@tangle-network/agent-eval 0.52.0 → 0.53.0

package/CHANGELOG.md

@@ -4,6 +4,29 @@ All notable changes to `@tangle-network/agent-eval` and its sibling `agent-eval-
 
 ---
 
+## [0.53.0] — 2026-05-27 — prior-period comparison ("did my last change help?")
+
+### Added
+
+- **`analyzeRuns({ runs, baselineRuns?, baselineLabel? })`** — when `baselineRuns` is provided, `InsightReport` gains a `priorPeriodComparison` block. Two-sample Welch comparison (unpaired — the two windows do NOT need to share scenarios) on: composite score, cost, duration, token usage, and every judge dimension present in both windows.
+- **`PriorPeriodComparison` + `MetricDelta` types** — per-metric `current`, `baseline`, `delta`, Welch 95% CI, p-value, Cohen's d, `baselineN`/`currentN`, and `significant` boolean (p < 0.05 AND |d| ≥ 0.2 — conjunction prevents large-effect-but-noisy and significant-but-tiny from triggering).
+- **`regressedMetrics` + `improvedMetrics` lists** — direction-aware (cost/duration are lower-is-better; composite/dimensions are higher-is-better). Drives the recommendations engine.
+- **New recommendations** — `critical/investigate` fires per regressed metric with the full statistical detail in the rationale (`Welch CI95 = [..], p=.., Cohen's d=..`). `low/ship` fires per improved metric so consumers see what to celebrate without noise.
+
+### Why this matters
+
+"Did my last change help?" is the conversion question for every observability prospect. LangSmith / Braintrust / Phoenix ship scorecards without paired-CI deltas. Hermes has no comparison at all. Our `priorPeriodComparison` answers the question with a falsifiable, statistically-rigorous delta. The block lands in the existing `InsightReport` so every consumer of `analyzeRuns` picks it up automatically.
+
+### Architectural context
+
+Part of the self-improvement-protocol design (`docs/design/self-improvement-protocol.md`). This is 0.53.0 of the roadmap that ends at 1.0.0 (profile-versioning + composite driver) and 1.1.0 (empirical-proof publication).
+
+### Notes
+
+Pure additive surface. `priorPeriodComparison?` is optional; existing consumers untouched. 10 new tests under `tests/prior-period-comparison.test.ts` cover: no-comparison-when-omitted, significant improvement, significant regression, direction-awareness for cost/duration, noise rejection, per-dimension comparison, empty windows, CI bracket-the-truth, both recommendation types. Full suite 1454/1454 green.
+
+---
+
 ## [0.52.0] — 2026-05-27 — honest drivers + profile-versioning architecture
 
 ### Honest correction

package/dist/adapters/otel.d.ts

@@ -1,4 +1,4 @@
-import { T as TraceSpanEvent, H as HostedClient } from '../index-DQHtWQ57.js';
+import { T as TraceSpanEvent, H as HostedClient } from '../index-C7RhhEME.js';
 import '../types-Dbj5gu8n.js';
 import '../summary-report-B7gNRX-r.js';
 import '../run-record-BGY6bHRh.js';

package/dist/contract/index.d.ts

@@ -3,8 +3,8 @@ export { f as CampaignAggregates, g as CampaignArtifactWriter, h as CampaignCell
 import { C as CampaignStorage, e as RunImprovementLoopResult } from '../run-improvement-loop-Cc7oZlRP.js';
 export { D as DefaultProductionGateOptions, E as EvolutionaryDriverOptions, a as GepaDriverOptions, H as HeldOutGateOptions, R as RunCampaignOptions, c as RunEvalOptions, d as RunImprovementLoopOptions, h as composeGate, j as defaultProductionGate, k as evolutionaryDriver, m as fsCampaignStorage, n as gepaDriver, o as heldOutGate, p as inMemoryCampaignStorage, r as runCampaign, s as runEval, t as runImprovementLoop } from '../run-improvement-loop-Cc7oZlRP.js';
 export { D as DeploymentOutcome, F as FileSystemOutcomeStore, b as FileSystemOutcomeStoreOptions, I as InMemoryOutcomeStore, O as OutcomeStore } from '../outcome-store-D6KWmYvj.js';
-import { a as HostedTenant, I as InsightReport, T as TraceSpanEvent } from '../index-DQHtWQ57.js';
-export { F as FailureClusterInsight, b as InterRaterInsight, J as JudgeInsight, L as LiftInsight, O as OutcomeCorrelationInsight, R as Recommendation, c as ReleaseSummary, S as ScalarDistribution } from '../index-DQHtWQ57.js';
+import { a as HostedTenant, I as InsightReport, T as TraceSpanEvent } from '../index-C7RhhEME.js';
+export { F as FailureClusterInsight, b as InterRaterInsight, J as JudgeInsight, L as LiftInsight, O as OutcomeCorrelationInsight, R as Recommendation, c as ReleaseSummary, S as ScalarDistribution } from '../index-C7RhhEME.js';
 import { A as AnalystRegistry } from '../registry-8KAs18kY.js';
 import { a as DatasetScenario } from '../dataset-BlwAtYYf.js';
 import { R as RunRecord, a as RunSplitTag } from '../run-record-BGY6bHRh.js';
@@ -297,6 +297,18 @@ interface AnalyzeRunsOptions {
      *  about. Used by the recommendations engine to call ship vs hold.
