +71,7 @@ function traceSpanKindToOpenInferenceKind(kind) {
     case "retrieval":
       return "CHAIN";
     case "judge":
-      return "GUARDRAIL";
+      return "EVALUATOR";
     case "sandbox":
       return "CHAIN";
     case "agent":
@@ -53,6 +80,13 @@ function traceSpanKindToOpenInferenceKind(kind) {
       return "SPAN";
   }
 }
+function firstStringAttribute(attributes, keys) {
+  for (const key of keys) {
+    const value = attributes[key];
+    if (typeof value === "string" && value.length > 0) return value;
+  }
+  return void 0;
+}
 function stringifyTraceValue(value) {
   if (value === void 0) return "null";
   if (typeof value === "string") return value;
@@ -124,7 +158,7 @@ function inferOtlpKind(attrs) {
   const opik = firstStringAttr(attrs, SPAN_KIND_ATTR_KEYS);
   if (opik) {
     const upper = opik.toUpperCase();
-    if (upper === "AGENT" || upper === "LLM" || upper === "TOOL" || upper === "CHAIN" || upper === "GUARDRAIL" || upper === "SPAN") {
+    if (upper === "AGENT" || upper === "LLM" || upper === "TOOL" || upper === "CHAIN" || upper === "EVALUATOR" || upper === "GUARDRAIL" || upper === "SPAN") {
       return upper;
     }
   }
@@ -1101,6 +1135,8 @@ function validateLoopLimits(options) {
 }
 
 export {
+  classifyOtlpSpanRole,
+  isOtlpModelCall,
   applyToolSpanOtlpAttributes,
   traceSpanKindToOpenInferenceKind,
   projectOtlpFlatLine,
@@ -1122,4 +1158,4 @@ export {
   buildTraceAnalystTools,
   traceAnalystFunctionGroup
 };
-//# sourceMappingURL=chunk-IR3KBHOY.js.map
+//# sourceMappingURL=chunk-P6FYH6K4.js.map

package/dist/chunk-P6FYH6K4.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/trace/otlp-attributes.ts","../src/trace-analyst/otlp-span.ts","../src/trace-analyst/types.ts","../src/trace-analyst/store-otlp.ts","../src/trace-analyst/store.ts","../src/trace-analyst/tools.ts","../src/trace-analyst/loop.ts"],"sourcesContent":["/** Canonical OpenInference-over-OTLP attribute vocabulary used at the trace boundary. */\n\nimport {\n  INPUT_VALUE,\n  LLM_MODEL_ATTR_KEYS,\n  OUTPUT_VALUE,\n  SPAN_KIND_ATTR_KEYS,\n  TOOL_ARGS_CAPTURED,\n  TOOL_LATENCY_MS,\n  TOOL_NAME,\n  TOOL_NAME_ATTR_KEYS,\n} from './attribute-vocabulary'\nimport type { ToolSpan } from './schema'\n\nexport * from './attribute-vocabulary'\n\nconst TOOL_SPAN_ATTRIBUTE_KEYS = [\n  TOOL_NAME,\n  TOOL_ARGS_CAPTURED,\n  TOOL_LATENCY_MS,\n  INPUT_VALUE,\n  OUTPUT_VALUE,\n] as const\n\nexport type ToolSpanOtlpInput = Pick<\n  ToolSpan,\n  'toolName' | 'args' | 'argsCaptured' | 'result' | 'latencyMs'\n>\n\nexport type OtlpSpanRole =\n  | 'AGENT'\n  | 'CHAIN'\n  | 'EVALUATOR'\n  | 'GUARDRAIL'\n  | 'LLM'\n  | 'SPAN'\n  | 'TOOL'\n  | 'UNKNOWN'\n\nexport interface OtlpSpanRoleInput {\n  name: string\n  attributes: Record<string, unknown>\n  kind?: string | null\n}\n\nconst EXPLICIT_SPAN_ROLES = new Set<OtlpSpanRole>([\n  'AGENT',\n  'CHAIN',\n  'EVALUATOR',\n  'GUARDRAIL',\n  'LLM',\n  'SPAN',\n  'TOOL',\n])\n\n/**\n * Classify a span once for both measurement and error accounting.\n * An explicit OpenInference kind wins; untyped spans use the same tool and\n * model signals in online and offline intake.\n */\nexport function classifyOtlpSpanRole(input: OtlpSpanRoleInput): OtlpSpanRole {\n  const explicitKind = input.kind ?? firstStringAttribute(input.attributes, SPAN_KIND_ATTR_KEYS)\n  if (explicitKind) {\n    const normalized = explicitKind.toUpperCase() as OtlpSpanRole\n    if (EXPLICIT_SPAN_ROLES.has(normalized)) return normalized\n  }\n\n  if (\n    firstStringAttribute(input.attributes, TOOL_NAME_ATTR_KEYS) !== undefined ||\n    /^(?:function|tool)[.:/]/i.test(input.name)\n  ) {\n    return 'TOOL'\n  }\n\n  const spanType = input.attributes['span.type']\n  if (\n    (typeof spanType === 'string' && spanType.toLowerCase() === 'llm_request') ||\n    /(?:^|[.:/_-])(?:chat[._-]?completions?|llm)(?:$|[.:/_-])/i.test(input.name) ||\n    firstStringAttribute(input.attributes, LLM_MODEL_ATTR_KEYS) !== undefined ||\n    typeof input.attributes['gen_ai.operation.name'] === 'string'\n  ) {\n    return 'LLM'\n  }\n\n  return 'UNKNOWN'\n}\n\nexport function isOtlpModelCall(input: OtlpSpanRoleInput): boolean {\n  return classifyOtlpSpanRole(input) === 'LLM'\n}\n\nfunction toolSpanOtlpAttributes(\n  span: ToolSpanOtlpInput,\n): Record<string, string | number | boolean> {\n  const argsCaptured = span.argsCaptured !== false\n  const attributes: Record<string, string | number | boolean> = {\n    [TOOL_NAME]: span.toolName,\n    [TOOL_ARGS_CAPTURED]: argsCaptured,\n  }\n  if (span.latencyMs !== undefined) attributes[TOOL_LATENCY_MS] = span.latencyMs\n  if (argsCaptured) attributes[INPUT_VALUE] = stringifyTraceValue(span.args)\n  if (span.result !== undefined) attributes[OUTPUT_VALUE] = stringifyTraceValue(span.result)\n  return attributes\n}\n\nexport function applyToolSpanOtlpAttributes(\n  attributes: Record<string, unknown>,\n  span: ToolSpanOtlpInput,\n): void {\n  for (const key of TOOL_SPAN_ATTRIBUTE_KEYS) delete attributes[key]\n  Object.assign(attributes, toolSpanOtlpAttributes(span))\n}\n\nexport function traceSpanKindToOpenInferenceKind(kind: string): string {\n  switch (kind) {\n    case 'llm':\n      return 'LLM'\n    case 'tool':\n      return 'TOOL'\n    case 'retrieval':\n      return 'CHAIN'\n    case 'judge':\n      return 'EVALUATOR'\n    case 'sandbox':\n      return 'CHAIN'\n    case 'agent':\n      return 'AGENT'\n    default:\n      return 'SPAN'\n  }\n}\n\nfunction firstStringAttribute(\n  attributes: Record<string, unknown>,\n  keys: readonly string[],\n): string | undefined {\n  for (const key of keys) {\n    const value = attributes[key]\n    if (typeof value === 'string' && value.length > 0) return value\n  }\n  return undefined\n}\n\nfunction stringifyTraceValue(value: unknown): string {\n  if (value === undefined) return 'null'\n  if (typeof value === 'string') return value\n  try {\n    return JSON.stringify(value) ?? String(value)\n  } catch {\n    return String(value)\n  }\n}\n","/**\n * Canonical OTLP-flat-line readers shared by every consumer of the\n * OTLP-JSONL wire shape (one OTLP span per line; the form\n * `flattenOtlpExportToNdjson` produces and the form AppWorld / HALO\n * emit via their OpenInference OTLP exporter).\n *\n * `OtlpFileTraceStore` indexes spans with these; `otlpToRunRecords`\n * aggregates spans into `RunRecord`s with the same readers. One parser,\n * one vocabulary — a divergence between the analyst's view of a trace and\n * the RunRecord projected from it is a class of bug this consolidation\n * removes by construction.\n *\n * Vocabulary. The readers understand BOTH dialects that appear in the\n * wild:\n *   - the substrate's own `llm.*` / `tool.*` / `span.kind` attributes\n *     (`flattenSpanAttributes` in `trace/otel.ts`), and\n *   - the OpenInference / inference-export attributes AppWorld / HALO\n *     emit (`openinference.span.kind`, `inference.observation_kind`,\n *     `inference.llm.input_tokens`, `llm.token_count.prompt`, …).\n *\n * Pure, no I/O.\n */\n\nimport {\n  LLM_MODEL_ATTR_KEYS,\n  SPAN_KIND_ATTR_KEYS,\n  TOOL_NAME_ATTR_KEYS,\n} from '../trace/otlp-attributes'\nimport type { TraceAnalystSpanKind, TraceAnalystSpanStatus } from './types'\n\n/**\n * The structural fields a flat OTLP-JSONL line projects to. `attributes`\n * is the merged resource+span attribute map (span overrides resource);\n * the named fields are the pivots every reader of a trace needs without\n * paying the full attribute materialisation.\n */\nexport interface ProjectedOtlpSpan {\n  trace_id: string\n  span_id: string\n  parent_span_id: string | null\n  name: string\n  kind: TraceAnalystSpanKind\n  start_time: string\n  end_time: string\n  duration_ms: number\n  status: TraceAnalystSpanStatus\n  status_message: string | undefined\n  service_name: string | null\n  agent_name: string | null\n  model_name: string | null\n  tool_name: string | null\n  /** Merged resource + span attributes, span winning on overlap. */\n  attributes: Record<string, unknown>\n}\n\n/**\n * Project one parsed OTLP-JSONL object to `ProjectedOtlpSpan`, or `null`\n * when the line is missing the mandatory `trace_id` + `span_id`.\n */\nexport function projectOtlpFlatLine(raw: Record<string, unknown>): ProjectedOtlpSpan | null {\n  const trace_id = stringField(raw, 'trace_id') ?? stringField(raw, 'traceId')\n  const span_id = stringField(raw, 'span_id') ?? stringField(raw, 'spanId')\n  if (!trace_id || !span_id) return null\n\n  const rawParentId = stringField(raw, 'parent_span_id') ?? stringField(raw, 'parentSpanId') ?? null\n  const parent_id = normalizeParentSpanId(trace_id, span_id, rawParentId)\n  const name = stringField(raw, 'name') ?? 'unknown'\n  const start_time = stringField(raw, 'start_time') ?? stringField(raw, 'startTime') ?? ''\n  const end_time = stringField(raw, 'end_time') ?? stringField(raw, 'endTime') ?? start_time\n\n  const status = readOtlpStatus(raw)\n  const attributes = extractOtlpAttributes(raw)\n\n  const service_name =\n    asString(attributes['service.name']) ??\n    asString(attributes['resource.attributes.service.name']) ??\n    null\n  const agent_name =\n    asString(attributes['agent.name']) ??\n    asString(attributes['inference.agent.name']) ??\n    asString(attributes['inference.agent_name']) ??\n    null\n  const model_name = firstStringAttr(attributes, LLM_MODEL_ATTR_KEYS)\n  const tool_name = firstStringAttr(attributes, TOOL_NAME_ATTR_KEYS)\n\n  const kind = inferOtlpKind(attributes)\n\n  let duration_ms = 0\n  if (start_time && end_time) {\n    const a = spanEpochMillis(start_time)\n    const b = spanEpochMillis(end_time)\n    if (a !== null && b !== null) duration_ms = Math.max(0, b - a)\n  }\n\n  return {\n    trace_id,\n    span_id,\n    parent_span_id: parent_id && parent_id.length > 0 ? parent_id : null,\n    name,\n    kind,\n    start_time,\n    end_time,\n    duration_ms,\n    status: status.code,\n    status_message: status.message,\n    service_name,\n    agent_name,\n    model_name,\n    tool_name,\n    attributes,\n  }\n}\n\nfunction normalizeParentSpanId(\n  traceId: string,\n  spanId: string,\n  parentId: string | null,\n): string | null {\n  if (!parentId) return null\n  const prefix = `${traceId}:`\n  return spanId.startsWith(prefix) && !parentId.startsWith(prefix)\n    ? `${prefix}${parentId}`\n    : parentId\n}\n\nexport function readOtlpStatus(raw: Record<string, unknown>): {\n  code: TraceAnalystSpanStatus\n  message: string | undefined\n} {\n  const status = raw.status\n  if (status && typeof status === 'object' && !Array.isArray(status)) {\n    const codeRaw = (status as Record<string, unknown>).code\n    const code: TraceAnalystSpanStatus =\n      codeRaw === 'STATUS_CODE_OK' || codeRaw === 'OK'\n        ? 'OK'\n        : codeRaw === 'STATUS_CODE_ERROR' || codeRaw === 'ERROR'\n          ? 'ERROR'\n          : 'UNSET'\n    const messageRaw = (status as Record<string, unknown>).message\n    const message = typeof messageRaw === 'string' && messageRaw.length > 0 ? messageRaw : undefined\n    return { code, message }\n  }\n  return { code: 'UNSET', message: undefined }\n}\n\nexport function inferOtlpKind(attrs: Record<string, unknown>): TraceAnalystSpanKind {\n  const opik = firstStringAttr(attrs, SPAN_KIND_ATTR_KEYS)\n  if (opik) {\n    const upper = opik.toUpperCase()\n    if (\n      upper === 'AGENT' ||\n      upper === 'LLM' ||\n      upper === 'TOOL' ||\n      upper === 'CHAIN' ||\n      upper === 'EVALUATOR' ||\n      upper === 'GUARDRAIL' ||\n      upper === 'SPAN'\n    ) {\n      return upper as TraceAnalystSpanKind\n    }\n  }\n  return 'UNKNOWN'\n}\n\n/**\n * Flatten OTLP `attributes` + `resource.attributes` into a single\n * dotted-key map. Span attributes override resource attributes when keys\n * overlap. Nested objects/arrays are preserved as-is.\n */\nexport function extractOtlpAttributes(raw: Record<string, unknown>): Record<string, unknown> {\n  const out: Record<string, unknown> = {}\n  const resource = raw.resource\n  if (resource && typeof resource === 'object' && !Array.isArray(resource)) {\n    const ra = (resource as Record<string, unknown>).attributes\n    if (ra && typeof ra === 'object' && !Array.isArray(ra)) {\n      for (const [k, v] of Object.entries(ra as Record<string, unknown>)) {\n        out[k] = v\n      }\n    }\n  }\n  const spanAttrs = raw.attributes\n  if (spanAttrs && typeof spanAttrs === 'object' && !Array.isArray(spanAttrs)) {\n    for (const [k, v] of Object.entries(spanAttrs as Record<string, unknown>)) {\n      out[k] = v\n    }\n  }\n  return out\n}\n\nexport function stringField(raw: Record<string, unknown>, key: string): string | undefined {\n  const v = raw[key]\n  return typeof v === 'string' ? v : undefined\n}\n\nexport function asString(v: unknown): string | null {\n  return typeof v === 'string' && v.length > 0 ? v : null\n}\n\n/** First non-empty string value across a list of candidate attribute keys. */\nexport function firstStringAttr(\n  attrs: Record<string, unknown>,\n  keys: readonly string[],\n): string | null {\n  for (const k of keys) {\n    const s = asString(attrs[k])\n    if (s !