      *  Default 0.02. */
     decisionThreshold?: number;
+    /** Optional prior-period runs. When set, the report includes
+     *  `priorPeriodComparison` with per-metric Welch-CI deltas and
+     *  recommendations fire on statistically significant regressions.
+     *  The two windows do NOT have to share scenarios — the comparison
+     *  is two-sample unpaired (the substrate's `lift` field uses paired
+     *  bootstrap on shared (experimentId, seed) tuples; this is the
+     *  shape for "this week vs last week" rather than "candidate vs
+     *  baseline within a campaign"). */
+    baselineRuns?: RunRecord[];
+    /** Human-readable label for the baseline window, e.g. "vs prior 7
+     *  days", "vs v3.1 release". Surfaces in recommendations + UI. */
+    baselineLabel?: string;
 }
 declare function analyzeRuns(opts: AnalyzeRunsOptions): Promise<InsightReport>;
 

package/dist/contract/index.js

@@ -78,6 +78,7 @@ async function analyzeRuns(opts) {
   const contamination = opts.canaryScenarios ? computeContamination(runs, opts.canaryScenarios) : void 0;
   const outcomeCorrelation = opts.outcomeSignal ? computeOutcomeCorrelation(runs, opts.outcomeSignal, split) : void 0;
   const release = buildReleaseScorecard(composite, lift, contamination);
+  const priorPeriodComparison = opts.baselineRuns ? computePriorPeriodComparison(runs, opts.baselineRuns, split, opts.baselineLabel) : void 0;
   const recommendations = buildRecommendations({
     composite,
     judges,
@@ -86,6 +87,7 @@ async function analyzeRuns(opts) {
     failureClusters,
     contamination,
     outcomeCorrelation,
+    priorPeriodComparison,
     threshold
   });
   return {
@@ -100,9 +102,127 @@ async function analyzeRuns(opts) {
     contamination,
     outcomeCorrelation,
     release,
+    ...priorPeriodComparison ? { priorPeriodComparison } : {},
     recommendations
   };
 }
+function computePriorPeriodComparison(current, baseline, split, windowLabel) {
+  if (current.length === 0 || baseline.length === 0) return void 0;
+  const metrics = {};
+  const directions = {};
+  const compositeCurrent = current.map((r) => compositeOf(r, split)).filter(Number.isFinite);
+  const compositeBaseline = baseline.map((r) => compositeOf(r, split)).filter(Number.isFinite);
+  if (compositeCurrent.length > 0 && compositeBaseline.length > 0) {
+    metrics.composite = welchCompare(compositeBaseline, compositeCurrent);
+    directions.composite = "higher-is-better";
+  }
+  const costCurrent = current.map((r) => r.costUsd).filter(Number.isFinite);
+  const costBaseline = baseline.map((r) => r.costUsd).filter(Number.isFinite);
+  if (costCurrent.length > 0 && costBaseline.length > 0) {
+    metrics.cost = welchCompare(costBaseline, costCurrent);
+    directions.cost = "lower-is-better";
+  }
+  const durCurrent = current.map((r) => r.wallMs).filter(Number.isFinite);
+  const durBaseline = baseline.map((r) => r.wallMs).filter(Number.isFinite);
+  if (durCurrent.length > 0 && durBaseline.length > 0) {
+    metrics.duration = welchCompare(durBaseline, durCurrent);
+    directions.duration = "lower-is-better";
+  }
+  const tokCurrent = current.map((r) => (r.tokenUsage.input ?? 0) + (r.tokenUsage.output ?? 0)).filter(Number.isFinite);
+  const tokBaseline = baseline.map((r) => (r.tokenUsage.input ?? 0) + (r.tokenUsage.output ?? 0)).filter(Number.isFinite);
+  if (tokCurrent.length > 0 && tokBaseline.length > 0) {
+    metrics.tokenUsage = welchCompare(tokBaseline, tokCurrent);
+    directions.tokenUsage = "lower-is-better";
+  }
+  const dimsCurrent = collectPerDimension(current);
+  const dimsBaseline = collectPerDimension(baseline);
+  for (const dim of Object.keys(dimsCurrent)) {
+    const b = dimsBaseline[dim];
+    const c = dimsCurrent[dim];
+    if (!b || b.length === 0 || !c || c.length === 0) continue;
+    metrics[`dim.${dim}`] = welchCompare(b, c);
+    directions[`dim.${dim}`] = "higher-is-better";
+  }
+  const regressedMetrics = [];
+  const improvedMetrics = [];
+  for (const [name, delta] of Object.entries(metrics)) {
+    if (!delta.significant) continue;
+    const dir = directions[name] ?? "higher-is-better";
+    const better = dir === "higher-is-better" ? delta.delta > 0 : delta.delta < 0;
+    if (better) improvedMetrics.push(name);
+    else regressedMetrics.push(name);
+  }
+  return {
+    baselineN: baseline.length,
+    currentN: current.length,
+    ...windowLabel ? { windowLabel } : {},
+    metrics,
+    regressedMetrics,
+    improvedMetrics
+  };
+}
+function collectPerDimension(runs) {
+  const out = {};
+  for (const r of runs) {
+    const perDim = r.outcome.judgeScores?.perDimMean;
+    if (!perDim) continue;
+    for (const [dim, value] of Object.entries(perDim)) {
+      if (!Number.isFinite(value)) continue;
+      if (!out[dim]) out[dim] = [];
+      out[dim].push(value);
+    }
+  }
+  return out;
+}
+function welchCompare(baseline, current) {
+  const baselineMean = mean(baseline);
+  const currentMean = mean(current);
+  const baselineVar = sampleVariance(baseline, baselineMean);
+  const currentVar = sampleVariance(current, currentMean);
+  const baselineN = baseline.length;
+  const currentN = current.length;
+  const delta = currentMean - baselineMean;
+  const se = Math.sqrt(baselineVar / baselineN + currentVar / currentN);
+  const halfWidth = 1.96 * (se > 0 ? se : 0);
+  const ci95 = [delta - halfWidth, delta + halfWidth];
+  const t = se > 0 ? delta / se : 0;
+  const pValue = se > 0 ? 2 * (1 - standardNormalCdf(Math.abs(t))) : 1;
+  const pooledStddev = Math.sqrt(
+    ((baselineN - 1) * baselineVar + (currentN - 1) * currentVar) / Math.max(1, baselineN + currentN - 2)
+  );
+  const cohensD2 = pooledStddev > 0 ? delta / pooledStddev : 0;
+  const significant = pValue < 0.05 && Math.abs(cohensD2) >= 0.2;
+  return {
+    current: currentMean,
+    baseline: baselineMean,
+    delta,
+    ci95,
+    pValue,
+    cohensD: cohensD2,
+    baselineN,
+    currentN,
+    significant
+  };
+}
+function sampleVariance(xs, xsMean) {
+  if (xs.length < 2) return 0;
+  let s = 0;
+  for (const x of xs) s += (x - xsMean) ** 2;
+  return s / (xs.length - 1);
+}
+function standardNormalCdf(z) {
+  const a1 = 0.254829592;
+  const a2 = -0.284496736;
+  const a3 = 1.421413741;
+  const a4 = -1.453152027;
+  const a5 = 1.061405429;
+  const p = 0.3275911;
+  const sign = z < 0 ? -1 : 1;
+  const x = Math.abs(z) / Math.SQRT2;
+  const t = 1 / (1 + p * x);
+  const y = 1 - ((((a5 * t + a4) * t + a3) * t + a2) * t + a1) * t * Math.exp(-x * x);
+  return 0.5 * (1 + sign * y);
+}
 function resolveSplit(runs, pref) {
   if (pref !== "auto") return pref;
   const hasHoldout = runs.some((r) => Number.isFinite(r.outcome.holdoutScore));
@@ -492,6 +612,32 @@ function buildReleaseScorecard(composite, lift, contamination) {
 }
 function buildRecommendations(ctx) {
   const out = [];
+  if (ctx.priorPeriodComparison) {
+    const ppc = ctx.priorPeriodComparison;
+    const label = ppc.windowLabel ?? "baseline period";
+    for (const name of ppc.regressedMetrics) {
+      const d = ppc.metrics[name];
+      if (!d) continue;
+      out.push({
+        priority: "critical",
+        kind: "investigate",
+        title: `${name} regressed from ${d.baseline.toFixed(3)} \u2192 ${d.current.toFixed(3)} vs ${label}`,