== null) return s\n  }\n  return null\n}\n\n/**\n * Parse a span timestamp to epoch millis, or null when empty/unparseable. The\n * OTLP readers accept BOTH ISO-8601 and epoch-millis-string dialects, so raw\n * string comparison (`<`, `localeCompare`) mis-orders across dialects and\n * `Date.parse` returns NaN for a bare epoch-millis string.\n */\nexport function spanEpochMillis(ts: string | undefined | null): number | null {\n  if (!ts) return null\n  if (/^\\d+$/.test(ts)) return Number(ts)\n  const n = Date.parse(ts)\n  return Number.isNaN(n) ? null : n\n}\n\n/**\n * Order comparator for span timestamps across mixed ISO/epoch dialects.\n * Unparseable timestamps sort as epoch 0 (earliest), never NaN (which would\n * make the sort non-deterministic).\n */\nexport function compareSpanTime(a: string, b: string): number {\n  return (spanEpochMillis(a) ?? 0) - (spanEpochMillis(b) ?? 0)\n}\n","/**\n * Shared types for the trace-analyst module.\n *\n * Wire format. The store interface speaks `OtlpSpanLike` rows — one JSONL\n * line per span, OTLP-shaped. We do NOT depend on a specific tracing\n * vendor at the type level. Adapter\n * layers map upstream shapes onto this interface.\n *\n * Design constraint. Every read operation that can return arbitrary\n * payload must carry a byte budget so the agent's tool result stays\n * bounded regardless of input trace size. Oversized responses\n * substitute a deterministic summary instead of bytes — see\n * `ViewTraceOversized`.\n */\n\n/** OTLP span kind (subset we actually use). */\nexport type TraceAnalystSpanKind =\n  | 'AGENT'\n  | 'LLM'\n  | 'TOOL'\n  | 'CHAIN'\n  | 'EVALUATOR'\n  | 'GUARDRAIL'\n  | 'SPAN'\n  | 'UNKNOWN'\n\nexport type TraceAnalystSpanStatus = 'OK' | 'ERROR' | 'UNSET'\n\n/** Subset of OTLP span fields the analyst exposes to the agent. The\n *  store's job is to project upstream's full span shape down to this\n *  view — the analyst never sees vendor extensions directly. */\nexport interface TraceAnalystSpan {\n  trace_id: string\n  span_id: string\n  parent_span_id: string | null\n  name: string\n  kind: TraceAnalystSpanKind\n  start_time: string\n  end_time: string\n  duration_ms: number\n  status: TraceAnalystSpanStatus\n  status_message?: string\n  service_name: string | null\n  agent_name: string | null\n  model_name: string | null\n  tool_name: string | null\n  /** Raw JSON-serialisable attribute map. May contain large strings;\n   *  callers must respect the per-attribute byte cap. */\n  attributes: Record<string, unknown>\n}\n\nexport interface TraceAnalystTraceSummary {\n  trace_id: string\n  service_name: string | null\n  agent_name: string | null\n  span_count: number\n  has_errors: boolean\n  start_time: string\n  end_time: string\n  duration_ms: number\n  raw_jsonl_bytes: number\n  models: string[]\n  tools: string[]\n}\n\nexport interface TraceAnalystFilters {\n  /** Restrict to traces that contain at least one error span. */\n  has_errors?: boolean\n  /** Match if any span's `service.name` is in this list. */\n  service_names?: string[]\n  /** Match if any span's `agent.name` is in this list. */\n  agent_names?: string[]\n  /** Match if any LLM span's `llm.model_name` is in this list. */\n  model_names?: string[]\n  /** Match if any tool span's `tool.name` is in this list. */\n  tool_names?: string[]\n  /** ISO-8601 lower bound on the trace's earliest start time. */\n  start_time_after?: string\n  /** ISO-8601 upper bound on the trace's earliest start time. */\n  start_time_before?: string\n  /** Single regex applied to raw JSONL bytes for the trace. Opt-in;\n   *  expensive on large datasets. Use the indexed filters above first. */\n  regex_pattern?: string\n}\n\n/** One distinct error signature across the dataset — the deterministic unit of\n *  failure coverage. Signatures normalize volatile tokens (digits, hex/uuids,\n *  paths, durations) out of the span `status_message` so semantically identical\n *  failures collapse into one cluster. An analyst that accounts for every\n *  cluster has, by construction, covered every distinct failure mode. */\nexport interface ErrorCluster {\n  /** Normalized status_message — the cluster key. */\n  signature: string\n  /** A verbatim, un-normalized exemplar message (for exact-string citation). */\n  status_message_sample: string\n  /** The span name that most often carries this signature, if any. */\n  span_name: string | null\n  /** The tool that most often carries this signature, if any. */\n  tool_name: string | null\n  trace_count: number\n  span_count: number\n  /** trace_count / total error traces in the matched set (0..1). */\n  prevalence: number\n  /** Real trace ids carrying this signature (capped), passable to view/search. */\n  exemplar_trace_ids: string[]\n  /** Real span ids carrying this signature (capped). */\n  exemplar_span_ids: string[]\n}\n\nexport interface DatasetOverview {\n  total_traces: number\n  raw_jsonl_bytes: number\n  services: string[]\n  agents: string[]\n  models: string[]\n  tool_names: string[]\n  /** Up to 20 real trace ids the agent may pass to view/search tools. */\n  sample_trace_ids: string[]\n  errors: { trace_count: number; span_count: number }\n  /** The COMPLETE deterministic error-signature population, sorted by\n   *  trace_count desc. This is the failure-coverage checklist: an analysis is\n   *  complete only when every cluster here is accounted for. Empty when the\n   *  matched set has no error spans. */\n  error_clusters: ErrorCluster[]\n  time_range: { earliest: string; latest: string } | null\n}\n\nexport interface QueryTracesPage {\n  traces: TraceAnalystTraceSummary[]\n  total: number\n  has_more: boolean\n}\n\n/** Full-trace view. When the response would exceed the per-call byte\n *  budget, `oversized` is populated INSTEAD of `spans` so the agent\n *  knows to switch to `searchTrace` / `viewSpans`. */\nexport interface ViewTraceResult {\n  trace_id: string\n  spans?: TraceAnalystSpan[]\n  oversized?: ViewTraceOversized\n}\n\nexport interface ViewTraceOversized {\n  span_count: number\n  /** Names with their counts, sorted desc. Capped at 20 entries. */\n  top_span_names: Array<[string, number]>\n  /** Largest single span body (bytes after attribute-cap projection). */\n  span_response_bytes_max: number\n  error_span_count: number\n}\n\nexport interface ViewSpansResult {\n  trace_id: string\n  spans: TraceAnalystSpan[]\n  /** Number of requested span ids that were not found in the trace. */\n  missing_span_ids: string[]\n  /** Number of attribute fields truncated to fit the per-attribute cap. */\n  truncated_attribute_count: number\n}\n\nexport interface SpanMatchRecord {\n  trace_id: string\n  span_id: string\n  span_name: string\n  span_kind: TraceAnalystSpanKind\n  /** JSON pointer-style path to the matched value, e.g.\n   *  `attributes.\"llm.input_messages\"[2].content`. */\n  attribute_path: string\n  matched_text: string\n  context_before: string\n  context_after: string\n  match_offset: number\n}\n\nexport interface SearchTraceResult {\n  trace_id: string\n  hits: SpanMatchRecord[]\n  total_matches: number\n  has_more: boolean\n}\n\nexport interface SearchSpanResult {\n  trace_id: string\n  span_id: string\n  hits: SpanMatchRecord[]\n  total_matches: number\n  has_more: boolean\n}\n\n/** Tunable byte budgets for bounded RLM tool output. */\nexport interface TraceAnalystByteBudgets {\n  /** Max bytes any single tool response may emit. Hard ceiling enforced\n   *  by the store; oversized → summary. Default 150_000. */\n  perCallByteCeiling: number\n  /** Per-attribute string truncation cap on `viewTrace` (discovery scan).\n   *  Default 4096. */\n  perAttributeViewBudget: number\n  /** Per-attribute string truncation cap on `viewSpans` (surgical reads).\n   *  Default 16384. */\n  perAttributeSpanBudget: number\n  /** Per-attribute cap on a single match record's `matched_text` and\n   *  context window. Default 1024. */\n  perMatchTextBudget: number\n}\n\nexport const DEFAULT_TRACE_ANALYST_BUDGETS: TraceAnalystByteBudgets = {\n  perCallByteCeiling: 150_000,\n  perAttributeViewBudget: 4_096,\n  perAttributeSpanBudget: 16_384,\n  perMatchTextBudget: 1_024,\n}\n\n/** Marker substituted in place of truncated string payloads. Callers\n *  parsing tool output can detect it deterministically. */\nexport const TRACE_ANALYST_TRUNCATION_MARKER_PREFIX = '[trace-analyst truncated:'\n","/**\n * `OtlpFileTraceStore` — read-only OTLP-JSONL trace store for the\n * trace-analyst.\n *\n * Wire shape. Each line of the input file is one OTLP-shaped span. The\n * store understands flattened OTLP JSONL plus the OpenInference vocab.\n * We project upstream's full\n * span shape down to `TraceAnalystSpan` lazily — full materialisation\n * only happens for the spans the agent actually requests.\n *\n * Indexing. On first read the store builds an in-memory index keyed\n * by `trace_id` carrying:\n *   - byte offsets + lengths for each span line (for surgical reads\n *     without re-parsing the whole file)\n *   - a `TraceAnalystTraceSummary` rollup\n *   - sets of services / agents / models / tools / has_errors\n *   - byte size of the trace's JSONL slab\n *\n * Memory bound. The index keeps span metadata only — names, kinds,\n * offsets, status. Attribute payloads stay on disk until requested.\n * For a 50MB JSONL with 50k spans, the index is ~5MB.\n *\n * Concurrency. The store builds the index once on first read and\n * caches it. Subsequent reads reuse the index. The file is opened on\n * each read; we never hold a long-lived FD.\n */\n\nimport { readFile, stat } from 'node:fs/promises'\nimport { NotFoundError } from '../errors'\nimport {\n  compareSpanTime,\n  extractOtlpAttributes,\n  projectOtlpFlatLine,\n  spanEpochMillis,\n} from './otlp-span'\nimport { compileSearchRegex, type TraceAnalysisStore, truncateForBudget } from './store'\nimport {\n  type DatasetOverview,\n  DEFAULT_TRACE_ANALYST_BUDGETS,\n  type ErrorCluster,\n  type QueryTracesPage,\n  type SearchSpanResult,\n  type SearchTraceResult,\n  type SpanMatchRecord,\n  type TraceAnalystFilters,\n  type TraceAnalystSpan,\n  type TraceAnalystSpanKind,\n  type TraceAnalystSpanStatus,\n  type TraceAnalystTraceSummary,\n  type ViewSpansResult,\n  type ViewTraceOversized,\n  type ViewTraceResult,\n} from './types'\n\n/** Lines indexed between event-loop yields. Bounded synchronous work per\n *  tick keeps the index build from starving other tasks on large files\n *  while staying coarse enough that the yields are cheap. */\nconst INDEX_YIELD_LINES = 5000\n\n/** Hand control back to the event loop without busy-waiting. */\nfunction yieldToEventLoop(): Promise<void> {\n  return new Promise((resolve) => setImmediate(resolve))\n}\n\ninterface SpanIndexEntry {\n  span_id: string\n  parent_span_id: string | null\n  name: string\n  kind: TraceAnalystSpanKind\n  start_time: string\n  end_time: string\n  duration_ms: number\n  status: TraceAnalystSpanStatus\n  status_message: string | undefined\n  service_name: string | null\n  agent_name: string | null\n  model_name: string | null\n  tool_name: string | null\n  /** Byte offset in the raw JSONL file to the start of this span's line. */\n  line_byte_offset: number\n  /** Length of this line in bytes (excluding the trailing newline). */\n  line_byte_length: number\n}\n\ninterface TraceIndexEntry {\n  trace_id: string\n  service_name: string | null\n  agent_name: string | null\n  span_count: number\n  has_errors: boolean\n  start_time: string\n  end_time: string\n  duration_ms: number\n  raw_jsonl_bytes: number\n  models: Set<string>\n  tools: Set<string>\n  spans: SpanIndexEntry[]\n  /** Sorted by line offset for stable iteration. */\n}\n\ninterface DatasetIndex {\n  byTrace: Map<string, TraceIndexEntry>\n  totalRawBytes: number\n  // Pre-computed sorted trace_ids for sample/query stability.\n  sortedTraceIds: string[]\n}\n\nexport interface OtlpFileTraceStoreOptions {\n  /** Path to the OTLP-JSONL file. */\n  path: string\n  /** Override the discovery (`viewTrace`) per-attribute byte cap. */\n  perAttributeViewBudget?: number\n  /** Override the surgical (`viewSpans`) per-attribute byte cap. */\n  perAttributeSpanBudget?: number\n  /** Override the per-call ceiling that triggers oversized summaries. */\n  perCallByteCeiling?: number\n  /** Override the per-match text budget. */\n  perMatchTextBudget?: number\n  /**\n   * Hard ceiling on the trace file size in bytes. The store reads the\n   * whole file into one Buffer and indexes it in memory, so an\n   * unbounded file OOMs the process. Above this size the store fails\n   * loud with `TraceFileTooLargeError` instead of degrading silently.\n   * Default 256 MiB.