+        detail: `Welch CI95 = [${d.ci95[0].toFixed(3)}, ${d.ci95[1].toFixed(3)}], p=${d.pValue.toFixed(4)}, Cohen's d=${d.cohensD.toFixed(2)} (n_current=${d.currentN}, n_baseline=${d.baselineN}). The regression is statistically significant at p<0.05 with at-least-small effect size.`,
+        evidencePath: `priorPeriodComparison.metrics.${name}`
+      });
+    }
+    for (const name of ppc.improvedMetrics) {
+      const d = ppc.metrics[name];
+      if (!d) continue;
+      out.push({
+        priority: "low",
+        kind: "ship",
+        title: `${name} improved from ${d.baseline.toFixed(3)} \u2192 ${d.current.toFixed(3)} vs ${label}`,
+        detail: `Welch CI95 = [${d.ci95[0].toFixed(3)}, ${d.ci95[1].toFixed(3)}], p=${d.pValue.toFixed(4)}, Cohen's d=${d.cohensD.toFixed(2)} (n_current=${d.currentN}, n_baseline=${d.baselineN}). Statistically significant improvement worth flagging.`,
+        evidencePath: `priorPeriodComparison.metrics.${name}`
+      });
+    }
+  }
   if (ctx.composite.n > 0) {
     if (ctx.composite.mean < 0.3) {
       const tail = ctx.composite.tailRuns ?? [];

package/dist/contract/index.js.map

@@ -1 +1 @@
-{"version":3,"sources":["../../src/contract/analyze-runs.ts","../../src/contract/self-improve.ts","../../src/contract/intake/feedback-table.ts","../../src/contract/intake/otel-spans.ts"],"sourcesContent":["/**\n * # `analyzeRuns()` — turn a set of agent runs into an actionable decision packet.\n *\n * Wires the substrate's statistical, calibration, clustering, Pareto, and\n * release-confidence primitives into one `InsightReport`. Two top-level\n * entry points use this function:\n *\n *   - `selfImprove()` calls it on the campaign output to attach a packet\n *     to every run.\n *   - Consumers with observed `RunRecord[]` (production traces, gold\n *     corpora, approve/reject tables) call it directly via `analyzeRuns()`\n *     for analysis without a closed loop.\n *\n * Every section is opt-in based on what the input data supports — the\n * function never invents signal. If runs carry no judge scores, `judges`\n * is empty. If there's no baseline/candidate split, `lift` is undefined.\n * If no `analyst` is wired, `failureClusters` is undefined.\n *\n * The `recommendations` array is the human-readable layer; everything\n * else is the evidence backing each recommendation.\n */\n\nimport type { AnalystRegistry } from '../analyst/registry'\nimport type { AnalystFinding } from '../analyst/types'\nimport { checkCanaries } from '../contamination-guard'\nimport type { DatasetScenario } from '../dataset'\nimport type { RunRecord } from '../run-record'\nimport { cohensD, pairedBootstrap, pairedMde, pairedTTest, requiredSampleSize } from '../statistics'\nimport { type ParetoFigureSpec, paretoChart } from '../summary-report'\n\nimport type {\n  FailureClusterInsight,\n  InsightReport,\n  InterRaterInsight,\n  JudgeInsight,\n  LiftInsight,\n  OutcomeCorrelationInsight,\n  Recommendation,\n  ScalarDistribution,\n} from './insight-report'\n\n// ── Public API ───────────────────────────────────────────────────────\n\nexport interface AnalyzeRunsOptions {\n  /** The runs to analyze. */\n  runs: RunRecord[]\n  /** Which split to score against when reading composite from RunOutcome.\n   *  Default: holdout when ANY run has a `holdoutScore`, else search. */\n  split?: 'search' | 'holdout' | 'auto'\n  /** Pairwise analysis configuration. When both `baselineCandidateId` and\n   *  `candidateCandidateId` are present, lift is computed on paired\n   *  (experimentId, seed) tuples shared between the two sides. */\n  baselineCandidateId?: string\n  candidateCandidateId?: string\n  /** Canary scenarios — checked against every run's raw output for\n   *  holdout contamination. */\n  canaryScenarios?: DatasetScenario[]\n  /** Analyst registry for failure clustering. When omitted, the\n   *  `failureClusters` section is left undefined. */\n  analyst?: AnalystRegistry\n  /** Downstream outcome metric per run (e.g. engagement rate, approval\n   *  rate, downstream pass rate). When present, the report includes\n   *  `outcomeCorrelation` + a simple linear reward model fit. */\n  outcomeSignal?: {\n    metric: string\n    valueByRunId: Record<string, number>\n  }\n  /** Multi-rater feedback for inter-rater agreement. Each entry is one\n   *  rater's score for one run. Two or more raters → kappa + disagreement\n   *  triage list. */\n  raterScores?: Array<{ runId: string; rater: string; score: number }>\n  /** Number of histogram bins for distributional summaries. Default 12. */\n  histogramBins?: number\n  /** Decision threshold — the smallest composite lift the caller cares\n   *  about. Used by the recommendations engine to call ship vs hold.\n   *  Default 0.02. */\n  decisionThreshold?: number\n}\n\nexport async function analyzeRuns(opts: AnalyzeRunsOptions): Promise<InsightReport> {\n  const runs = opts.runs\n  const bins = opts.histogramBins ?? 12\n  const threshold = opts.decisionThreshold ?? 0.02\n  const split = resolveSplit(runs, opts.split ?? 'auto')\n\n  const compositeWithIds = runs\n    .map((r) => ({ runId: r.runId, score: compositeOf(r, split) }))\n    .filter((p) => Number.isFinite(p.score))\n  const composite = distributionOf(\n    compositeWithIds.map((p) => p.score),\n    bins,\n    compositeWithIds,\n  )\n\n  const perDimension = computePerDimension(runs, bins)\n\n  const costs = runs.map((r) => r.costUsd).filter(Number.isFinite)\n  const costDist = distributionOf(costs, bins)\n  const pareto = paretoChart(runs, { split })\n  const degraded: { cost?: string; pareto?: string } = {}\n  if (costs.length === 0 || costs.every((c) => c === 0)) {\n    degraded.cost = 'no costUsd values recorded — cost axis carries no signal'\n  }\n  if (pareto.points.length < 2) {\n    degraded.pareto =\n      pareto.points.length === 0\n        ? 'no candidates — Pareto unavailable'\n        : 'single candidate — Pareto is a single point, not a frontier'\n  }\n  const costQuality = {\n    cost: costDist,\n    pareto,\n    ...(degraded.cost || degraded.pareto ? { degraded } : {}),\n  }\n\n  const judges = computeJudgeInsights(runs)\n\n  const interRater = opts.raterScores ? computeInterRater(opts.raterScores) : undefined\n\n  const lift = computeLift(runs, opts.baselineCandidateId, opts.candidateCandidateId, split)\n\n  const failureClusters = opts.analyst\n    ? await computeFailureClusters(runs, opts.analyst, split)\n    : undefined\n\n  const contamination = opts.canaryScenarios\n    ? computeContamination(runs, opts.canaryScenarios)\n    : undefined\n\n  const outcomeCorrelation = opts.outcomeSignal\n    ? computeOutcomeCorrelation(runs, opts.outcomeSignal, split)\n    : undefined\n\n  const release = buildReleaseScorecard(composite, lift, contamination)\n\n  const recommendations = buildRecommendations({\n    composite,\n    judges,\n    interRater,\n    lift,\n    failureClusters,\n    contamination,\n    outcomeCorrelation,\n    threshold,\n  })\n\n  return {\n    n: runs.length,\n    composite,\n    perDimension,\n    costQuality,\n    judges,\n    interRater,\n    lift,\n    failureClusters,\n    contamination,\n    outcomeCorrelation,\n    release,\n    recommendations,\n  }\n}\n\n// ── Composite + split selection ─────────────────────────────────────\n\nfunction resolveSplit(\n  runs: RunRecord[],\n  pref: 'search' | 'holdout' | 'auto',\n): 'search' | 'holdout' {\n  if (pref !== 'auto') return pref\n  const hasHoldout = runs.some((r) => Number.isFinite(r.outcome.holdoutScore))\n  return hasHoldout ? 