\n   */\n  maxFileBytes?: number\n}\n\n/** Default ceiling for {@link OtlpFileTraceStoreOptions.maxFileBytes}. */\nexport const DEFAULT_MAX_TRACE_FILE_BYTES = 256 * 1024 * 1024\n\nexport class OtlpFileTraceStore implements TraceAnalysisStore {\n  private readonly path: string\n  private readonly perAttributeViewBudget: number\n  private readonly perAttributeSpanBudget: number\n  private readonly perCallByteCeiling: number\n  private readonly perMatchTextBudget: number\n  private readonly maxFileBytes: number\n  private indexPromise?: Promise<DatasetIndex>\n  /** Cached UTF-8 buffer of the file. We pin it once because every\n   *  read needs slice access and re-reading on each call balloons the\n   *  syscall count. */\n  private bufferPromise?: Promise<Buffer>\n\n  constructor(opts: OtlpFileTraceStoreOptions) {\n    this.path = opts.path\n    this.perAttributeViewBudget =\n      opts.perAttributeViewBudget ?? DEFAULT_TRACE_ANALYST_BUDGETS.perAttributeViewBudget\n    this.perAttributeSpanBudget =\n      opts.perAttributeSpanBudget ?? DEFAULT_TRACE_ANALYST_BUDGETS.perAttributeSpanBudget\n    this.perCallByteCeiling =\n      opts.perCallByteCeiling ?? DEFAULT_TRACE_ANALYST_BUDGETS.perCallByteCeiling\n    this.perMatchTextBudget =\n      opts.perMatchTextBudget ?? DEFAULT_TRACE_ANALYST_BUDGETS.perMatchTextBudget\n    this.maxFileBytes = opts.maxFileBytes ?? DEFAULT_MAX_TRACE_FILE_BYTES\n  }\n\n  // ─── Public API ────────────────────────────────────────────────────\n\n  async getOverview(filters?: TraceAnalystFilters): Promise<DatasetOverview> {\n    const idx = await this.index()\n    const matched = await this.matchedTraces(idx, filters)\n\n    const services = new Set<string>()\n    const agents = new Set<string>()\n    const models = new Set<string>()\n    const tools = new Set<string>()\n    let rawBytes = 0\n    let earliest: string | null = null\n    let latest: string | null = null\n    let errorTraceCount = 0\n    let errorSpanCount = 0\n    const clusters = new Map<string, ErrorClusterAccumulator>()\n\n    for (const t of matched) {\n      if (t.service_name) services.add(t.service_name)\n      if (t.agent_name) agents.add(t.agent_name)\n      for (const m of t.models) models.add(m)\n      for (const tn of t.tools) tools.add(tn)\n      rawBytes += t.raw_jsonl_bytes\n      if (!earliest || compareSpanTime(t.start_time, earliest) < 0) earliest = t.start_time\n      if (!latest || compareSpanTime(t.end_time, latest) > 0) latest = t.end_time\n      if (t.has_errors) {\n        errorTraceCount += 1\n        for (const s of t.spans) {\n          if (s.status !== 'ERROR') continue\n          errorSpanCount += 1\n          accumulateErrorCluster(clusters, t.trace_id, s)\n        }\n      }\n    }\n\n    const sample_trace_ids = matched.slice(0, 20).map((t) => t.trace_id)\n    return {\n      total_traces: matched.length,\n      raw_jsonl_bytes: rawBytes,\n      services: [...services].sort(),\n      agents: [...agents].sort(),\n      models: [...models].sort(),\n      tool_names: [...tools].sort(),\n      sample_trace_ids,\n      errors: { trace_count: errorTraceCount, span_count: errorSpanCount },\n      error_clusters: finalizeErrorClusters(clusters, errorTraceCount),\n      time_range: earliest && latest ? { earliest, latest } : null,\n    }\n  }\n\n  async queryTraces(opts: {\n    filters?: TraceAnalystFilters\n    limit: number\n    offset?: number\n  }): Promise<QueryTracesPage> {\n    if (!Number.isInteger(opts.limit) || opts.limit < 1 || opts.limit > 200) {\n      throw new RangeError(`queryTraces.limit must be 1..200, got ${opts.limit}`)\n    }\n    const offset = opts.offset ?? 0\n    if (!Number.isInteger(offset) || offset < 0) {\n      throw new RangeError(`queryTraces.offset must be >=0, got ${offset}`)\n    }\n\n    const idx = await this.index()\n    const matched = await this.matchedTraces(idx, opts.filters)\n    const slice = matched.slice(offset, offset + opts.limit)\n    return {\n      traces: slice.map((t) => this.toSummary(t)),\n      total: matched.length,\n      has_more: offset + slice.length < matched.length,\n    }\n  }\n\n  async countTraces(filters?: TraceAnalystFilters): Promise<number> {\n    const idx = await this.index()\n    const matched = await this.matchedTraces(idx, filters)\n    return matched.length\n  }\n\n  async viewTrace(opts: {\n    trace_id: string\n    per_attribute_byte_cap?: number\n  }): Promise<ViewTraceResult> {\n    const idx = await this.index()\n    const trace = idx.byTrace.get(opts.trace_id)\n    if (!trace) {\n      throw new TraceNotFoundError(opts.trace_id)\n    }\n    const cap = opts.per_attribute_byte_cap ?? this.perAttributeViewBudget\n\n    // Probe size first — if the materialised payload would exceed\n    // the per-call ceiling we return an oversized summary rather than\n    // blowing the agent's context.\n    const buf = await this.buffer()\n    const spans: TraceAnalystSpan[] = []\n    let runningBytes = 0\n    let span_response_bytes_max = 0\n    const counter: TruncationCounter = { value: 0 }\n    for (const s of trace.spans) {\n      const projected = this.projectSpan(buf, trace.trace_id, s, cap, counter)\n      const bytes = Buffer.byteLength(JSON.stringify(projected), 'utf8')\n      span_response_bytes_max = Math.max(span_response_bytes_max, bytes)\n      runningBytes += bytes\n      if (runningBytes > this.perCallByteCeiling) {\n        return {\n          trace_id: trace.trace_id,\n          oversized: this.buildOversizedSummary(trace, span_response_bytes_max),\n        }\n      }\n      spans.push(projected)\n    }\n    return { trace_id: trace.trace_id, spans }\n  }\n\n  async viewSpans(opts: {\n    trace_id: string\n    span_ids: readonly string[]\n    per_attribute_byte_cap?: number\n  }): Promise<ViewSpansResult> {\n    const idx = await this.index()\n    const trace = idx.byTrace.get(opts.trace_id)\n    if (!trace) throw new TraceNotFoundError(opts.trace_id)\n    if (opts.span_ids.length === 0) {\n      return {\n        trace_id: trace.trace_id,\n        spans: [],\n        missing_span_ids: [],\n        truncated_attribute_count: 0,\n      }\n    }\n    if (opts.span_ids.length > 100) {\n      throw new RangeError(`viewSpans.span_ids cap is 100, got ${opts.span_ids.length}`)\n    }\n    const cap = opts.per_attribute_byte_cap ?? this.perAttributeSpanBudget\n\n    const wantSet = new Set(opts.span_ids)\n    const found = trace.spans.filter((s) => wantSet.has(s.span_id))\n    const missing = opts.span_ids.filter((id) => !found.some((f) => f.span_id === id))\n\n    const buf = await this.buffer()\n    const spans: TraceAnalystSpan[] = []\n    const counter: TruncationCounter = { value: 0 }\n    let runningBytes = 0\n    for (const s of found) {\n      const projected = this.projectSpan(buf, trace.trace_id, s, cap, counter)\n      const bytes = Buffer.byteLength(JSON.stringify(projected), 'utf8')\n      runningBytes += bytes\n      if (runningBytes > this.perCallByteCeiling) {\n        // Stop adding further spans rather than truncate mid-list.\n        // Callers can refetch the rest with a smaller `span_ids`.\n        break\n      }\n      spans.push(projected)\n    }\n    return {\n      trace_id: trace.trace_id,\n      spans,\n      missing_span_ids: missing,\n      truncated_attribute_count: counter.value,\n    }\n  }\n\n  async searchTrace(opts: {\n    trace_id: string\n    regex_pattern: string\n    max_matches?: number\n  }): Promise<SearchTraceResult> {\n    const max_matches = opts.max_matches ?? 50\n    if (!Number.isInteger(max_matches) || max_matches < 1 || max_matches > 500) {\n      throw new RangeError(`searchTrace.max_matches must be 1..500, got ${max_matches}`)\n    }\n    const idx = await this.index()\n    const trace = idx.byTrace.get(opts.trace_id)\n    if (!trace) throw new TraceNotFoundError(opts.trace_id)\n    const re = compileSearchRegex(opts.regex_pattern)\n\n    const buf = await this.buffer()\n    const hits: SpanMatchRecord[] = []\n    let total = 0\n    let capped = false\n    for (const s of trace.spans) {\n      const remaining = max_matches - hits.length\n      const localHits = await this.scanSpanForMatches(\n        buf,\n        trace.trace_id,\n        s,\n        re,\n        this.perMatchTextBudget,\n        remaining,\n      )\n      total += localHits.total\n      for (const h of localHits.records) {\n        if (hits.length >= max_matches) break\n        hits.push(h)\n      }\n      if (hits.length >= max_matches) {\n        // Capped: we stopped scanning, so `total` is a lower bound on the\n        // real match count — never report it as exact. has_more signals\n        // \"more exist\"; total_matches mirrors hits so callers don't read a\n        // fabricated number.\n        capped = true\n        break\n      }\n    }\n    return {\n      trace_id: trace.trace_id,\n      hits,\n      // Uncapped: every span scanned to completion, so `total` is exact.\n      total_matches: capped ? hits.length : total,\n      has_more: capped || total > hits.length,\n    }\n  }\n\n  async searchSpan(opts: {\n    trace_id: string\n    span_id: string\n    regex_pattern: string\n    max_matches?: number\n  }): Promise<SearchSpanResult> {\n    const max_matches = opts.max_matches ?? 50\n    if (!Number.isInteger(max_matches) || max_matches < 1 || max_matches > 500) {\n      throw new RangeError(`searchSpan.max_matches must be 1..500, got ${max_matches}`)\n    }\n    const idx = await this.index()\n    const trace = idx.byTrace.get(opts.trace_id)\n    if (!trace) throw new TraceNotFoundError(opts.trace_id)\n    const span = trace.spans.find((s) => s.span_id === opts.span_id)\n    if (!span) {\n      throw new SpanNotFoundError(opts.trace_id, opts.span_id)\n    }\n    const re = compileSearchRegex(opts.regex_pattern)\n    const buf = await this.buffer()\n    const localHits = await this.scanSpanForMatches(\n      buf,\n      trace.trace_id,\n      span,\n      re,\n      this.perMatchTextBudget,\n      max_matches,\n    )\n    return {\n      trace_id: trace.trace_id,\n      span_id: span.span_id,\n      hits: localHits.records,\n      total_matches: localHits.total,\n      has_more: localHits.total > localHits.records.length,\n    }\n  }\n\n  // ─── Index building ────────────────────────────────────────────────\n\n  /** Force the index to materialise. Useful to amortise startup cost\n   *  before the first agent call. */\n  async ensureIndexed(): Promise<void> {\n    await this.index()\n  }\n\n  private async buffer(): Promise<Buffer> {\n    if (!this.bufferPromise) {\n      this.bufferPromise = this.readGuarded()\n    }\n    return this.bufferPromise\n  }\n\n  /** Stat-then-read so an oversized file fails loud BEFORE we allocate a\n   *  multi-hundred-MB Buffer and OOM the process. A missing file surfaces\n   *  as TraceFileMissingError; any other stat/read error propagates. */\n  private async readGuarded(): Promise<Buffer> {\n    let stats: Awaited<ReturnType<typeof stat>>\n    try {\n      stats = await stat(this.path)\n    } catch (err) {\n      if ((err as NodeJS.ErrnoException)?.code === 'ENOENT') {\n        throw new TraceFileMissingError(this.path)\n      }\n      throw err\n    }\n    if (stats.size > this.maxFileBytes) {\n      throw new TraceFileTooLargeError(this.path, stats.size, this.maxFileBytes)\n    }\n    return readFile(this.path)\n  }\n\n  private async index(): Promise<DatasetIndex> {\n    if (!this.indexPromise) {\n      this.indexPromise = this.buildIndex()\n    }\n    return this.indexPromise\n  }\n\n  private async buildIndex(): Promise<DatasetIndex> {\n    // readGuarded surfaces missing/oversized files as typed errors.\n    const buf = await this.buffer()\n\n    const byTrace = new Map<string, TraceIndexEntry>()\n    let cursor = 0\n    let sinceYield = 0\n    while (cursor < buf.length) {\n      // Yield to the event loop every INDEX_YIELD_LINES lines so a huge\n      // file doesn't monopolise the thread for the whole index build.\n      if (++sinceYield >= INDEX_YIELD_LINES) {\n        sinceYield = 0\n        await yieldToEventLoop()\n      }\n      const newlineIndex = buf.indexOf(0x0a, cursor) // \\n\n      const lineEnd = newlineIndex === -1 ? buf.length : newlineIndex\n      const lineLength = lineEnd - cursor\n      if (lineLength === 0) {\n        cursor = lineEnd + 1\n        continue\n      }\n      const lineSlice = buf.subarray(cursor, lineEnd).toString('utf8')\n      const lineOffset = cursor\n      cursor = lineEnd + 1\n\n      let parsed: unknown\n      try {\n        parsed = JSON.parse(lineSlice)\n      } catch {\n        // Skip malformed lines silently. The agent shouldn't see them\n        // but we don't want one bad line to nuke an entire dataset.\n        continue\n      }\n      if (!parsed || typeof parsed !== 'object') continue\n      const span = projectOtlpFlatLine(parsed as Record<string, unknown>)\n      if (!span) continue\n\n      let entry = byTrace.get(span.trace_id)\n      if (!entry) {\n        entry = {\n          trace_id: span.trace_id,\n          service_name: span.service_name,\n          agent_name: span.agent_name,\n          span_count: 0,\n          has_errors: false,\n          start_time: span.start_time,\n          end_time: span.end_time,\n          duration_ms: 0,\n          raw_jsonl_bytes: 0,\n          models: new Set(),\n          tools: new Set(),\n          spans: [],\n        }\n        byTrace.set(span.trace_id, entry)\n      } else {\n        // Pin the trace's service/agent to the first AGENT span we\n        // Prefer the first agent/service fields that appear in the trace.