'holdout' : 'search'\n}\n\nfunction compositeOf(run: RunRecord, split: 'search' | 'holdout'): number {\n  const primary = split === 'holdout' ? run.outcome.holdoutScore : run.outcome.searchScore\n  if (Number.isFinite(primary)) return primary as number\n  // Fall through to the other split if the preferred one is missing —\n  // analyzeRuns shouldn't refuse to summarise a run just because the\n  // caller asked for the split that wasn't recorded.\n  const alt = split === 'holdout' ? run.outcome.searchScore : run.outcome.holdoutScore\n  return Number.isFinite(alt) ? (alt as number) : Number.NaN\n}\n\n// ── Distribution helpers ────────────────────────────────────────────\n\nfunction distributionOf(\n  values: number[],\n  bins: number,\n  withIds?: Array<{ runId: string; score: number }>,\n): ScalarDistribution {\n  if (values.length === 0) {\n    return {\n      n: 0,\n      mean: 0,\n      p50: 0,\n      p95: 0,\n      stddev: 0,\n      min: 0,\n      max: 0,\n      histogram: [],\n    }\n  }\n  const sorted = [...values].sort((a, b) => a - b)\n  const n = sorted.length\n  const mean = sorted.reduce((s, v) => s + v, 0) / n\n  const variance = sorted.reduce((s, v) => s + (v - mean) ** 2, 0) / n\n  const stddev = Math.sqrt(variance)\n  const tailRuns = withIds\n    ? [...withIds].sort((a, b) => a.score - b.score).slice(0, Math.min(5, withIds.length))\n    : undefined\n  return {\n    n,\n    mean,\n    p50: percentile(sorted, 0.5),\n    p95: percentile(sorted, 0.95),\n    stddev,\n    min: sorted[0]!,\n    max: sorted[n - 1]!,\n    histogram: histogram(sorted, bins),\n    ...(tailRuns ? { tailRuns } : {}),\n  }\n}\n\nfunction percentile(sorted: number[], q: number): number {\n  if (sorted.length === 0) return 0\n  if (sorted.length === 1) return sorted[0]!\n  const idx = (sorted.length - 1) * q\n  const lo = Math.floor(idx)\n  const hi = Math.ceil(idx)\n  if (lo === hi) return sorted[lo]!\n  const w = idx - lo\n  return sorted[lo]! * (1 - w) + sorted[hi]! * w\n}\n\n/** Even-width histogram over the value range. Returns inclusive-lo /\n *  exclusive-hi bins (closed on right for the last bin) compatible with\n *  the substrate's `GainDistributionBin` shape. */\nfunction histogram(sorted: number[], bins: number): ScalarDistribution['histogram'] {\n  if (sorted.length === 0 || bins < 1) return []\n  const min = sorted[0]!\n  const max = sorted[sorted.length - 1]!\n  if (min === max) return [{ lo: min, hi: max, count: sorted.length }]\n  const width = (max - min) / bins\n  const out: ScalarDistribution['histogram'] = []\n  for (let i = 0; i < bins; i++) {\n    const lo = min + i * width\n    const hi = i === bins - 1 ? max : lo + width\n    out.push({ lo, hi, count: 0 })\n  }\n  for (const v of sorted) {\n    const idx = Math.min(bins - 1, Math.floor((v - min) / width))\n    out[idx]!.count++\n  }\n  return out\n}\n\nfunction computePerDimension(runs: RunRecord[], bins: number): Record<string, ScalarDistribution> {\n  // JudgeScoresRecord pre-aggregates `perDimMean` (mean across judges per\n  // dimension). We collect those means across runs to produce a per-dim\n  // distribution at the corpus level. Consumers who want per-judge\n  // dimension values reach into `perJudge[judgeId][dim]` themselves.\n  const byDim = new Map<string, number[]>()\n  for (const run of runs) {\n    const scores = run.outcome.judgeScores\n    if (!scores) continue\n    for (const [dim, value] of Object.entries(scores.perDimMean ?? {})) {\n      if (!Number.isFinite(value)) continue\n      const arr = byDim.get(dim) ?? []\n      arr.push(value)\n      byDim.set(dim, arr)\n    }\n  }\n  const out: Record<string, ScalarDistribution> = {}\n  for (const [dim, values] of byDim) out[dim] = distributionOf(values, bins)\n  return out\n}\n\n// ── Judge insights ──────────────────────────────────────────────────\n\nfunction computeJudgeInsights(runs: RunRecord[]): Record<string, JudgeInsight> {\n  // Each judge's per-run mean is the average of its per-dimension scores\n  // for that run. We aggregate those means across all runs each judge\n  // scored — giving consumers a \"this judge's typical verdict\" reading.\n  const out: Record<string, JudgeInsight> = {}\n  const byJudge = new Map<string, number[]>()\n  for (const run of runs) {\n    const scores = run.outcome.judgeScores\n    if (!scores?.perJudge) continue\n    for (const [judgeId, dims] of Object.entries(scores.perJudge)) {\n      const dimValues = Object.values(dims).filter(Number.isFinite) as number[]\n      if (dimValues.length === 0) continue\n      const judgeMean = dimValues.reduce((s, v) => s + v, 0) / dimValues.length\n      const arr = byJudge.get(judgeId) ?? []\n      arr.push(judgeMean)\n      byJudge.set(judgeId, arr)\n    }\n  }\n  for (const [judgeId, values] of byJudge) {\n    out[judgeId] = {\n      n: values.length,\n      meanScore: values.reduce((s, v) => s + v, 0) / values.length,\n    }\n  }\n  return out\n}\n\n// ── Inter-rater agreement ───────────────────────────────────────────\n\nfunction computeInterRater(\n  ratings: Array<{ runId: string; rater: string; score: number }>,\n): InterRaterInsight | undefined {\n  const byRun = new Map<string, Array<{ rater: string; score: number }>>()\n  for (const r of ratings) {\n    if (!Number.isFinite(r.score)) continue\n    const list = byRun.get(r.runId) ?? []\n    list.push({ rater: r.rater, score: r.score })\n    byRun.set(r.runId, list)\n  }\n  const raters = new Set(ratings.map((r) => r.rater))\n  const jointlyRated: string[] = []\n  for (const [runId, ratersForRun] of byRun) {\n    const seen = new Set(ratersForRun.map((r) => r.rater))\n    let all = true\n    for (const r of raters) if (!seen.has(r)) all = false\n    if (all) jointlyRated.push(runId)\n  }\n  if (raters.size < 2 || jointlyRated.length === 0) return undefined\n\n  const raterList = [...raters].sort()\n  const perPair: Record<string, number> = {}\n  for (let i = 0; i < raterList.length; i++) {\n    for (let j = i + 1; j < raterList.length; j++) {\n      const a = raterList[i]!\n      const b = raterList[j]!\n      const aScores: number[] = []\n      const bScores: number[] = []\n      for (const runId of jointlyRated) {\n        const ratersForRun = byRun.get(runId)!\n        const sa = ratersForRun.find((r) => r.rater === a)?.score\n        const sb = ratersForRun.find((r) => r.rater === b)?.score\n        if (sa !== undefined && sb !== undefined) {\n          aScores.push(sa)\n          bScores.push(sb)\n        }\n      }\n      perPair[`${a}::${b}`] = pearson(aScores, bScores)\n    }\n  }\n  const pairKappas = Object.values(perPair)\n  const kappa =\n    pairKappas.length === 0 ? 0 : pairKappas.reduce((s, v) => s + v, 0) / pairKappas.length\n\n  const disagreementCases = jointlyRated\n    .map((runId) => {\n      const ratersForRun = byRun.get(runId)!\n      const scores = ratersForRun.map((r) => r.score)\n      const range = Math.max(...scores) - Math.min(...scores)\n      return { runId, ratings: ratersForRun, range }\n    })\n    .sort((a, b) => b.range - a.range)\n    .slice(0, 20)\n\n  return {\n    raters: raters.size,\n    jointlyRated: jointlyRated.length,\n    kappa,\n    perPair,\n    disagreementCases,\n  }\n}\n\nfunction pearson(a: number[], b: number[]): number {\n  if (a.length !