\n        if (!entry.service_name && span.service_name) entry.service_name = span.service_name\n        if (!entry.agent_name && span.agent_name) entry.agent_name = span.agent_name\n      }\n\n      const indexEntry: SpanIndexEntry = {\n        span_id: span.span_id,\n        parent_span_id: span.parent_span_id,\n        name: span.name,\n        kind: span.kind,\n        start_time: span.start_time,\n        end_time: span.end_time,\n        duration_ms: span.duration_ms,\n        status: span.status,\n        status_message: span.status_message,\n        service_name: span.service_name,\n        agent_name: span.agent_name,\n        model_name: span.model_name,\n        tool_name: span.tool_name,\n        line_byte_offset: lineOffset,\n        line_byte_length: lineLength,\n      }\n      entry.spans.push(indexEntry)\n      entry.span_count += 1\n      entry.raw_jsonl_bytes += lineLength + 1 // +1 newline byte\n      if (span.status === 'ERROR') entry.has_errors = true\n      if (compareSpanTime(span.start_time, entry.start_time) < 0) entry.start_time = span.start_time\n      if (compareSpanTime(span.end_time, entry.end_time) > 0) entry.end_time = span.end_time\n      if (span.model_name) entry.models.add(span.model_name)\n      if (span.tool_name) entry.tools.add(span.tool_name)\n    }\n\n    // Compute trace duration once, sort spans by start time for\n    // stable iteration.\n    let totalRawBytes = 0\n    for (const t of byTrace.values()) {\n      totalRawBytes += t.raw_jsonl_bytes\n      t.spans.sort(\n        (a, b) =>\n          compareSpanTime(a.start_time, b.start_time) || a.line_byte_offset - b.line_byte_offset,\n      )\n      // Duration is 0 unless BOTH bounds parse — a missing/garbage timestamp\n      // yields 0, never a NaN (→ null in JSON) or a bogus epoch-from-zero span.\n      const startMs = spanEpochMillis(t.start_time)\n      const endMs = spanEpochMillis(t.end_time)\n      t.duration_ms = startMs === null || endMs === null ? 0 : Math.max(0, endMs - startMs)\n    }\n    const sortedTraceIds = [...byTrace.keys()].sort()\n\n    return { byTrace, totalRawBytes, sortedTraceIds }\n  }\n\n  // ─── Filter pipeline ───────────────────────────────────────────────\n\n  private async matchedTraces(\n    idx: DatasetIndex,\n    filters: TraceAnalystFilters | undefined,\n  ): Promise<TraceIndexEntry[]> {\n    const traces = idx.sortedTraceIds.map((id) => idx.byTrace.get(id)).filter(isPresent)\n    if (!filters) return traces\n\n    const indexedFiltered = traces.filter((t) => {\n      if (filters.has_errors !== undefined && t.has_errors !== filters.has_errors) return false\n      if (filters.service_names && filters.service_names.length > 0) {\n        if (!t.service_name || !filters.service_names.includes(t.service_name)) return false\n      }\n      if (filters.agent_names && filters.agent_names.length > 0) {\n        if (!t.agent_name || !filters.agent_names.includes(t.agent_name)) return false\n      }\n      if (filters.model_names && filters.model_names.length > 0) {\n        if (![...t.models].some((m) => filters.model_names!.includes(m))) return false\n      }\n      if (filters.tool_names && filters.tool_names.length > 0) {\n        if (![...t.tools].some((tn) => filters.tool_names!.includes(tn))) return false\n      }\n      if (filters.start_time_after && t.start_time < filters.start_time_after) return false\n      if (filters.start_time_before && t.start_time > filters.start_time_before) return false\n      return true\n    })\n\n    if (!filters.regex_pattern) return indexedFiltered\n\n    // Opt-in raw-bytes scan — only over the already-narrowed set.\n    const re = compileSearchRegex(filters.regex_pattern)\n    const buf = await this.buffer()\n    const out: TraceIndexEntry[] = []\n    for (const t of indexedFiltered) {\n      let matched = false\n      for (const s of t.spans) {\n        const slice = buf.subarray(s.line_byte_offset, s.line_byte_offset + s.line_byte_length)\n        // Buffer.toString allocates; tolerate it because regex_pattern\n        // is opt-in. Future optimisation: byte-level fast-path for\n        // ASCII-only patterns.\n        if (re.test(slice.toString('utf8'))) {\n          matched = true\n          break\n        }\n      }\n      if (matched) out.push(t)\n    }\n    return out\n  }\n\n  private toSummary(t: TraceIndexEntry): TraceAnalystTraceSummary {\n    return {\n      trace_id: t.trace_id,\n      service_name: t.service_name,\n      agent_name: t.agent_name,\n      span_count: t.span_count,\n      has_errors: t.has_errors,\n      start_time: t.start_time,\n      end_time: t.end_time,\n      duration_ms: t.duration_ms,\n      raw_jsonl_bytes: t.raw_jsonl_bytes,\n      models: [...t.models].sort(),\n      tools: [...t.tools].sort(),\n    }\n  }\n\n  // ─── Span projection (lazy attribute reads) ────────────────────────\n\n  private projectSpan(\n    buf: Buffer,\n    trace_id: string,\n    s: SpanIndexEntry,\n    perAttrCap: number,\n    counter: TruncationCounter,\n  ): TraceAnalystSpan {\n    const slice = buf\n      .subarray(s.line_byte_offset, s.line_byte_offset + s.line_byte_length)\n      .toString('utf8')\n    let raw: Record<string, unknown> = {}\n    try {\n      const parsed = JSON.parse(slice)\n      if (parsed && typeof parsed === 'object') raw = parsed as Record<string, unknown>\n    } catch {\n      // Should not happen — index pre-validated.\n    }\n    const attrs = extractOtlpAttributes(raw)\n    const projected: Record<string, unknown> = {}\n    for (const [k, v] of Object.entries(attrs)) {\n      if (typeof v === 'string') {\n        const trunc = truncateForBudget(v, perAttrCap)\n        if (trunc !== v) counter.value += 1\n        projected[k] = trunc\n      } else if (Array.isArray(v) || (v && typeof v === 'object')) {\n        const json = JSON.stringify(v)\n        const trunc = truncateForBudget(json, perAttrCap)\n        if (trunc !== json) {\n          counter.value += 1\n          projected[k] = trunc\n        } else {\n          projected[k] = v\n        }\n      } else {\n        projected[k] = v\n      }\n    }\n    return {\n      trace_id,\n      span_id: s.span_id,\n      parent_span_id: s.parent_span_id,\n      name: s.name,\n      kind: s.kind,\n      start_time: s.start_time,\n      end_time: s.end_time,\n      duration_ms: s.duration_ms,\n      status: s.status,\n      status_message: s.status_message,\n      service_name: s.service_name,\n      agent_name: s.agent_name,\n      model_name: s.model_name,\n      tool_name: s.tool_name,\n      attributes: projected,\n    }\n  }\n\n  private buildOversizedSummary(\n    t: TraceIndexEntry,\n    span_response_bytes_max: number,\n  ): ViewTraceOversized {\n    const counts = new Map<string, number>()\n    let errorCount = 0\n    for (const s of t.spans) {\n      counts.set(s.name, (counts.get(s.name) ?? 0) + 1)\n      if (s.status === 'ERROR') errorCount += 1\n    }\n    const top = [...counts.entries()].sort((a, b) => b[1] - a[1]).slice(0, 20)\n    return {\n      span_count: t.span_count,\n      top_span_names: top,\n      span_response_bytes_max,\n      error_span_count: errorCount,\n    }\n  }\n\n  private async scanSpanForMatches(\n    buf: Buffer,\n    trace_id: string,\n    s: SpanIndexEntry,\n    re: RegExp,\n    textBudget: number,\n    recordCap: number,\n  ): Promise<{ records: SpanMatchRecord[]; total: number; hasMore: boolean }> {\n    // We scan against the original raw JSONL slice for each span and\n    // record byte positions; the matched_text + context window is\n    // truncated to `textBudget` bytes per record so total tool output\n    // stays bounded even if hits cluster.\n    const slice = buf\n      .subarray(s.line_byte_offset, s.line_byte_offset + s.line_byte_length)\n      .toString('utf8')\n    const records: SpanMatchRecord[] = []\n    const globalRe = new RegExp(re.source, re.flags.includes('g') ? re.flags : `${re.flags}g`)\n    let total = 0\n    let hasMore = false\n    let m: RegExpExecArray | null = globalRe.exec(slice)\n    while (m !== null) {\n      total += 1\n      if (m.index === globalRe.lastIndex) globalRe.lastIndex += 1 // zero-width guard\n      if (records.length >= recordCap) {\n        hasMore = true\n        break\n      }\n      const before = slice.slice(Math.max(0, m.index - textBudget / 2), m.index)\n      const after = slice.slice(\n        m.index + m[0].length,\n        m.index + m[0].length + Math.floor(textBudget / 2),\n      )\n      records.push({\n        trace_id,\n        span_id: s.span_id,\n        span_name: s.name,\n        span_kind: s.kind,\n        attribute_path: bestAttributePathForOffset(slice, m.index) ?? 'span.raw',\n        matched_text: truncateForBudget(m[0], textBudget),\n        context_before: truncateForBudget(before, textBudget),\n        context_after: truncateForBudget(after, textBudget),\n        match_offset: m.index,\n      })\n      m = globalRe.exec(slice)\n    }\n    return { records, total, hasMore }\n  }\n}\n\n// ─── Errors ──────────────────────────────────────────────────────────\n\nexport class TraceFileMissingError extends NotFoundError {\n  constructor(path: string) {\n    super(`trace file not found: ${path}`)\n  }\n}\nexport class TraceFileTooLargeError extends Error {\n  readonly path: string\n  readonly size_bytes: number\n  readonly max_bytes: number\n  constructor(path: string, size_bytes: number, max_bytes: number) {\n    super(\n      `trace file ${path} is ${size_bytes} bytes, over the ${max_bytes}-byte limit; ` +\n        'raise OtlpFileTraceStoreOptions.maxFileBytes or pre-split the file',\n    )\n    this.path = path\n    this.size_bytes = size_bytes\n    this.max_bytes = max_bytes\n  }\n}\nexport class TraceNotFoundError extends NotFoundError {\n  readonly trace_id: string\n  constructor(trace_id: string) {\n    super(`trace not found: ${trace_id}`)\n    this.trace_id = trace_id\n  }\n}\nexport class SpanNotFoundError extends NotFoundError {\n  readonly trace_id: string\n  readonly span_id: string\n  constructor(trace_id: string, span_id: string) {\n    super(`span ${span_id} not found in trace ${trace_id}`)\n    this.trace_id = trace_id\n    this.span_id = span_id\n  }\n}\n\n// ─── OTLP shape readers ──────────────────────────────────────────────\n//\n// The per-line projection lives in `./otlp-span` so the index here and\n// `otlpToRunRecords` read the same vocabulary off the same parser.\n\nfunction isPresent<T>(v: T | undefined): v is T {\n  return v !== undefined\n}\n\n// Per-call truncation counter. Each public read that projects spans\n// owns one of these and threads it through projectSpan; a store-keyed\n// counter would let two concurrent reads on the same store report each\n// other's truncation counts.\ninterface TruncationCounter {\n  value: number\n}\n\n/** A `\"` at `idx` is a real JSON delimiter only when the run of `\\`\n *  immediately preceding it is even-length; an odd run means the quote\n *  is escaped (`\\\"`) and is part of a string value, not a boundary. */\nfunction isUnescapedQuote(slice: string, idx: number): boolean {\n  if (slice[idx] !== '\"') return false\n  let backslashes = 0\n  let b = idx - 1\n  while (b >= 0 && slice[b] === '\\\\') {\n    backslashes += 1\n    b -= 1\n  }\n  return backslashes % 2 === 0\n}\n\n/** Scan backwards from `from` (inclusive) for the nearest UNescaped `\"`.\n *  Returns its index, or -1 when none is found. */\nfunction prevUnescapedQuote(slice: string, from: number): number {\n  for (let i = from; i >= 0; i -= 1) {\n    if (slice[i] === '\"' && isUnescapedQuote(slice, i)) return i\n  }\n  return -1\n}\n\n/**\n * Best-effort: locate the JSON path for the substring at `offset` in\n * a single span's JSONL slice. The slice is '...,\"key\":\"value...\"' — we\n * walk back from `offset` to the value-opening quote, past the `:`, to\n * the key's closing then opening quote, skipping `\\\"`-escaped quotes that\n * live inside string values. Returns `null` when the offset doesn't fall\n * inside a recognisable string field.\n */\nfunction bestAttributePathForOffset(slice: string, offset: number): string | null {\n  // Value-opening quote: nearest unescaped '\"' at or before the offset.\n  const valueQuote = prevUnescapedQuote(slice, Math.min(offset, slice.length - 1))\n  if (valueQuote < 1) return null\n  // The ':' separating key and value sits before the value quote.\n  let j = valueQuote - 1\n  while (j >= 0 && slice[j] !== ':') j -= 1\n  if (j < 1) return null\n  // Key closing quote, then key opening quote — both unescaped.\n  const keyClose = prevUnescapedQuote(slice, j - 1)\n  if (keyClose < 1) return null\n  const keyOpen = prevUnescapedQuote(slice, keyClose - 1)\n  if (keyOpen < 0) return null\n  return slice.slice(keyOpen + 1, keyClose)\n}\n\n// ─── Error-cluster extraction ────────────────────────────────────────\n//\n// Deterministic failure-coverage population. The error-span loop in\n// getOverview already visits every ERROR span; bucketing them by a\n// normalized status_message signature turns \"N error spans\" into \"K\n// distinct failure modes\" — the checklist an analyst must close. No LLM.\n\nconst ERROR_CLUSTER_MAX = 50\nconst ERROR_CLUSTER_EXEMPLARS = 5\nconst SIGNATURE_MAX_CHARS = 160\n\ninterface ErrorClusterAccumulator {\n  signature: string\n  sample: string\n  traceIds: Set<string>\n  spanIds: string[]\n  spanCount: number\n  spanNames: Map<string, number>\n  toolNames: Map<string, number>\n}\n\n/** Collapse volatile tokens so semantically identical failures share a key:\n *  hex/uuid ids → <id>, numbers → #, quoted/abs paths → <path>, durations →\n *  <dur>, whitespace collapsed. Empty/absent messages fall back to the span\n *  name so a no-message error still forms a real cluster. */\nfunction normalizeErrorSignature(message: string | undefined, spanName: string): string {\n  const raw = (message ?? '').trim()\n  const base = raw.length > 0 ? raw : `(${spanName || 'error'} — no message)`\n  const norm = base\n    .replace(/[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}/gi, '<id>')\n    .replace(/\\b[0-9a-f]{12,}\\b/gi, '<id>')\n    .replace(/(?:\\/[\\w.\\-@]+){2,}/g, '<path>')\n    .replace(/\\b\\d+(?:\\.\\d+)?(ms|s|m|h|kb|mb|gb)?\\b/gi, (_m, u) => (u ? `#${u}` : '#'))\n    .replace(/\\s+/g, ' ')\n    .trim()\n  return norm.length > SIGNATURE_MAX_CHARS ? `${norm.slice(0, SIGNATURE_MAX_CHARS)}…` : norm\n}\n\nfunction bump(map: Map<string, number>, key: string | null): void {\n  if (!key) return\n  map.set(key, (map.get(key) ?? 0) + 1)\n}\n\nfunction topKey(map: Map<string, number>): string | null {\n  let best: string | null = null\n  let bestN = 0\n  for (const [k, n] of map)\n    if (n > bestN) {\n      best = k\n      bestN = n\n    }\n  return best\n}\n\nfunction accumulateErrorCluster(\n  clusters: Map<string, ErrorClusterAccumulator>,\n  traceId: string,\n  span: SpanIndexEntry,\n): void {\n  const signature = normalizeErrorSignature(span.status_message, span.name)\n  let acc = clusters.get(signature)\n  if (!acc) {\n    acc = {\n      signature,\n      sample: (span.status_message ?? span.name ?? '').slice(0, 500),\n      traceIds: new Set(),\n      spanIds: [],\n      spanCount: 0,\n      spanNames: new Map(),\n      toolNames: new Map(),\n    }\n    clusters.set(signature, acc)\n  }\n  acc.traceIds.add(traceId)\n  acc.spanCount += 1\n  if (acc.spanIds.length < ERROR_CLUSTER_EXEMPLARS && !acc.spanIds.includes(span.span_id)) {\n    acc.spanIds.push(span.span_id)\n  }\n  bump(acc.spanNames, span.name)\n  bump(acc.toolNames, span.tool_name)\n}\n\nfunction finalizeErrorClusters(\n  clusters: Map<string, ErrorClusterAccumulator>,\n  errorTraceCount: number,\n): ErrorCluster[] {\n  const out = [...clusters.values()].map(\n    (acc): ErrorCluster => ({\n      signature: acc.signature,\n      status_message_sample: acc.sample,\n      span_name: topKey(acc.spanNames),\n      tool_name: topKey(acc.toolNames),\n      trace_count: acc.traceIds.size,\n      span_count: acc.spanCount,\n      prevalence: errorTraceCount > 0 ? acc.traceIds.size / errorTraceCount : 0,\n      exemplar_trace_ids: [...acc.traceIds].slice(0, ERROR_CLUSTER_EXEMPLARS),\n      exemplar_span_ids: acc.spanIds.slice(0, ERROR_CLUSTER_EXEMPLARS),\n    }),\n  )\n  out.sort((a, b) => b.trace_count - a.trace_count || b.span_count - a.span_count)\n  return out.slice(0, ERROR_CLUSTER_MAX)\n}\n","/**\n * `TraceAnalysisStore` — read-side interface the trace-analyst calls\n * through. Six operations, all bounded:\n *\n *   - `getOverview(filters?)` — dataset rollup + sample trace ids.\n *   - `queryTraces(filters?, limit, offset)` — paginated summaries.\n *   - `countTraces(filters?)` — cheap count without materialisation.\n *   - `viewTrace(trace_id, perAttrCap)` — full span list, oversized → summary.\n *   - `viewSpans(trace_id, span_ids, perAttrCap)` — surgical span fetch.\n *   - `searchTrace(trace_id, regex, max_matches)` — bounded regex hits.\n *   - `searchSpan(trace_id, span_id, regex, max_matches)` — single-span search.\n *\n * Multiple implementations ship in the core (`OtlpFileTraceStore`).\n * Downstream callers can supply their own — e.g. a DuckDB-backed\n * adapter or an in-memory adapter for tests — by implementing this\n * interface.\n *\n * Filters compose with AND semantics. Empty/undefined fields impose\n * no constraint. `regex_pattern` is the only opt-in raw-bytes scan —\n * implementations may skip it via `count`/`overview` when not set.\n */\n\nimport type {\n  DatasetOverview,\n  QueryTracesPage,\n  SearchSpanResult,\n  SearchTraceResult,\n  TraceAnalystFilters,\n  ViewSpansResult,\n  ViewTraceResult,\n} from './types'\n\nexport interface TraceAnalysisStore {\n  getOverview(filters?: TraceAnalystFilters): Promise<DatasetOverview>\n\n  queryTraces(opts: {\n    filters?: TraceAnalystFilters\n    limit: number\n    offset?: number\n  }): Promise<QueryTracesPage>\n\n  countTraces(filters?: TraceAnalystFilters): Promise<number>\n\n  viewTrace(opts: {\n    trace_id: string\n    /** Override per-attribute byte cap. Defaults to discovery budget. */\n    per_attribute_byte_cap?: number\n  }): Promise<ViewTraceResult>\n\n  viewSpans(opts: {\n    trace_id: string\n    span_ids: readonly string[]\n    /** Override per-attribute byte cap. Defaults to surgical budget. */\n    per_attribute_byte_cap?: number\n  }): Promise<ViewSpansResult>\n\n  searchTrace(opts: {\n    trace_id: string\n    regex_pattern: string\n    /** Hard cap on matches returned. Default 50. */\n    max_matches?: number\n  }): Promise<SearchTraceResult>\n\n  searchSpan(opts: {\n    trace_id: string\n    span_id: string\n    regex_pattern: string\n    max_matches?: number\n  }): Promise<SearchSpanResult>\n}\n\n/** Compile a regex with the same anchoring + flags semantics across\n *  implementations. Throws on invalid pattern — callers should surface\n *  that to the agent so it can refine instead of looping. */\nexport function compileSearchRegex(pattern: string): RegExp {\n  let source = pattern\n  let flags = 'm'\n  if (source.startsWith('(?i)')) {\n    source = source.slice(4)\n    flags += 'i'\n  }\n  return new RegExp(source, flags)\n}\n\n/** Truncate string payload deterministically for tool responses.\n *  Marker is parseable so downstream consumers can detect truncation\n *  and decide whether to fetch surgically. */\nexport function truncateForBudget(value: string, byteCap: number): string {\n  // We measure UTF-8 byte length conservatively via Buffer.byteLength;\n  // for predictability the truncation point is in CHARS, never inside\n  // a code point.\n  const original = Buffer.byteLength(value, 'utf8')\n  if (original <= byteCap) return value\n\n  // Step back from the cap until we're at a valid char boundary.\n  // Slice by char count proportional to byte ratio, then re-measure.\n  const ratio = byteCap / original\n  let cut = Math.max(0, Math.floor(value.length * ratio))\n  while (cut > 0 && Buffer.byteLength(value.slice(0, cut), 'utf8') > byteCap) {\n    cut -= 1\n  }\n  return `${value.slice(0, cut)}\\n[trace-analyst truncated: original ${original} bytes]`\n}\n","/**\n * Trace-analyst tool surface — six namespaced AxFunctions the analyst\n * agent calls from generated JS code via `traces.<name>(...)`.\n *\n * Discovery → narrow → deep-read protocol. Tool names + ordering\n * support RLM discovery:\n *\n *   1. `getDatasetOverview` (cheap)  — first call, sizes the dataset\n *   2. `queryTraces`                 — paginated summaries with `raw_jsonl_bytes`\n *   3. `countTraces`                 — cheap pre-flight before regex\n *   4. `viewTrace`                   — full span list, oversized → summary\n *   5. `viewSpans`                   — surgical 16KB-cap reads\n *   6. `searchTrace` / `searchSpan`  — bounded regex hits\n *\n * Failure mode. Tool handlers throw on bad input (invalid trace ids,\n * out-of-range pagination, malformed regex). Ax converts thrown errors\n * into actor-visible `[ERROR]` strings so the analyst can adjust on\n * the next turn instead of looping.\n */\n\nimport type { AxFunction } from '@ax-llm/ax'\nimport { f, fn } from '@ax-llm/ax'\n\nimport type { TraceAnalysisStore } from './store'\nimport type { TraceAnalystFilters } from './types'\n\nconst NAMESPACE = 'traces'\n\ninterface BuildTraceAnalystToolsOpts {\n  store: TraceAnalysisStore\n  /** Override the default sample-trace-id slot count (20). Mostly for tests. */\n  sampleTraceLimit?: number\n}\n\nconst filtersField = f\n  .json('Filter set. ALL fields are AND-composed. Leave empty to scan everything.')\n  .optional()\n\n/**\n * Build the trace-analyst function set. Pass the result into\n * `agent(...).functions.local`.\n */\nexport function buildTraceAnalystTools(opts: BuildTraceAnalystToolsOpts): AxFunction[] {\n  const { store } = opts\n\n  const getDatasetOverview = fn('getDatasetOverview')\n    .description(\n      'Dataset rollup: total traces, raw_jsonl_bytes, services, agents, ' +\n        'models, tools, and sample_trace_ids (real ids passable to ' +\n        'view/search). Always call this FIRST without a regex_pattern.',\n    )\n    .namespace(NAMESPACE)\n    .arg('filters', filtersField)\n    .returns(f.json('DatasetOverview'))\n    .handler(async ({ filters }) => store.getOverview(parseFilters(filters)))\n    .build()\n\n  const queryTraces = fn('queryTraces')\n    .description(\n      'Paginated trace summaries. Each summary carries raw_jsonl_bytes — ' +\n        'use it to size traces BEFORE calling viewTrace. Narrow with indexed ' +\n        'filters before adding regex_pattern.',\n    )\n    .namespace(NAMESPACE)\n    .arg('filters', filtersField)\n    .arg('limit', f.number('Page size, 1..200'))\n    .arg('offset', f.number('Page offset; default 0').optional())\n    .returns(f.json('QueryTracesPage'))\n    .handler(async ({ filters, limit, offset }) =>\n      store.queryTraces({\n        filters: parseFilters(filters),\n        limit: assertPageLimit(limit),\n        offset: assertOffset(offset),\n      }),\n    )\n    .build()\n\n  const countTraces = fn('countTraces')\n    .description(\n      'Count traces matching `filters`. Use as a cheap pre-flight ' +\n        'before opting into a regex_pattern scan.',\n    )\n    .namespace(NAMESPACE)\n    .arg('filters', filtersField)\n    .returns(f.number('count'))\n    .handler(async ({ filters }) => store.countTraces(parseFilters(filters)))\n    .build()\n\n  const viewTrace = fn('viewTrace')\n    .description(\n      'Return ALL spans for a single trace, with each attribute capped at ' +\n        '~4KB. If the response would exceed the per-call ceiling the result ' +\n        'carries `oversized` instead of `spans` — DO NOT retry with the same ' +\n        'trace_id; switch to searchTrace / viewSpans.',\n    )\n    .namespace(NAMESPACE)\n    .arg('trace_id', f.string('Real trace id from a prior overview/query'))\n    .returns(f.json('ViewTraceResult'))\n    .handler(async ({ trace_id }) =>\n      store.viewTrace({ trace_id: assertString(trace_id, 'trace_id') }),\n    )\n    .build()\n\n  const viewSpans = fn('viewSpans')\n    .description(\n      'Surgical read of specific spans within a trace, with each ' +\n        'attribute capped at ~16KB (4× the discovery cap). Use after ' +\n        'searchTrace narrows to specific span_ids.',\n    )\n    .namespace(NAMESPACE)\n    .arg('trace_id', f.string('Real trace id'))\n    .arg('span_ids', f.string('Span ids to fetch').array())\n    .returns(f.json('ViewSpansResult'))\n    .handler(async ({ trace_id, span_ids }) =>\n      store.viewSpans({\n        trace_id: assertString(trace_id, 'trace_id'),\n        span_ids: assertStringArray(span_ids, 'span_ids'),\n      }),\n    )\n    .build()\n\n  const searchTrace = fn('searchTrace')\n    .description(\n      'Regex search across all spans of one trace. Returns ' +\n        '`{trace_id, hits: SpanMatchRecord[], total_matches, has_more}`. ' +\n        '**Iterate `result.hits`, NOT `result.matches`** — the field is ' +\n        '`hits`. Each hit has `{span_id, span_name, span_kind, ' +\n        'attribute_path, matched_text, context_before, context_after, ' +\n        'match_offset}`. Bounded regardless of trace size by max_matches ' +\n        '(1..500, default 50). If has_more=true, REFINE the regex rather ' +\n        'than blindly raising max_matches.',\n    )\n    .namespace(NAMESPACE)\n    .arg('trace_id', f.string('Real trace id'))\n    .arg('regex_pattern', f.string('JS-compatible regex, multiline'))\n    .arg('max_matches', f.number('Max records returned, 1..500; default 50').optional())\n    .returns(f.json('SearchTraceResult'))\n    .handler(async ({ trace_id, regex_pattern, max_matches }) =>\n      store.searchTrace({\n        trace_id: assertString(trace_id, 'trace_id'),\n        regex_pattern: assertRegex(regex_pattern),\n        max_matches: assertMaxMatches(max_matches),\n      }),\n    )\n    .build()\n\n  const searchSpan = fn('searchSpan')\n    .description(\n      'Regex search inside a single span. Use when viewSpans returned ' +\n        'a 16KB-truncated payload and you need to narrow further. ' +\n        'Returns `{trace_id, span_id, hits: SpanMatchRecord[], ' +\n        'total_matches, has_more}` — iterate `result.hits`, NOT ' +\n        '`result.matches`.',\n    )\n    .namespace(NAMESPACE)\n    .arg('trace_id', f.string('Real trace id'))\n    .arg('span_id', f.string('Real span id within trace'))\n    .arg('regex_pattern', f.string('JS-compatible regex, multiline'))\n    .arg('max_matches', f.number('Max records, 1..500; default 50').optional())\n    .returns(f.json('SearchSpanResult'))\n    .handler(async ({ trace_id, span_id, regex_pattern, max_matches }) =>\n      store.searchSpan({\n        trace_id: assertString(trace_id, 'trace_id'),\n        span_id: assertString(span_id, 'span_id'),\n        regex_pattern: assertRegex(regex_pattern),\n        max_matches: assertMaxMatches(max_matches),\n      }),\n    )\n    .build()\n\n  return [\n    getDatasetOverview,\n    queryTraces,\n    countTraces,\n    viewTrace,\n    viewSpans,\n    searchTrace,\n    searchSpan,\n  ]\n}\n\n/**\n * Convenience: same shape as `buildTraceAnalystTools` but returns the\n * grouped form expected when registering trace tools alongside other\n * agent function modules. */\nexport function traceAnalystFunctionGroup(opts: BuildTraceAnalystToolsOpts): {\n  namespace: string\n  title: string\n  selectionCriteria: string\n  description: string\n  functions: AxFunction[]\n} {\n  return {\n    namespace: NAMESPACE,\n    title: 'Trace Analysis',\n    selectionCriteria: 'Use for any inspection of OTLP-shaped trace data.',\n    description:\n      'Discovery → narrow → deep-read tools over a JSONL trace dataset. ' +\n      'Always call getDatasetOverview first.',\n    functions: buildTraceAnalystTools(opts),\n  }\n}\n\n// ─── Argument validation ─────────────────────────────────────────────\n\nfunction parseFilters(input: unknown): TraceAnalystFilters | undefined {\n  if (input == null) return undefined\n  if (typeof input !== 'object' || Array.isArray(input)) {\n    throw new TypeError(`filters must be an object, got ${typeof input}`)\n  }\n  const f = input as Record<string, unknown>\n  const out: TraceAnalystFilters = {}\n  if (typeof f.has_errors === 'boolean') out.has_errors = f.has_errors\n  out.service_names = stringArrayOrUndefined(f.service_names, 'service_names')\n  out.agent_names = stringArrayOrUndefined(f.agent_names, 'agent_names')\n  out.model_names = stringArrayOrUndefined(f.model_names, 'model_names')\n  out.tool_names = stringArrayOrUndefined(f.tool_names, 'tool_names')\n  if (typeof f.start_time_after === 'string') out.start_time_after = f.start_time_after\n  if (typeof f.start_time_before === 'string') out.start_time_before = f.start_time_before\n  if (typeof f.regex_pattern === 'string') {\n    if (f.regex_pattern.length === 0) {\n      throw new TypeError('filters.regex_pattern cannot be empty')\n    }\n    out.regex_pattern = f.regex_pattern\n  }\n  return out\n}\n\nfunction stringArrayOrUndefined(v: unknown, label: string): string[] | undefined {\n  if (v === undefined || v === null) return undefined\n  if (!Array.isArray(v)) throw new TypeError(`${label} must be an array of strings`)\n  if (v.some((x) => typeof x !== 'string')) {\n    throw new TypeError(`${label} entries must be strings`)\n  }\n  return v as string[]\n}\n\nfunction assertPageLimit(limit: unknown): number {\n  if (typeof limit !== 'number' || !Number.isInteger(limit) || limit < 1 || limit > 200) {\n    throw new RangeError(`limit must be an integer 1..200`)\n  }\n  return limit\n}\nfunction assertOffset(offset: unknown): number | undefined {\n  if (offset === undefined) return undefined\n  if (typeof offset !== 'number' || !Number.isInteger(offset) || offset < 0) {\n    throw new RangeError(`offset must be a non-negative integer`)\n  }\n  return offset\n}\nfunction assertRegex(pattern: unknown): string {\n  if (typeof pattern !== 'string' || pattern.length === 0) {\n    throw new TypeError(`regex_pattern must be a non-empty string`)\n  }\n  // Compile-and-discard to fail fast — store will recompile, but we\n  // want a deterministic error from the agent's side rather than\n  // a downstream exception.\n  // eslint-disable-next-line no-new\n  new RegExp(pattern, 'm')\n  return pattern\n}\nfunction assertMaxMatches(n: unknown): number | undefined {\n  if (n === undefined) return undefined\n  if (typeof n !== 'number' || !Number.isInteger(n) || n < 1 || n > 500) {\n    throw new RangeError(`max_matches must be an integer 1..500`)\n  }\n  return n\n}\n\nfunction assertString(v: unknown, label: string): string {\n  if (typeof v !== 'string' || v.length === 0) {\n    throw new TypeError(`${label} must be a non-empty string`)\n  }\n  return v\n}\n\nfunction assertStringArray(v: unknown, label: string): string[] {\n  if (!Array.isArray(v)) throw new TypeError(`${label} must be an array of strings`)\n  if (v.some((x) => typeof x !== 'string')) {\n    throw new TypeError(`${label} entries must be strings`)\n  }\n  return v as string[]\n}\n","import {\n  type AxAgentActorTurnCallback,\n  type AxAIService,\n  type AxFunction,\n  AxJSRuntime,\n  agent,\n} from '@ax-llm/ax'\nimport { TraceFileMissingError } from './store-otlp'\n\nexport const TRACE_ANALYSIS_FINAL_TASK = 'Submit the completed trace analysis.'\n\nconst TRACE_ANALYSIS_COMPLETION_INSTRUCTION = `This host consumes your executor result directly; there is no downstream responder for this run. Ignore generic executor guidance that says a responder will format the answer. You must produce the final report and findings yourself.\n\nReturn exactly one executable JavaScript program per turn. Never emit multiple JavaScript or code fences; put every tool call and the final call in that one program.\n\nWhen the analysis is complete, call \\`await final(${JSON.stringify(TRACE_ANALYSIS_FINAL_TASK)}, { report, findings })\\` exactly once. Do not return a one-argument string from \\`final(...)\\`.`\n\nexport class TraceAnalysisTurnLimitError extends Error {\n  readonly analystId: string\n  readonly maxTurns: number\n\n  constructor(analystId: string, maxTurns: number, cause: unknown) {\n    super(\n      `Trace analyst '${analystId}' reached maxTurns=${maxTurns} without a structured final result`,\n      { cause },\n    )\n    this.name = 'TraceAnalysisTurnLimitError'\n    this.analystId = analystId\n    this.maxTurns = maxTurns\n  }\n}\n\nexport interface TraceAnalysisLoopResult<TFinding> {\n  report: string\n  findings: TFinding[]\n  usage: readonly unknown[]\n  chatLog: readonly unknown[]\n  turnCount: number\n}\n\ninterface TraceAnalysisLoopOptions {\n  id: string\n  description: string\n  prompt: string\n  question: string\n  ai: AxAIService\n  model?: string\n  tools: readonly AxFunction[]\n  maxSubqueries: number\n  maxParallelSubqueries: number\n  maxTurns: number\n  maxRuntimeChars: number\n  signal?: AbortSignal\n  onTurn?: AxAgentActorTurnCallback\n}\n\nexport function runTraceAnalysisLoop(\n  options: TraceAnalysisLoopOptions & { findingType: 'string' },\n): Promise<TraceAnalysisLoopResult<string>>\nexport function runTraceAnalysisLoop(\n  options: TraceAnalysisLoopOptions & { findingType: 'object' },\n): Promise<TraceAnalysisLoopResult<Record<string, unknown>>>\nexport async function runTraceAnalysisLoop(\n  options: TraceAnalysisLoopOptions & { findingType: 'string' | 'object' },\n): Promise<TraceAnalysisLoopResult<string | Record<string, unknown>>> {\n  validateLoopLimits(options)\n\n  const config = {\n    agentIdentity: { name: options.id, description: options.description },\n    contextFields: [] as const,\n    runtime: new AxJSRuntime({\n      permissions: [],\n      blockDynamicImport: true,\n      allowedModules: [],\n      freezeIntrinsics: true,\n      blockShadowRealm: true,\n      preventGlobalThisExtensions: false,\n    }),\n    maxSubAgentCalls: options.maxSubqueries,\n    maxTurns: options.maxTurns,\n    maxRuntimeChars: options.maxRuntimeChars,\n    maxBatchedLlmQueryConcurrency: options.maxParallelSubqueries,\n    promptLevel: 'detailed' as const,\n    contextPolicy: { preset: 'full' as const, budget: 'balanced' as const },\n    functions: options.tools,\n    executorOptions: {\n      description: `${options.prompt.trim()}\\n\\n${TRACE_ANALYSIS_COMPLETION_INSTRUCTION}`,\n      ...(options.model ? { model: options.model } : {}),\n      showThoughts: false,\n      thinkingTokenBudget: 'none' as const,\n    },\n    ...(options.onTurn ? { actorTurnCallback: options.onTurn } : {}),\n    bubbleErrors: [TraceFileMissingError],\n  }\n  const analyst =\n    options.findingType === 'string'\n      ? agent('question:string -> report:string, findings:string[]', config)\n      : agent('question:string -> report:string, findings:json[]', config)\n\n  const state = await analyst.executor.run(\n    options.ai,\n    { question: options.question },\n    options.signal ? { abortSignal: options.signal } : undefined,\n  )\n\n  let completed: CompletedTraceAnalysis<string | Record<string, unknown>>\n  try {\n    completed =\n      options.findingType === 'string'\n        ? readTraceAnalysisCompletion(state.executorResult, 'string')\n        : readTraceAnalysisCompletion(state.executorResult, 'object')\n  } catch (error) {\n    if (state.turnCount >= options.maxTurns) {\n      throw new TraceAnalysisTurnLimitError(options.id, options.maxTurns, error)\n    }\n    throw new Error(`Trace analyst '${options.id}' stopped without a structured final result`, {\n      cause: error,\n    })\n  }\n\n  return {\n    ...completed,\n    usage: analyst.executor.getUsage(),\n    chatLog: analyst.executor.getChatLog(),\n    turnCount: state.turnCount,\n  }\n}\n\ninterface CompletedTraceAnalysis<TFinding> {\n  report: string\n  findings: TFinding[]\n}\n\nexport function readTraceAnalysisCompletion(\n  value: unknown,\n  findingType: 'string',\n): CompletedTraceAnalysis<string>\nexport function readTraceAnalysisCompletion(\n  value: unknown,\n  findingType: 'object',\n): CompletedTraceAnalysis<Record<string, unknown>>\nexport function readTraceAnalysisCompletion(\n  value: unknown,\n  findingType: 'string' | 'object',\n): CompletedTraceAnalysis<string | Record<string, unknown>> {\n  if (!value || typeof value !== 'object') {\n    throw new Error('Trace analyst did not return a structured final result')\n  }\n  const completion = value as { type?: unknown; args?: unknown }\n  if (\n    completion.type !== 'final' ||\n    !Array.isArray(completion.args) ||\n    completion.args.length !== 2 ||\n    completion.args[0] !== TRACE_ANALYSIS_FINAL_TASK\n  ) {\n    throw new Error('Trace analyst did not return a structured final result')\n  }\n\n  const payload = completion.args[1]\n  if (!payload || typeof payload !== 'object') {\n    throw new Error('Trace analyst final result must contain report and findings')\n  }\n  const { report, findings } = payload as { report?: unknown; findings?: unknown }\n  if (typeof report !== 'string' || !Array.isArray(findings)) {\n    throw new Error('Trace analyst final result must contain report and findings')\n  }\n  if (findingType === 'string') {\n    if (findings.some((finding) => typeof finding !== 'string')) {\n      throw new Error('Trace analyst final result must contain string findings')\n    }\n    return { report, findings: findings as string[] }\n  }\n  if (\n    findings.some((finding) => !finding || typeof finding !== 'object' || Array.isArray(finding))\n  ) {\n    throw new Error('Trace analyst final result must contain object findings')\n  }\n  return { report, findings: findings as Record<string, unknown>[] }\n}\n\nfunction validateLoopLimits(options: TraceAnalysisLoopOptions): void {\n  if (!Number.isSafeInteger(options.maxSubqueries) || options.maxSubqueries < 0) {\n    throw new TypeError('maxSubqueries must be a non-negative integer')\n  }\n  if (!Number.isSafeInteger(options.maxParallelSubqueries) || options.maxParallelSubqueries < 1) {\n    throw new TypeError('maxParallelSubqueries must be a positive integer')\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;AAgBA,IAAM,2BAA2B;AAAA,EAC/B;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF;AAuBA,IAAM,sBAAsB,oBAAI,IAAkB;AAAA,EAChD;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF,CAAC;AAOM,SAAS,qBAAqB,OAAwC;AAC3E,QAAM,eAAe,MAAM,QAAQ,qBAAqB,MAAM,YAAY,mBAAmB;AAC7F,MAAI,cAAc;AAChB,UAAM,aAAa,aAAa,YAAY;AAC5C,QAAI,oBAAoB,IAAI,UAAU,EAAG,QAAO;AAAA,EAClD;AAEA,MACE,qBAAqB,MAAM,YAAY,mBAAmB,MAAM,UAChE,2BAA2B,KAAK,MAAM,IAAI,GAC1C;AACA,WAAO;AAAA,EACT;AAEA,QAAM,WAAW,MAAM,WAAW,WAAW;AAC7C,MACG,OAAO,aAAa,YAAY,SAAS,YAAY,MAAM,iBAC5D,4DAA4D,KAAK,MAAM,IAAI,KAC3E,qBAAqB,MAAM,YAAY,mBAAmB,MAAM,UAChE,OAAO,MAAM,WAAW,uBAAuB,MAAM,UACrD;AACA,WAAO;AAAA,EACT;AAEA,SAAO;AACT;AAEO,SAAS,gBAAgB,OAAmC;AACjE,SAAO,qBAAqB,KAAK,MAAM;AACzC;AAEA,SAAS,uBACP,MAC2C;AAC3C,QAAM,eAAe,KAAK,iBAAiB;AAC3C,QAAM,aAAwD;AAAA,IAC5D,CAAC,SAAS,GAAG,KAAK;AAAA,IAClB,CAAC,kBAAkB,GAAG;AAAA,EACxB;AACA,MAAI,KAAK,cAAc,OAAW,YAAW,eAAe,IAAI,KAAK;AACrE,MAAI,aAAc,YAAW,WAAW,IAAI,oBAAoB,KAAK,IAAI;AACzE,MAAI,KAAK,WAAW,OAAW,YAAW,YAAY,IAAI,oBAAoB,KAAK,MAAM;AACzF,SAAO;AACT;AAEO,SAAS,4BACd,YACA,MACM;AACN,aAAW,OAAO,yBAA0B,QAAO,WAAW,GAAG;AACjE,SAAO,OAAO,YAAY,uBAAuB,IAAI,CAAC;AACxD;AAEO,SAAS,iCAAiC,MAAsB;AACrE,UAAQ,MAAM;AAAA,IACZ,KAAK;AACH,aAAO;AAAA,IACT,KAAK;AACH,aAAO;AAAA,IACT,KAAK;AACH,aAAO;AAAA,IACT,KAAK;AACH,aAAO;AAAA,IACT,KAAK;AACH,aAAO;AAAA,IACT,KAAK;AACH,aAAO;AAAA,IACT;AACE,aAAO;AAAA,EACX;AACF;AAEA,SAAS,qBACP,YACA,MACoB;AACpB,aAAW,OAAO,MAAM;AACtB,UAAM,QAAQ,WAAW,GAAG;AAC5B,QAAI,OAAO,UAAU,YAAY,MAAM,SAAS,EAAG,QAAO;AAAA,EAC5D;AACA,SAAO;AACT;AAEA,SAAS,oBAAoB,OAAwB;AACnD,MAAI,UAAU,OAAW,QAAO;AAChC,MAAI,OAAO,UAAU,SAAU,QAAO;AACtC,MAAI;AACF,WAAO,KAAK,UAAU,KAAK,KAAK,OAAO,KAAK;AAAA,EAC9C,QAAQ;AACN,WAAO,OAAO,KAAK;AAAA,EACrB;AACF;;;AC5FO,SAAS,oBAAoB,KAAwD;AAC1F,QAAM,WAAW,YAAY,KAAK,UAAU,KAAK,YAAY,KAAK,SAAS;AAC3E,QAAM,UAAU,YAAY,KAAK,SAAS,KAAK,YAAY,KAAK,QAAQ;AACxE,MAAI,CAAC,YAAY,CAAC,QAAS,QAAO;AAElC,QAAM,cAAc,YAAY,KAAK,gBAAgB,KAAK,YAAY,KAAK,cAAc,KAAK;AAC9F,QAAM,YAAY,sBAAsB,UAAU,SAAS,WAAW;AACtE,QAAM,OAAO,YAAY,KAAK,MAAM,KAAK;AACzC,QAAM,aAAa,YAAY,KAAK,YAAY,KAAK,YAAY,KAAK,WAAW,KAAK;AACtF,QAAM,WAAW,YAAY,KAAK,UAAU,KAAK,YAAY,KAAK,SAAS,KAAK;AAEhF,QAAM,SAAS,eAAe,GAAG;AACjC,QAAM,aAAa,sBAAsB,GAAG;AAE5C,QAAM,eACJ,SAAS,WAAW,cAAc,CAAC,KACnC,SAAS,WAAW,kCAAkC,CAAC,KACvD;AACF,QAAM,aACJ,SAAS,WAAW,YAAY,CAAC,KACjC,SAAS,WAAW,sBAAsB,CAAC,KAC3C,SAAS,WAAW,sBAAsB,CAAC,KAC3C;AACF,QAAM,aAAa,gBAAgB,YAAY,mBAAmB;AAClE,QAAM,YAAY,gBAAgB,YAAY,mBAAmB;AAEjE,QAAM,OAAO,cAAc,UAAU;AAErC,MAAI,cAAc;AAClB,MAAI,cAAc,UAAU;AAC1B,UAAM,IAAI,gBAAgB,UAAU;AACpC,UAAM,IAAI,gBAAgB,QAAQ;AAClC,QAAI,MAAM,QAAQ,MAAM,KAAM,eAAc,KAAK,IAAI,GAAG,IAAI,CAAC;AAAA,EAC/D;AAEA,SAAO;AAAA,IACL;AAAA,IACA;AAAA,IACA,gBAAgB,aAAa,UAAU,SAAS,IAAI,YAAY;AAAA,IAChE;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA,QAAQ,OAAO;AAAA,IACf,gBAAgB,OAAO;AAAA,IACvB;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AACF;AAEA,SAAS,sBACP,SACA,QACA,UACe;AACf,MAAI,CAAC,SAAU,QAAO;AACtB,QAAM,SAAS,GAAG,OAAO;AACzB,SAAO,OAAO,WAAW,MAAM,KAAK,CAAC,SAAS,WAAW,MAAM,IAC3D,GAAG,MAAM,GAAG,QAAQ,KACpB;AACN;AAEO,SAAS,eAAe,KAG7B;AACA,QAAM,SAAS,IAAI;AACnB,MAAI,UAAU,OAAO,WAAW,YAAY,CAAC,MAAM,QAAQ,MAAM,GAAG;AAClE,UAAM,UAAW,OAAmC;AACpD,UAAM,OACJ,YAAY,oBAAoB,YAAY,OACxC,OACA,YAAY,uBAAuB,YAAY,UAC7C,UACA;AACR,UAAM,aAAc,OAAmC;AACvD,UAAM,UAAU,OAAO,eAAe,YAAY,WAAW,SAAS,IAAI,aAAa;AACvF,WAAO,EAAE,MAAM,QAAQ;AAAA,EACzB;AACA,SAAO,EAAE,MAAM,SAAS,SAAS,OAAU;AAC7C;AAEO,SAAS,cAAc,OAAsD;AAClF,QAAM,OAAO,gBAAgB,OAAO,mBAAmB;AACvD,MAAI,MAAM;AACR,UAAM,QAAQ,KAAK,YAAY;AAC/B,QACE,UAAU,WACV,UAAU,SACV,UAAU,UACV,UAAU,WACV,UAAU,eACV,UAAU,eACV,UAAU,QACV;AACA,aAAO;AAAA,IACT;AAAA,EACF;AACA,SAAO;AACT;AAOO,SAAS,sBAAsB,KAAuD;AAC3F,QAAM,MAA+B,CAAC;AACtC,QAAM,WAAW,IAAI;AACrB,MAAI,YAAY,OAAO,aAAa,YAAY,CAAC,MAAM,QAAQ,QAAQ,GAAG;AACxE,UAAM,KAAM,SAAqC;AACjD,QAAI,MAAM,OAAO,OAAO,YAAY,CAAC,MAAM,QAAQ,EAAE,GAAG;AACtD,iBAAW,CAAC,GAAG,CAAC,KAAK,OAAO,QAAQ,EAA6B,GAAG;AAClE,YAAI,CAAC,IAAI;AAAA,MACX;AAAA,IACF;AAAA,EACF;AACA,QAAM,YAAY,IAAI;AACtB,MAAI,aAAa,OAAO,cAAc,YAAY,CAAC,MAAM,QAAQ,SAAS,GAAG;AAC3E,eAAW,CAAC,GAAG,CAAC,KAAK,OAAO,QAAQ,SAAoC,GAAG;AACzE,UAAI,CAAC,IAAI;AAAA,IACX;AAAA,EACF;AACA,SAAO;AACT;AAEO,SAAS,YAAY,KAA8B,KAAiC;AACzF,QAAM,IAAI,IAAI,GAAG;AACjB,SAAO,OAAO,MAAM,WAAW,IAAI;AACrC;AAEO,SAAS,SAAS,GAA2B;AAClD,SAAO,OAAO,MAAM,YAAY,EAAE,SAAS,IAAI,IAAI;AACrD;AAGO,SAAS,gBACd,OACA,MACe;AACf,aAAW,KAAK,MAAM;AACpB,UAAM,IAAI,SAAS,MAAM,CAAC,CAAC;AAC3B,QAAI,MAAM,KAAM,QAAO;AAAA,EACzB;AACA,SAAO;AACT;AAQO,SAAS,gBAAgB,IAA8C;AAC5E,MAAI,CAAC,GAAI,QAAO;AAChB,MAAI,QAAQ,KAAK,EAAE,EAAG,QAAO,OAAO,EAAE;AACtC,QAAM,IAAI,KAAK,MAAM,EAAE;AACvB,SAAO,OAAO,MAAM,CAAC,IAAI,OAAO;AAClC;AAOO,SAAS,gBAAgB,GAAW,GAAmB;AAC5D,UAAQ,gBAAgB,CAAC,KAAK,MAAM,gBAAgB,CAAC,KAAK;AAC5D;;;ACzBO,IAAM,gCAAyD;AAAA,EACpE,oBAAoB;AAAA,EACpB,wBAAwB;AAAA,EACxB,wBAAwB;AAAA,EACxB,oBAAoB;AACtB;AAIO,IAAM,yCAAyC;;;AC3LtD,SAAS,UAAU,YAAY;;;AC+CxB,SAAS,mBAAmB,SAAyB;AAC1D,MAAI,SAAS;AACb,MAAI,QAAQ;AACZ,MAAI,OAAO,WAAW,MAAM,GAAG;AAC7B,aAAS,OAAO,MAAM,CAAC;AACvB,aAAS;AAAA,EACX;AACA,SAAO,IAAI,OAAO,QAAQ,KAAK;AACjC;AAKO,SAAS,kBAAkB,OAAe,SAAyB;AAIxE,QAAM,WAAW,OAAO,WAAW,OAAO,MAAM;AAChD,MAAI,YAAY,QAAS,QAAO;AAIhC,QAAM,QAAQ,UAAU;AACxB,MAAI,MAAM,KAAK,IAAI,GAAG,KAAK,MAAM,MAAM,SAAS,KAAK,CAAC;AACtD,SAAO,MAAM,KAAK,OAAO,WAAW,MAAM,MAAM,GAAG,GAAG,GAAG,MAAM,IAAI,SAAS;AAC1E,WAAO;AAAA,EACT;AACA,SAAO,GAAG,MAAM,MAAM,GAAG,GAAG,CAAC;AAAA,qCAAwC,QAAQ;AAC/E;;;AD7CA,IAAM,oBAAoB;AAG1B,SAAS,mBAAkC;AACzC,SAAO,IAAI,QAAQ,CAAC,YAAY,aAAa,OAAO,CAAC;AACvD;AAmEO,IAAM,+BAA+B,MAAM,OAAO;AAElD,IAAM,qBAAN,MAAuD;AAAA,EAC3C;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACT;AAAA;AAAA;AAAA;AAAA,EAIA;AAAA,EAER,YAAY,MAAiC;AAC3C,SAAK,OAAO,KAAK;AACjB,SAAK,yBACH,KAAK,0BAA0B,8BAA8B;AAC/D,SAAK,yBACH,KAAK,0BAA0B,8BAA8B;AAC/D,SAAK,qBACH,KAAK,sBAAsB,8BAA8B;AAC3D,SAAK,qBACH,KAAK,sBAAsB,8BAA8B;AAC3D,SAAK,eAAe,KAAK,gBAAgB;AAAA,EAC3C;AAAA;AAAA,EAIA,MAAM,YAAY,SAAyD;AACzE,UAAM,MAAM,MAAM,KAAK,MAAM;AAC7B,UAAM,UAAU,MAAM,KAAK,cAAc,KAAK,OAAO;AAErD,UAAM,WAAW,oBAAI,IAAY;AACjC,UAAM,SAAS,oBAAI,IAAY;AAC/B,UAAM,SAAS,oBAAI,IAAY;AAC/B,UAAM,QAAQ,oBAAI,IAAY;AAC9B,QAAI,WAAW;AACf,QAAI,WAA0B;AAC9B,QAAI,SAAwB;AAC5B,QAAI,kBAAkB;AACtB,QAAI,iBAAiB;AACrB,UAAM,WAAW,oBAAI,IAAqC;AAE1D,eAAW,KAAK,SAAS;AACvB,UAAI,EAAE,aAAc,UAAS,IAAI,EAAE,YAAY;AAC/C,UAAI,EAAE,WAAY,QAAO,IAAI,EAAE,UAAU;AACzC,iBAAW,KAAK,EAAE,OAAQ,QAAO,IAAI,CAAC;AACtC,iBAAW,MAAM,EAAE,MAAO,OAAM,IAAI,EAAE;AACtC,kBAAY,EAAE;AACd,UAAI,CAAC,YAAY,gBAAgB,EAAE,YAAY,QAAQ,IAAI,EAAG,YAAW,EAAE;AAC3E,UAAI,CAAC,UAAU,gBAAgB,EAAE,UAAU,MAAM,IAAI,EAAG,UAAS,EAAE;AACnE,UAAI,EAAE,YAAY;AAChB,2BAAmB;AACnB,mBAAW,KAAK,EAAE,OAAO;AACvB,cAAI,EAAE,WAAW,QAAS;AAC1B,4BAAkB;AAClB,iCAAuB,UAAU,EAAE,UAAU,CAAC;AAAA,QAChD;AAAA,MACF;AAAA,IACF;AAEA,UAAM,mBAAmB,QAAQ,MAAM,GAAG,EAAE,EAAE,IAAI,CAAC,MAAM,EAAE,QAAQ;AACnE,WAAO;AAAA,MACL,cAAc,QAAQ;AAAA,MACtB,iBAAiB;AAAA,MACjB,UAAU,CAAC,GAAG,QAAQ,EAAE,KAAK;AAAA,MAC7B,QAAQ,CAAC,GAAG,MAAM,EAAE,KAAK;AAAA,MACzB,QAAQ,CAAC,GAAG,MAAM,EAAE,KAAK;AAAA,MACzB,YAAY,CAAC,GAAG,KAAK,EAAE,KAAK;AAAA,MAC5B;AAAA,MACA,QAAQ,EAAE,aAAa,iBAAiB,YAAY,eAAe;AAAA,MACnE,gBAAgB,sBAAsB,UAAU,eAAe;AAAA,MAC/D,YAAY,YAAY,SAAS,EAAE,UAAU,OAAO,IAAI;AAAA,IAC1D;AAAA,EACF;AAAA,EAEA,MAAM,YAAY,MAIW;AAC3B,QAAI,CAAC,OAAO,UAAU,KAAK,KAAK,KAAK,KAAK,QAAQ,KAAK,KAAK,QAAQ,KAAK;AACvE,YAAM,IAAI,WAAW,yCAAyC,KAAK,KAAK,EAAE;AAAA,IAC5E;AACA,UAAM,SAAS,KAAK,UAAU;AAC9B,QAAI,CAAC,OAAO,UAAU,MAAM,KAAK,SAAS,GAAG;AAC3C,YAAM,IAAI,WAAW,uCAAuC,MAAM,EAAE;AAAA,IACtE;AAEA,UAAM,MAAM,MAAM,KAAK,MAAM;AAC7B,UAAM,UAAU,MAAM,KAAK,cAAc,KAAK,KAAK,OAAO;AAC1D,UAAM,QAAQ,QAAQ,MAAM,QAAQ,SAAS,KAAK,KAAK;AACvD,WAAO;AAAA,MACL,QAAQ,MAAM,IAAI,CAAC,MAAM,KAAK,UAAU,CAAC,CAAC;AAAA,MAC1C,OAAO,QAAQ;AAAA,MACf,UAAU,SAAS,MAAM,SAAS,QAAQ;AAAA,IAC5C;AAAA,EACF;AAAA,EAEA,MAAM,YAAY,SAAgD;AAChE,UAAM,MAAM,MAAM,KAAK,MAAM;AAC7B,UAAM,UAAU,MAAM,KAAK,cAAc,KAAK,OAAO;AACrD,WAAO,QAAQ;AAAA,EACjB;AAAA,EAEA,MAAM,UAAU,MAGa;AAC3B,UAAM,MAAM,MAAM,KAAK,MAAM;AAC7B,UAAM,QAAQ,IAAI,QAAQ,IAAI,KAAK,QAAQ;AAC3C,QAAI,CAAC,OAAO;AACV,YAAM,IAAI,mBAAmB,KAAK,QAAQ;AAAA,IAC5C;AACA,UAAM,MAAM,KAAK,0BAA0B,KAAK;AAKhD,UAAM,MAAM,MAAM,KAAK,OAAO;AAC9B,UAAM,QAA4B,CAAC;AACnC,QAAI,eAAe;AACnB,QAAI,0BAA0B;AAC9B,UAAM,UAA6B,EAAE,OAAO,EAAE;AAC9C,eAAW,KAAK,MAAM,OAAO;AAC3B,YAAM,YAAY,KAAK,YAAY,KAAK,MAAM,UAAU,GAAG,KAAK,OAAO;AACvE,YAAM,QAAQ,OAAO,WAAW,KAAK,UAAU,SAAS,GAAG,MAAM;AACjE,gCAA0B,KAAK,IAAI,yBAAyB,KAAK;AACjE,sBAAgB;AAChB,UAAI,eAAe,KAAK,oBAAoB;AAC1C,eAAO;AAAA,UACL,UAAU,MAAM;AAAA,UAChB,WAAW,KAAK,sBAAsB,OAAO,uBAAuB;AAAA,QACtE;AAAA,MACF;AACA,YAAM,KAAK,SAAS;AAAA,IACtB;AACA,WAAO,EAAE,UAAU,MAAM,UAAU,MAAM;AAAA,EAC3C;AAAA,EAEA,MAAM,UAAU,MAIa;AAC3B,UAAM,MAAM,MAAM,KAAK,MAAM;AAC7B,UAAM,QAAQ,IAAI,QAAQ,IAAI,KAAK,QAAQ;AAC3C,QAAI,CAAC,MAAO,OAAM,IAAI,mBAAmB,KAAK,QAAQ;AACtD,QAAI,KAAK,SAAS,WAAW,GAAG;AAC9B,aAAO;AAAA,QACL,UAAU,MAAM;AAAA,QAChB,OAAO,CAAC;AAAA,QACR,kBAAkB,CAAC;AAAA,QACnB,2BAA2B;AAAA,MAC7B;AAAA,IACF;AACA,QAAI,KAAK,SAAS,SAAS,KAAK;AAC9B,YAAM,IAAI,WAAW,sCAAsC,KAAK,SAAS,MAAM,EAAE;AAAA,IACnF;AACA,UAAM,MAAM,KAAK,0BAA0B,KAAK;AAEhD,UAAM,UAAU,IAAI,IAAI,KAAK,QAAQ;AACrC,UAAM,QAAQ,MAAM,MAAM,OAAO,CAAC,MAAM,QAAQ,IAAI,EAAE,OAAO,CAAC;AAC9D,UAAM,UAAU,KAAK,SAAS,OAAO,CAAC,OAAO,CAAC,MAAM,KAAK,CAACA,OAAMA,GAAE,YAAY,EAAE,CAAC;AAEjF,UAAM,MAAM,MAAM,KAAK,OAAO;AAC9B,UAAM,QAA4B,CAAC;AACnC,UAAM,UAA6B,EAAE,OAAO,EAAE;AAC9C,QAAI,eAAe;AACnB,eAAW,KAAK,OAAO;AACrB,YAAM,YAAY,KAAK,YAAY,KAAK,MAAM,UAAU,GAAG,KAAK,OAAO;AACvE,YAAM,QAAQ,OAAO,WAAW,KAAK,UAAU,SAAS,GAAG,MAAM;AACjE,sBAAgB;AAChB,UAAI,eAAe,KAAK,oBAAoB;AAG1C;AAAA,MACF;AACA,YAAM,KAAK,SAAS;AAAA,IACtB;AACA,WAAO;AAAA,MACL,UAAU,MAAM;AAAA,MAChB;AAAA,MACA,kBAAkB;AAAA,MAClB,2BAA2B,QAAQ;AAAA,IACrC;AAAA,EACF;AAAA,EAEA,MAAM,YAAY,MAIa;AAC7B,UAAM,cAAc,KAAK,eAAe;AACxC,QAAI,CAAC,OAAO,UAAU,WAAW,KAAK,cAAc,KAAK,cAAc,KAAK;AAC1E,YAAM,IAAI,WAAW,+CAA+C,WAAW,EAAE;AAAA,IACnF;AACA,UAAM,MAAM,MAAM,KAAK,MAAM;AAC7B,UAAM,QAAQ,IAAI,QAAQ,IAAI,KAAK,QAAQ;AAC3C,QAAI,CAAC,MAAO,OAAM,IAAI,mBAAmB,KAAK,QAAQ;AACtD,UAAM,KAAK,mBAAmB,KAAK,aAAa;AAEhD,UAAM,MAAM,MAAM,KAAK,OAAO;AAC9B,UAAM,OAA0B,CAAC;AACjC,QAAI,QAAQ;AACZ,QAAI,SAAS;AACb,eAAW,KAAK,MAAM,OAAO;AAC3B,YAAM,YAAY,cAAc,KAAK;AACrC,YAAM,YAAY,MAAM,KAAK;AAAA,QAC3B;AAAA,QACA,MAAM;AAAA,QACN;AAAA,QACA;AAAA,QACA,KAAK;AAAA,QACL;AAAA,MACF;AACA,eAAS,UAAU;AACnB,iBAAW,KAAK,UAAU,SAAS;AACjC,YAAI,KAAK,UAAU,YAAa;AAChC,aAAK,KAAK,CAAC;AAAA,MACb;AACA,UAAI,KAAK,UAAU,aAAa;AAK9B,iBAAS;AACT;AAAA,MACF;AAAA,IACF;AACA,WAAO;AAAA,MACL,UAAU,MAAM;AAAA,MAChB;AAAA;AAAA,MAEA,eAAe,SAAS,KAAK,SAAS;AAAA,MACtC,UAAU,UAAU,QAAQ,KAAK;AAAA,IACnC;AAAA,EACF;AAAA,EAEA,MAAM,WAAW,MAKa;AAC5B,UAAM,cAAc,KAAK,eAAe;AACxC,QAAI,CAAC,OAAO,UAAU,WAAW,KAAK,cAAc,KAAK,cAAc,KAAK;AAC1E,YAAM,IAAI,WAAW,8CAA8C,WAAW,EAAE;AAAA,IAClF;AACA,UAAM,MAAM,MAAM,KAAK,MAAM;AAC7B,UAAM,QAAQ,IAAI,QAAQ,IAAI,KAAK,QAAQ;AAC3C,QAAI,CAAC,MAAO,OAAM,IAAI,mBAAmB,KAAK,QAAQ;AACtD,UAAM,OAAO,MAAM,MAAM,KAAK,CAAC,MAAM,EAAE,YAAY,KAAK,OAAO;AAC/D,QAAI,CAAC,MAAM;AACT,YAAM,IAAI,kBAAkB,KAAK,UAAU,KAAK,OAAO;AAAA,IACzD;AACA,UAAM,KAAK,mBAAmB,KAAK,aAAa;AAChD,UAAM,MAAM,MAAM,KAAK,OAAO;AAC9B,UAAM,YAAY,MAAM,KAAK;AAAA,MAC3B;AAAA,MACA,MAAM;AAAA,MACN;AAAA,MACA;AAAA,MACA,KAAK;AAAA,MACL;AAAA,IACF;AACA,WAAO;AAAA,MACL,UAAU,MAAM;AAAA,MAChB,SAAS,KAAK;AAAA,MACd,MAAM,UAAU;AAAA,MAChB,eAAe,UAAU;AAAA,MACzB,UAAU,UAAU,QAAQ,UAAU,QAAQ;AAAA,IAChD;AAAA,EACF;AAAA;AAAA;AAAA;AAAA,EAMA,MAAM,gBAA+B;AACnC,UAAM,KAAK,MAAM;AAAA,EACnB;AAAA,EAEA,MAAc,SAA0B;AACtC,QAAI,CAAC,KAAK,eAAe;AACvB,WAAK,gBAAgB,KAAK,YAAY;AAAA,IACxC;AACA,WAAO,KAAK;AAAA,EACd;AAAA;AAAA;AAAA;AAAA,EAKA,MAAc,cAA+B;AAC3C,QAAI;AACJ,QAAI;AACF,cAAQ,MAAM,KAAK,KAAK,IAAI;AAAA,IAC9B,SAAS,KAAK;AACZ,UAAK,KAA+B,SAAS,UAAU;AACrD,cAAM,IAAI,sBAAsB,KAAK,IAAI;AAAA,MAC3C;AACA,YAAM;AAAA,IACR;AACA,QAAI,MAAM,OAAO,KAAK,cAAc;AAClC,YAAM,IAAI,uBAAuB,KAAK,MAAM,MAAM,MAAM,KAAK,YAAY;AAAA,IAC3E;AACA,WAAO,SAAS,KAAK,IAAI;AAAA,EAC3B;AAAA,EAEA,MAAc,QAA+B;AAC3C,QAAI,CAAC,KAAK,cAAc;AACtB,WAAK,eAAe,KAAK,WAAW;AAAA,IACtC;AACA,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,MAAc,aAAoC;AAEhD,UAAM,MAAM,MAAM,KAAK,OAAO;AAE9B,UAAM,UAAU,oBAAI,IAA6B;AACjD,QAAI,SAAS;AACb,QAAI,aAAa;AACjB,WAAO,SAAS,IAAI,QAAQ;AAG1B,UAAI,EAAE,cAAc,mBAAmB;AACrC,qBAAa;AACb,cAAM,iBAAiB;AAAA,MACzB;AACA,YAAM,eAAe,IAAI,QAAQ,IAAM,MAAM;AAC7C,YAAM,UAAU,iBAAiB,KAAK,IAAI,SAAS;AACnD,YAAM,aAAa,UAAU;AAC7B,UAAI,eAAe,GAAG;AACpB,iBAAS,UAAU;AACnB;AAAA,MACF;AACA,YAAM,YAAY,IAAI,SAAS,QAAQ,OAAO,EAAE,SAAS,MAAM;AAC/D,YAAM,aAAa;AACnB,eAAS,UAAU;AAEnB,UAAI;AACJ,UAAI;AACF,iBAAS,KAAK,MAAM,SAAS;AAAA,MAC/B,QAAQ;AAGN;AAAA,MACF;AACA,UAAI,CAAC,UAAU,OAAO,WAAW,SAAU;AAC3C,YAAM,OAAO,oBAAoB,MAAiC;AAClE,UAAI,CAAC,KAAM;AAEX,UAAI,QAAQ,QAAQ,IAAI,KAAK,QAAQ;AACrC,UAAI,CAAC,OAAO;AACV,gBAAQ;AAAA,UACN,UAAU,KAAK;AAAA,UACf,cAAc,KAAK;AAAA,UACnB,YAAY,KAAK;AAAA,UACjB,YAAY;AAAA,UACZ,YAAY;AAAA,UACZ,YAAY,KAAK;AAAA,UACjB,UAAU,KAAK;AAAA,UACf,aAAa;AAAA,UACb,iBAAiB;AAAA,UACjB,QAAQ,oBAAI,IAAI;AAAA,UAChB,OAAO,oBAAI,IAAI;AAAA,UACf,OAAO,CAAC;AAAA,QACV;AACA,gBAAQ,IAAI,KAAK,UAAU,KAAK;AAAA,MAClC,OAAO;AAGL,YAAI,CAAC,MAAM,gBAAgB,KAAK,aAAc,OAAM,eAAe,KAAK;AACxE,YAAI,CAAC,MAAM,cAAc,KAAK,WAAY,OAAM,aAAa,KAAK;AAAA,MACpE;AAEA,YAAM,aAA6B;AAAA,QACjC,SAAS,KAAK;AAAA,QACd,gBAAgB,KAAK;AAAA,QACrB,MAAM,KAAK;AAAA,QACX,MAAM,KAAK;AAAA,QACX,YAAY,KAAK;AAAA,QACjB,UAAU,KAAK;AAAA,QACf,aAAa,KAAK;AAAA,QAClB,QAAQ,KAAK;AAAA,QACb,gBAAgB,KAAK;AAAA,QACrB,cAAc,KAAK;AAAA,QACnB,YAAY,KAAK;AAAA,QACjB,YAAY,KAAK;AAAA,QACjB,WAAW,KAAK;AAAA,QAChB,kBAAkB;AAAA,QAClB,kBAAkB;AAAA,MACpB;AACA,YAAM,MAAM,KAAK,UAAU;AAC3B,YAAM,cAAc;AACpB,YAAM,mBAAmB,aAAa;AACtC,UAAI,KAAK,WAAW,QAAS,OAAM,aAAa;AAChD,UAAI,gBAAgB,KAAK,YAAY,MAAM,UAAU,IAAI,EAAG,OAAM,aAAa,KAAK;AACpF,UAAI,gBAAgB,KAAK,UAAU,MAAM,QAAQ,IAAI,EAAG,OAAM,WAAW,KAAK;AAC9E,UAAI,KAAK,WAAY,OAAM,OAAO,IAAI,KAAK,UAAU;AACrD,UAAI,KAAK,UAAW,OAAM,MAAM,IAAI,KAAK,SAAS;AAAA,IACpD;AAIA,QAAI,gBAAgB;AACpB,eAAW,KAAK,QAAQ,OAAO,GAAG;AAChC,uBAAiB,EAAE;AACnB,QAAE,MAAM;AAAA,QACN,CAAC,GAAG,MACF,gBAAgB,EAAE,YAAY,EAAE,UAAU,KAAK,EAAE,mBAAmB,EAAE;AAAA,MAC1E;AAGA,YAAM,UAAU,gBAAgB,EAAE,UAAU;AAC5C,YAAM,QAAQ,gBAAgB,EAAE,QAAQ;AACxC,QAAE,cAAc,YAAY,QAAQ,UAAU,OAAO,IAAI,KAAK,IAAI,GAAG,QAAQ,OAAO;AAAA,IACtF;AACA,UAAM,iBAAiB,CAAC,GAAG,QAAQ,KAAK,CAAC,EAAE,KAAK;AAEhD,WAAO,EAAE,SAAS,eAAe,eAAe;AAAA,EAClD;AAAA;AAAA,EAIA,MAAc,cACZ,KACA,SAC4B;AAC5B,UAAM,SAAS,IAAI,eAAe,IAAI,CAAC,OAAO,IAAI,QAAQ,IAAI,EAAE,CAAC,EAAE,OAAO,SAAS;AACnF,QAAI,CAAC,QAAS,QAAO;AAErB,UAAM,kBAAkB,OAAO,OAAO,CAAC,MAAM;AAC3C,UAAI,QAAQ,eAAe,UAAa,EAAE,eAAe,QAAQ,WAAY,QAAO;AACpF,UAAI,QAAQ,iBAAiB,QAAQ,cAAc,SAAS,GAAG;AAC7D,YAAI,CAAC,EAAE,gBAAgB,CAAC,QAAQ,cAAc,SAAS,EAAE,YAAY,EAAG,QAAO;AAAA,MACjF;AACA,UAAI,QAAQ,eAAe,QAAQ,YAAY,SAAS,GAAG;AACzD,YAAI,CAAC,EAAE,cAAc,CAAC,QAAQ,YAAY,SAAS,EAAE,UAAU,EAAG,QAAO;AAAA,MAC3E;AACA,UAAI,QAAQ,eAAe,QAAQ,YAAY,SAAS,GAAG;AACzD,YAAI,CAAC,CAAC,GAAG,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,QAAQ,YAAa,SAAS,CAAC,CAAC,EAAG,QAAO;AAAA,MAC3E;AACA,UAAI,QAAQ,cAAc,QAAQ,WAAW,SAAS,GAAG;AACvD,YAAI,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,KAAK,CAAC,OAAO,QAAQ,WAAY,SAAS,EAAE,CAAC,EAAG,QAAO;AAAA,MAC3E;AACA,UAAI,QAAQ,oBAAoB,EAAE,aAAa,QAAQ,iBAAkB,QAAO;AAChF,UAAI,QAAQ,qBAAqB,EAAE,aAAa,QAAQ,kBAAmB,QAAO;AAClF,aAAO;AAAA,IACT,CAAC;AAED,QAAI,CAAC,QAAQ,cAAe,QAAO;AAGnC,UAAM,KAAK,mBAAmB,QAAQ,aAAa;AACnD,UAAM,MAAM,MAAM,KAAK,OAAO;AAC9B,UAAM,MAAyB,CAAC;AAChC,eAAW,KAAK,iBAAiB;AAC/B,UAAI,UAAU;AACd,iBAAW,KAAK,EAAE,OAAO;AACvB,cAAM,QAAQ,IAAI,SAAS,EAAE,kBAAkB,EAAE,mBAAmB,EAAE,gBAAgB;AAItF,YAAI,GAAG,KAAK,MAAM,SAAS,MAAM,CAAC,GAAG;AACnC,oBAAU;AACV;AAAA,QACF;AAAA,MACF;AACA,UAAI,QAAS,KAAI,KAAK,CAAC;AAAA,IACzB;AACA,WAAO;AAAA,EACT;AAAA,EAEQ,UAAU,GAA8C;AAC9D,WAAO;AAAA,MACL,UAAU,EAAE;AAAA,MACZ,cAAc,EAAE;AAAA,MAChB,YAAY,EAAE;AAAA,MACd,YAAY,EAAE;AAAA,MACd,YAAY,EAAE;AAAA,MACd,YAAY,EAAE;AAAA,MACd,UAAU,EAAE;AAAA,MACZ,aAAa,EAAE;AAAA,MACf,iBAAiB,EAAE;AAAA,MACnB,QAAQ,CAAC,GAAG,EAAE,MAAM,EAAE,KAAK;AAAA,MAC3B,OAAO,CAAC,GAAG,EAAE,KAAK,EAAE,KAAK;AAAA,IAC3B;AAAA,EACF;AAAA;AAAA,EAIQ,YACN,KACA,UACA,GACA,YACA,SACkB;AAClB,UAAM,QAAQ,IACX,SAAS,EAAE,kBAAkB,EAAE,mBAAmB,EAAE,gBAAgB,EACpE,SAAS,MAAM;AAClB,QAAI,MAA+B,CAAC;AACpC,QAAI;AACF,YAAM,SAAS,KAAK,MAAM,KAAK;AAC/B,UAAI,UAAU,OAAO,WAAW,SAAU,OAAM;AAAA,IAClD,QAAQ;AAAA,IAER;AACA,UAAM,QAAQ,sBAAsB,GAAG;AACvC,UAAM,YAAqC,CAAC;AAC5C,eAAW,CAAC,GAAG,CAAC,KAAK,OAAO,QAAQ,KAAK,GAAG;AAC1C,UAAI,OAAO,MAAM,UAAU;AACzB,cAAM,QAAQ,kBAAkB,GAAG,UAAU;AAC7C,YAAI,UAAU,EAAG,SAAQ,SAAS;AAClC,kBAAU,CAAC,IAAI;AAAA,MACjB,WAAW,MAAM,QAAQ,CAAC,KAAM,KAAK,OAAO,MAAM,UAAW;AAC3D,cAAM,OAAO,KAAK,UAAU,CAAC;AAC7B,cAAM,QAAQ,kBAAkB,MAAM,UAAU;AAChD,YAAI,UAAU,MAAM;AAClB,kBAAQ,SAAS;AACjB,oBAAU,CAAC,IAAI;AAAA,QACjB,OAAO;AACL,oBAAU,CAAC,IAAI;AAAA,QACjB;AAAA,MACF,OAAO;AACL,kBAAU,CAAC,IAAI;AAAA,MACjB;AAAA,IACF;AACA,WAAO;AAAA,MACL;AAAA,MACA,SAAS,EAAE;AAAA,MACX,gBAAgB,EAAE;AAAA,MAClB,MAAM,EAAE;AAAA,MACR,MAAM,EAAE;AAAA,MACR,YAAY,EAAE;AAAA,MACd,UAAU,EAAE;AAAA,MACZ,aAAa,EAAE;AAAA,MACf,QAAQ,EAAE;AAAA,MACV,gBAAgB,EAAE;AAAA,MAClB,cAAc,EAAE;AAAA,MAChB,YAAY,EAAE;AAAA,MACd,YAAY,EAAE;AAAA,MACd,WAAW,EAAE;AAAA,MACb,YAAY;AAAA,IACd;AAAA,EACF;AAAA,EAEQ,sBACN,GACA,yBACoB;AACpB,UAAM,SAAS,oBAAI,IAAoB;AACvC,QAAI,aAAa;AACjB,eAAW,KAAK,EAAE,OAAO;AACvB,aAAO,IAAI,EAAE,OAAO,OAAO,IAAI,EAAE,IAAI,KAAK,KAAK,CAAC;AAChD,UAAI,EAAE,WAAW,QAAS,eAAc;AAAA,IAC1C;AACA,UAAM,MAAM,CAAC,GAAG,OAAO,QAAQ,CAAC,EAAE,KAAK,CAAC,GAAG,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,CAAC,EAAE,MAAM,GAAG,EAAE;AACzE,WAAO;AAAA,MACL,YAAY,EAAE;AAAA,MACd,gBAAgB;AAAA,MAChB;AAAA,MACA,kBAAkB;AAAA,IACpB;AAAA,EACF;AAAA,EAEA,MAAc,mBACZ,KACA,UACA,GACA,IACA,YACA,WAC0E;AAK1E,UAAM,QAAQ,IACX,SAAS,EAAE,kBAAkB,EAAE,mBAAmB,EAAE,gBAAgB,EACpE,SAAS,MAAM;AAClB,UAAM,UAA6B,CAAC;AACpC,UAAM,WAAW,IAAI,OAAO,GAAG,QAAQ,GAAG,MAAM,SAAS,GAAG,IAAI,GAAG,QAAQ,GAAG,GAAG,KAAK,GAAG;AACzF,QAAI,QAAQ;AACZ,QAAI,UAAU;AACd,QAAI,IAA4B,SAAS,KAAK,KAAK;AACnD,WAAO,MAAM,MAAM;AACjB,eAAS;AACT,UAAI,EAAE,UAAU,SAAS,UAAW,UAAS,aAAa;AAC1D,UAAI,QAAQ,UAAU,WAAW;AAC/B,kBAAU;AACV;AAAA,MACF;AACA,YAAM,SAAS,MAAM,MAAM,KAAK,IAAI,GAAG,EAAE,QAAQ,aAAa,CAAC,GAAG,EAAE,KAAK;AACzE,YAAM,QAAQ,MAAM;AAAA,QAClB,EAAE,QAAQ,EAAE,CAAC,EAAE;AAAA,QACf,EAAE,QAAQ,EAAE,CAAC,EAAE,SAAS,KAAK,MAAM,aAAa,CAAC;AAAA,MACnD;AACA,cAAQ,KAAK;AAAA,QACX;AAAA,QACA,SAAS,EAAE;AAAA,QACX,WAAW,EAAE;AAAA,QACb,WAAW,EAAE;AAAA,QACb,gBAAgB,2BAA2B,OAAO,EAAE,KAAK,KAAK;AAAA,QAC9D,cAAc,kBAAkB,EAAE,CAAC,GAAG,UAAU;AAAA,QAChD,gBAAgB,kBAAkB,QAAQ,UAAU;AAAA,QACpD,eAAe,kBAAkB,OAAO,UAAU;AAAA,QAClD,cAAc,EAAE;AAAA,MAClB,CAAC;AACD,UAAI,SAAS,KAAK,KAAK;AAAA,IACzB;AACA,WAAO,EAAE,SAAS,OAAO,QAAQ;AAAA,EACnC;AACF;AAIO,IAAM,wBAAN,cAAoC,cAAc;AAAA,EACvD,YAAY,MAAc;AACxB,UAAM,yBAAyB,IAAI,EAAE;AAAA,EACvC;AACF;AACO,IAAM,yBAAN,cAAqC,MAAM;AAAA,EACvC;AAAA,EACA;AAAA,EACA;AAAA,EACT,YAAY,MAAc,YAAoB,WAAmB;AAC/D;AAAA,MACE,cAAc,IAAI,OAAO,UAAU,oBAAoB,SAAS;AAAA,IAElE;AACA,SAAK,OAAO;AACZ,SAAK,aAAa;AAClB,SAAK,YAAY;AAAA,EACnB;AACF;AACO,IAAM,qBAAN,cAAiC,cAAc;AAAA,EAC3C;AAAA,EACT,YAAY,UAAkB;AAC5B,UAAM,oBAAoB,QAAQ,EAAE;AACpC,SAAK,WAAW;AAAA,EAClB;AACF;AACO,IAAM,oBAAN,cAAgC,cAAc;AAAA,EAC1C;AAAA,EACA;AAAA,EACT,YAAY,UAAkB,SAAiB;AAC7C,UAAM,QAAQ,OAAO,uBAAuB,QAAQ,EAAE;AACtD,SAAK,WAAW;AAChB,SAAK,UAAU;AAAA,EACjB;AACF;AAOA,SAAS,UAAa,GAA0B;AAC9C,SAAO,MAAM;AACf;AAaA,SAAS,iBAAiB,OAAe,KAAsB;AAC7D,MAAI,MAAM,GAAG,MAAM,IAAK,QAAO;AAC/B,MAAI,cAAc;AAClB,MAAI,IAAI,MAAM;AACd,SAAO,KAAK,KAAK,MAAM,CAAC,MAAM,MAAM;AAClC,mBAAe;AACf,SAAK;AAAA,EACP;AACA,SAAO,cAAc,MAAM;AAC7B;AAIA,SAAS,mBAAmB,OAAe,MAAsB;AAC/D,WAAS,IAAI,MAAM,KAAK,GAAG,KAAK,GAAG;AACjC,QAAI,MAAM,CAAC,MAAM,OAAO,iBAAiB,OAAO,CAAC,EAAG,QAAO;AAAA,EAC7D;AACA,SAAO;AACT;AAUA,SAAS,2BAA2B,OAAe,QAA+B;AAEhF,QAAM,aAAa,mBAAmB,OAAO,KAAK,IAAI,QAAQ,MAAM,SAAS,CAAC,CAAC;AAC/E,MAAI,aAAa,EAAG,QAAO;AAE3B,MAAI,IAAI,aAAa;AACrB,SAAO,KAAK,KAAK,MAAM,CAAC,MAAM,IAAK,MAAK;AACxC,MAAI,IAAI,EAAG,QAAO;AAElB,QAAM,WAAW,mBAAmB,OAAO,IAAI,CAAC;AAChD,MAAI,WAAW,EAAG,QAAO;AACzB,QAAM,UAAU,mBAAmB,OAAO,WAAW,CAAC;AACtD,MAAI,UAAU,EAAG,QAAO;AACxB,SAAO,MAAM,MAAM,UAAU,GAAG,QAAQ;AAC1C;AASA,IAAM,oBAAoB;AAC1B,IAAM,0BAA0B;AAChC,IAAM,sBAAsB;AAgB5B,SAAS,wBAAwB,SAA6B,UAA0B;AACtF,QAAM,OAAO,WAAW,IAAI,KAAK;AACjC,QAAM,OAAO,IAAI,SAAS,IAAI,MAAM,IAAI,YAAY,OAAO;AAC3D,QAAM,OAAO,KACV,QAAQ,kEAAkE,MAAM,EAChF,QAAQ,uBAAuB,MAAM,EACrC,QAAQ,wBAAwB,QAAQ,EACxC,QAAQ,2CAA2C,CAAC,IAAI,MAAO,IAAI,IAAI,CAAC,KAAK,GAAI,EACjF,QAAQ,QAAQ,GAAG,EACnB,KAAK;AACR,SAAO,KAAK,SAAS,sBAAsB,GAAG,KAAK,MAAM,GAAG,mBAAmB,CAAC,WAAM;AACxF;AAEA,SAAS,KAAK,KAA0B,KAA0B;AAChE,MAAI,CAAC,IAAK;AACV,MAAI,IAAI,MAAM,IAAI,IAAI,GAAG,KAAK,KAAK,CAAC;AACtC;AAEA,SAAS,OAAO,KAAyC;AACvD,MAAI,OAAsB;AAC1B,MAAI,QAAQ;AACZ,aAAW,CAAC,GAAG,CAAC,KAAK;AACnB,QAAI,IAAI,OAAO;AACb,aAAO;AACP,cAAQ;AAAA,IACV;AACF,SAAO;AACT;AAEA,SAAS,uBACP,UACA,SACA,MACM;AACN,QAAM,YAAY,wBAAwB,KAAK,gBAAgB,KAAK,IAAI;AACxE,MAAI,MAAM,SAAS,IAAI,SAAS;AAChC,MAAI,CAAC,KAAK;AACR,UAAM;AAAA,MACJ;AAAA,MACA,SAAS,KAAK,kBAAkB,KAAK,QAAQ,IAAI,MAAM,GAAG,GAAG;AAAA,MAC7D,UAAU,oBAAI,IAAI;AAAA,MAClB,SAAS,CAAC;AAAA,MACV,WAAW;AAAA,MACX,WAAW,oBAAI,IAAI;AAAA,MACnB,WAAW,oBAAI,IAAI;AAAA,IACrB;AACA,aAAS,IAAI,WAAW,GAAG;AAAA,EAC7B;AACA,MAAI,SAAS,IAAI,OAAO;AACxB,MAAI,aAAa;AACjB,MAAI,IAAI,QAAQ,SAAS,2BAA2B,CAAC,IAAI,QAAQ,SAAS,KAAK,OAAO,GAAG;AACvF,QAAI,QAAQ,KAAK,KAAK,OAAO;AAAA,EAC/B;AACA,OAAK,IAAI,WAAW,KAAK,IAAI;AAC7B,OAAK,IAAI,WAAW,KAAK,SAAS;AACpC;AAEA,SAAS,sBACP,UACA,iBACgB;AAChB,QAAM,MAAM,CAAC,GAAG,SAAS,OAAO,CAAC,EAAE;AAAA,IACjC,CAAC,SAAuB;AAAA,MACtB,WAAW,IAAI;AAAA,MACf,uBAAuB,IAAI;AAAA,MAC3B,WAAW,OAAO,IAAI,SAAS;AAAA,MAC/B,WAAW,OAAO,IAAI,SAAS;AAAA,MAC/B,aAAa,IAAI,SAAS;AAAA,MAC1B,YAAY,IAAI;AAAA,MAChB,YAAY,kBAAkB,IAAI,IAAI,SAAS,OAAO,kBAAkB;AAAA,MACxE,oBAAoB,CAAC,GAAG,IAAI,QAAQ,EAAE,MAAM,GAAG,uBAAuB;AAAA,MACtE,mBAAmB,IAAI,QAAQ,MAAM,GAAG,uBAAuB;AAAA,IACjE;AAAA,EACF;AACA,MAAI,KAAK,CAAC,GAAG,MAAM,EAAE,cAAc,EAAE,eAAe,EAAE,aAAa,EAAE,UAAU;AAC/E,SAAO,IAAI,MAAM,GAAG,iBAAiB;AACvC;;;AEl6BA,SAAS,GAAG,UAAU;AAKtB,IAAM,YAAY;AAQlB,IAAM,eAAe,EAClB,KAAK,0EAA0E,EAC/E,SAAS;AAML,SAAS,uBAAuB,MAAgD;AACrF,QAAM,EAAE,MAAM,IAAI;AAElB,QAAM,qBAAqB,GAAG,oBAAoB,EAC/C;AAAA,IACC;AAAA,EAGF,EACC,UAAU,SAAS,EACnB,IAAI,WAAW,YAAY,EAC3B,QAAQ,EAAE,KAAK,iBAAiB,CAAC,EACjC,QAAQ,OAAO,EAAE,QAAQ,MAAM,MAAM,YAAY,aAAa,OAAO,CAAC,CAAC,EACvE,MAAM;AAET,QAAM,cAAc,GAAG,aAAa,EACjC;AAAA,IACC;AAAA,EAGF,EACC,UAAU,SAAS,EACnB,IAAI,WAAW,YAAY,EAC3B,IAAI,SAAS,EAAE,OAAO,mBAAmB,CAAC,EAC1C,IAAI,UAAU,EAAE,OAAO,wBAAwB,EAAE,SAAS,CAAC,EAC3D,QAAQ,EAAE,KAAK,iBAAiB,CAAC,EACjC;AAAA,IAAQ,OAAO,EAAE,SAAS,OAAO,OAAO,MACvC,MAAM,YAAY;AAAA,MAChB,SAAS,aAAa,OAAO;AAAA,MAC7B,OAAO,gBAAgB,KAAK;AAAA,MAC5B,QAAQ,aAAa,MAAM;AAAA,IAC7B,CAAC;AAAA,EACH,EACC,MAAM;AAET,QAAM,cAAc,GAAG,aAAa,EACjC;AAAA,IACC;AAAA,EAEF,EACC,UAAU,SAAS,EACnB,IAAI,WAAW,YAAY,EAC3B,QAAQ,EAAE,OAAO,OAAO,CAAC,EACzB,QAAQ,OAAO,EAAE,QAAQ,MAAM,MAAM,YAAY,aAAa,OAAO,CAAC,CAAC,EACvE,MAAM;AAET,QAAM,YAAY,GAAG,WAAW,EAC7B;AAAA,IACC;AAAA,EAIF,EACC,UAAU,SAAS,EACnB,IAAI,YAAY,EAAE,OAAO,2CAA2C,CAAC,EACrE,QAAQ,EAAE,KAAK,iBAAiB,CAAC,EACjC;AAAA,IAAQ,OAAO,EAAE,SAAS,MACzB,MAAM,UAAU,EAAE,UAAU,aAAa,UAAU,UAAU,EAAE,CAAC;AAAA,EAClE,EACC,MAAM;AAET,QAAM,YAAY,GAAG,WAAW,EAC7B;AAAA,IACC;AAAA,EAGF,EACC,UAAU,SAAS,EACnB,IAAI,YAAY,EAAE,OAAO,eAAe,CAAC,EACzC,IAAI,YAAY,EAAE,OAAO,mBAAmB,EAAE,MAAM,CAAC,EACrD,QAAQ,EAAE,KAAK,iBAAiB,CAAC,EACjC;AAAA,IAAQ,OAAO,EAAE,UAAU,SAAS,MACnC,MAAM,UAAU;AAAA,MACd,UAAU,aAAa,UAAU,UAAU;AAAA,MAC3C,UAAU,kBAAkB,UAAU,UAAU;AAAA,IAClD,CAAC;AAAA,EACH,EACC,MAAM;AAET,QAAM,cAAc,GAAG,aAAa,EACjC;AAAA,IACC;AAAA,EAQF,EACC,UAAU,SAAS,EACnB,IAAI,YAAY,EAAE,OAAO,eAAe,CAAC,EACzC,IAAI,iBAAiB,EAAE,OAAO,gCAAgC,CAAC,EAC/D,IAAI,eAAe,EAAE,OAAO,0CAA0C,EAAE,SAAS,CAAC,EAClF,QAAQ,EAAE,KAAK,mBAAmB,CAAC,EACnC;AAAA,IAAQ,OAAO,EAAE,UAAU,eAAe,YAAY,MACrD,MAAM,YAAY;AAAA,MAChB,UAAU,aAAa,UAAU,UAAU;AAAA,MAC3C,eAAe,YAAY,aAAa;AAAA,MACxC,aAAa,iBAAiB,WAAW;AAAA,IAC3C,CAAC;AAAA,EACH,EACC,MAAM;AAET,QAAM,aAAa,GAAG,YAAY,EAC/B;AAAA,IACC;AAAA,EAKF,EACC,UAAU,SAAS,EACnB,IAAI,YAAY,EAAE,OAAO,eAAe,CAAC,EACzC,IAAI,WAAW,EAAE,OAAO,2BAA2B,CAAC,EACpD,IAAI,iBAAiB,EAAE,OAAO,gCAAgC,CAAC,EAC/D,IAAI,eAAe,EAAE,OAAO,iCAAiC,EAAE,SAAS,CAAC,EACzE,QAAQ,EAAE,KAAK,kBAAkB,CAAC,EAClC;AAAA,IAAQ,OAAO,EAAE,UAAU,SAAS,eAAe,YAAY,MAC9D,MAAM,WAAW;AAAA,MACf,UAAU,aAAa,UAAU,UAAU;AAAA,MAC3C,SAAS,aAAa,SAAS,SAAS;AAAA,MACxC,eAAe,YAAY,aAAa;AAAA,MACxC,aAAa,iBAAiB,WAAW;AAAA,IAC3C,CAAC;AAAA,EACH,EACC,MAAM;AAET,SAAO;AAAA,IACL;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AACF;AAMO,SAAS,0BAA0B,MAMxC;AACA,SAAO;AAAA,IACL,WAAW;AAAA,IACX,OAAO;AAAA,IACP,mBAAmB;AAAA,IACnB,aACE;AAAA,IAEF,WAAW,uBAAuB,IAAI;AAAA,EACxC;AACF;AAIA,SAAS,aAAa,OAAiD;AACrE,MAAI,SAAS,KAAM,QAAO;AAC1B,MAAI,OAAO,UAAU,YAAY,MAAM,QAAQ,KAAK,GAAG;AACrD,UAAM,IAAI,UAAU,kCAAkC,OAAO,KAAK,EAAE;AAAA,EACtE;AACA,QAAMC,KAAI;AACV,QAAM,MAA2B,CAAC;AAClC,MAAI,OAAOA,GAAE,eAAe,UAAW,KAAI,aAAaA,GAAE;AAC1D,MAAI,gBAAgB,uBAAuBA,GAAE,eAAe,eAAe;AAC3E,MAAI,cAAc,uBAAuBA,GAAE,aAAa,aAAa;AACrE,MAAI,cAAc,uBAAuBA,GAAE,aAAa,aAAa;AACrE,MAAI,aAAa,uBAAuBA,GAAE,YAAY,YAAY;AAClE,MAAI,OAAOA,GAAE,qBAAqB,SAAU,KAAI,mBAAmBA,GAAE;AACrE,MAAI,OAAOA,GAAE,sBAAsB,SAAU,KAAI,oBAAoBA,GAAE;AACvE,MAAI,OAAOA,GAAE,kBAAkB,UAAU;AACvC,QAAIA,GAAE,cAAc,WAAW,GAAG;AAChC,YAAM,IAAI,UAAU,uCAAuC;AAAA,IAC7D;AACA,QAAI,gBAAgBA,GAAE;AAAA,EACxB;AACA,SAAO;AACT;AAEA,SAAS,uBAAuB,GAAY,OAAqC;AAC/E,MAAI,MAAM,UAAa,MAAM,KAAM,QAAO;AAC1C,MAAI,CAAC,MAAM,QAAQ,CAAC,EAAG,OAAM,IAAI,UAAU,GAAG,KAAK,8BAA8B;AACjF,MAAI,EAAE,KAAK,CAAC,MAAM,OAAO,MAAM,QAAQ,GAAG;AACxC,UAAM,IAAI,UAAU,GAAG,KAAK,0BAA0B;AAAA,EACxD;AACA,SAAO;AACT;AAEA,SAAS,gBAAgB,OAAwB;AAC/C,MAAI,OAAO,UAAU,YAAY,CAAC,OAAO,UAAU,KAAK,KAAK,QAAQ,KAAK,QAAQ,KAAK;AACrF,UAAM,IAAI,WAAW,iCAAiC;AAAA,EACxD;AACA,SAAO;AACT;AACA,SAAS,aAAa,QAAqC;AACzD,MAAI,WAAW,OAAW,QAAO;AACjC,MAAI,OAAO,WAAW,YAAY,CAAC,OAAO,UAAU,MAAM,KAAK,SAAS,GAAG;AACzE,UAAM,IAAI,WAAW,uCAAuC;AAAA,EAC9D;AACA,SAAO;AACT;AACA,SAAS,YAAY,SAA0B;AAC7C,MAAI,OAAO,YAAY,YAAY,QAAQ,WAAW,GAAG;AACvD,UAAM,IAAI,UAAU,0CAA0C;AAAA,EAChE;AAKA,MAAI,OAAO,SAAS,GAAG;AACvB,SAAO;AACT;AACA,SAAS,iBAAiB,GAAgC;AACxD,MAAI,MAAM,OAAW,QAAO;AAC5B,MAAI,OAAO,MAAM,YAAY,CAAC,OAAO,UAAU,CAAC,KAAK,IAAI,KAAK,IAAI,KAAK;AACrE,UAAM,IAAI,WAAW,uCAAuC;AAAA,EAC9D;AACA,SAAO;AACT;AAEA,SAAS,aAAa,GAAY,OAAuB;AACvD,MAAI,OAAO,MAAM,YAAY,EAAE,WAAW,GAAG;AAC3C,UAAM,IAAI,UAAU,GAAG,KAAK,6BAA6B;AAAA,EAC3D;AACA,SAAO;AACT;AAEA,SAAS,kBAAkB,GAAY,OAAyB;AAC9D,MAAI,CAAC,MAAM,QAAQ,CAAC,EAAG,OAAM,IAAI,UAAU,GAAG,KAAK,8BAA8B;AACjF,MAAI,EAAE,KAAK,CAAC,MAAM,OAAO,MAAM,QAAQ,GAAG;AACxC,UAAM,IAAI,UAAU,GAAG,KAAK,0BAA0B;AAAA,EACxD;AACA,SAAO;AACT;;;AC1RA;AAAA,EAIE;AAAA,EACA;AAAA,OACK;AAGA,IAAM,4BAA4B;AAEzC,IAAM,wCAAwC;AAAA;AAAA;AAAA;AAAA,oDAIM,KAAK,UAAU,yBAAyB,CAAC;AAEtF,IAAM,8BAAN,cAA0C,MAAM;AAAA,EAC5C;AAAA,EACA;AAAA,EAET,YAAY,WAAmB,UAAkB,OAAgB;AAC/D;AAAA,MACE,kBAAkB,SAAS,sBAAsB,QAAQ;AAAA,MACzD,EAAE,MAAM;AAAA,IACV;AACA,SAAK,OAAO;AACZ,SAAK,YAAY;AACjB,SAAK,WAAW;AAAA,EAClB;AACF;AAgCA,eAAsB,qBACpB,SACoE;AACpE,qBAAmB,OAAO;AAE1B,QAAM,SAAS;AAAA,IACb,eAAe,EAAE,MAAM,QAAQ,IAAI,aAAa,QAAQ,YAAY;AAAA,IACpE,eAAe,CAAC;AAAA,IAChB,SAAS,IAAI,YAAY;AAAA,MACvB,aAAa,CAAC;AAAA,MACd,oBAAoB;AAAA,MACpB,gBAAgB,CAAC;AAAA,MACjB,kBAAkB;AAAA,MAClB,kBAAkB;AAAA,MAClB,6BAA6B;AAAA,IAC/B,CAAC;AAAA,IACD,kBAAkB,QAAQ;AAAA,IAC1B,UAAU,QAAQ;AAAA,IAClB,iBAAiB,QAAQ;AAAA,IACzB,+BAA+B,QAAQ;AAAA,IACvC,aAAa;AAAA,IACb,eAAe,EAAE,QAAQ,QAAiB,QAAQ,WAAoB;AAAA,IACtE,WAAW,QAAQ;AAAA,IACnB,iBAAiB;AAAA,MACf,aAAa,GAAG,QAAQ,OAAO,KAAK,CAAC;AAAA;AAAA,EAAO,qCAAqC;AAAA,MACjF,GAAI,QAAQ,QAAQ,EAAE,OAAO,QAAQ,MAAM,IAAI,CAAC;AAAA,MAChD,cAAc;AAAA,MACd,qBAAqB;AAAA,IACvB;AAAA,IACA,GAAI,QAAQ,SAAS,EAAE,mBAAmB,QAAQ,OAAO,IAAI,CAAC;AAAA,IAC9D,cAAc,CAAC,qBAAqB;AAAA,EACtC;AACA,QAAM,UACJ,QAAQ,gBAAgB,WACpB,MAAM,uDAAuD,MAAM,IACnE,MAAM,qDAAqD,MAAM;AAEvE,QAAM,QAAQ,MAAM,QAAQ,SAAS;AAAA,IACnC,QAAQ;AAAA,IACR,EAAE,UAAU,QAAQ,SAAS;AAAA,IAC7B,QAAQ,SAAS,EAAE,aAAa,QAAQ,OAAO,IAAI;AAAA,EACrD;AAEA,MAAI;AACJ,MAAI;AACF,gBACE,QAAQ,gBAAgB,WACpB,4BAA4B,MAAM,gBAAgB,QAAQ,IAC1D,4BAA4B,MAAM,gBAAgB,QAAQ;AAAA,EAClE,SAAS,OAAO;AACd,QAAI,MAAM,aAAa,QAAQ,UAAU;AACvC,YAAM,IAAI,4BAA4B,QAAQ,IAAI,QAAQ,UAAU,KAAK;AAAA,IAC3E;AACA,UAAM,IAAI,MAAM,kBAAkB,QAAQ,EAAE,+CAA+C;AAAA,MACzF,OAAO;AAAA,IACT,CAAC;AAAA,EACH;AAEA,SAAO;AAAA,IACL,GAAG;AAAA,IACH,OAAO,QAAQ,SAAS,SAAS;AAAA,IACjC,SAAS,QAAQ,SAAS,WAAW;AAAA,IACrC,WAAW,MAAM;AAAA,EACnB;AACF;AAeO,SAAS,4BACd,OACA,aAC0D;AAC1D,MAAI,CAAC,SAAS,OAAO,UAAU,UAAU;AACvC,UAAM,IAAI,MAAM,wDAAwD;AAAA,EAC1E;AACA,QAAM,aAAa;AACnB,MACE,WAAW,SAAS,WACpB,CAAC,MAAM,QAAQ,WAAW,IAAI,KAC9B,WAAW,KAAK,WAAW,KAC3B,WAAW,KAAK,CAAC,MAAM,2BACvB;AACA,UAAM,IAAI,MAAM,wDAAwD;AAAA,EAC1E;AAEA,QAAM,UAAU,WAAW,KAAK,CAAC;AACjC,MAAI,CAAC,WAAW,OAAO,YAAY,UAAU;AAC3C,UAAM,IAAI,MAAM,6DAA6D;AAAA,EAC/E;AACA,QAAM,EAAE,QAAQ,SAAS,IAAI;AAC7B,MAAI,OAAO,WAAW,YAAY,CAAC,MAAM,QAAQ,QAAQ,GAAG;AAC1D,UAAM,IAAI,MAAM,6DAA6D;AAAA,EAC/E;AACA,MAAI,gBAAgB,UAAU;AAC5B,QAAI,SAAS,KAAK,CAAC,YAAY,OAAO,YAAY,QAAQ,GAAG;AAC3D,YAAM,IAAI,MAAM,yDAAyD;AAAA,IAC3E;AACA,WAAO,EAAE,QAAQ,SAA+B;AAAA,EAClD;AACA,MACE,SAAS,KAAK,CAAC,YAAY,CAAC,WAAW,OAAO,YAAY,YAAY,MAAM,QAAQ,OAAO,CAAC,GAC5F;AACA,UAAM,IAAI,MAAM,yDAAyD;AAAA,EAC3E;AACA,SAAO,EAAE,QAAQ,SAAgD;AACnE;AAEA,SAAS,mBAAmB,SAAyC;AACnE,MAAI,CAAC,OAAO,cAAc,QAAQ,aAAa,KAAK,QAAQ,gBAAgB,GAAG;AAC7E,UAAM,IAAI,UAAU,8CAA8C;AAAA,EACpE;AACA,MAAI,CAAC,OAAO,cAAc,QAAQ,qBAAqB,KAAK,QAAQ,wBAAwB,GAAG;AAC7F,UAAM,IAAI,UAAU,kDAAkD;AAAA,EACxE;AACF;","names":["f","f"]}