== b.length || a.length === 0) return 0\n  const n = a.length\n  const meanA = a.reduce((s, v) => s + v, 0) / n\n  const meanB = b.reduce((s, v) => s + v, 0) / n\n  let num = 0\n  let denomA = 0\n  let denomB = 0\n  for (let i = 0; i < n; i++) {\n    const da = a[i]! - meanA\n    const db = b[i]! - meanB\n    num += da * db\n    denomA += da * da\n    denomB += db * db\n  }\n  const denom = Math.sqrt(denomA * denomB)\n  return denom === 0 ? 0 : num / denom\n}\n\n// ── Lift ────────────────────────────────────────────────────────────\n\nfunction computeLift(\n  runs: RunRecord[],\n  baselineId: string | undefined,\n  candidateId: string | undefined,\n  split: 'search' | 'holdout',\n): LiftInsight | undefined {\n  let bId = baselineId\n  let cId = candidateId\n  if (!bId || !cId) {\n    // Auto-detect: when exactly two distinct candidateIds appear, treat the\n    // lower-mean side as baseline.\n    const ids = [...new Set(runs.map((r) => r.candidateId))]\n    if (ids.length !== 2) return undefined\n    const [idA, idB] = ids as [string, string]\n    const meanA = mean(runs.filter((r) => r.candidateId === idA).map((r) => compositeOf(r, split)))\n    const meanB = mean(runs.filter((r) => r.candidateId === idB).map((r) => compositeOf(r, split)))\n    bId = meanA <= meanB ? idA : idB\n    cId = meanA <= meanB ? idB : idA\n  }\n\n  const baseline = runs.filter((r) => r.candidateId === bId)\n  const candidate = runs.filter((r) => r.candidateId === cId)\n  if (baseline.length === 0 || candidate.length === 0) return undefined\n\n  // Pair on (experimentId, seed). When that key doesn't match, fall back\n  // to ordinal pairing — common for fresh runs from the same scenario list.\n  const baselineByKey = new Map(baseline.map((r) => [pairingKey(r), r]))\n  const pairedBaseline: number[] = []\n  const pairedCandidate: number[] = []\n  let usedKeyPairing = false\n  for (const cand of candidate) {\n    const b = baselineByKey.get(pairingKey(cand))\n    if (b) {\n      const bC = compositeOf(b, split)\n      const cC = compositeOf(cand, split)\n      if (Number.isFinite(bC) && Number.isFinite(cC)) {\n        pairedBaseline.push(bC)\n        pairedCandidate.push(cC)\n        usedKeyPairing = true\n      }\n    }\n  }\n  if (!usedKeyPairing) {\n    const n = Math.min(baseline.length, candidate.length)\n    for (let i = 0; i < n; i++) {\n      const bC = compositeOf(baseline[i]!, split)\n      const cC = compositeOf(candidate[i]!, split)\n      if (Number.isFinite(bC) && Number.isFinite(cC)) {\n        pairedBaseline.push(bC)\n        pairedCandidate.push(cC)\n      }\n    }\n  }\n  if (pairedBaseline.length === 0) return undefined\n\n  const baselineMean = mean(pairedBaseline)\n  const candidateMean = mean(pairedCandidate)\n  const delta = candidateMean - baselineMean\n\n  const bootstrap = pairedBootstrap(pairedBaseline, pairedCandidate, {\n    confidence: 0.95,\n    resamples: 2000,\n    statistic: 'mean',\n  })\n  const tTest = pairedTTest(pairedBaseline, pairedCandidate)\n  const d = cohensD(pairedBaseline, pairedCandidate)\n  const mde = pairedMde({ nPaired: pairedBaseline.length, power: 0.8, alpha: 0.05 })\n  const requiredN = requiredSampleSize({\n    effect: Math.max(Math.abs(delta), 1e-6),\n    power: 0.8,\n    alpha: 0.05,\n  })\n\n  return {\n    baselineMean,\n    candidateMean,\n    delta,\n    ci95: [bootstrap.low, bootstrap.high],\n    pValue: tTest.p,\n    n: pairedBaseline.length,\n    cohensD: d,\n    mde,\n    requiredN,\n  }\n}\n\nfunction pairingKey(r: RunRecord): string {\n  return `${r.experimentId}::${r.seed}`\n}\n\nfunction mean(arr: number[]): number {\n  return arr.length === 0 ? 0 : arr.reduce((s, v) => s + v, 0) / arr.length\n}\n\n// ── Failure clustering ──────────────────────────────────────────────\n\nasync function computeFailureClusters(\n  runs: RunRecord[],\n  analyst: AnalystRegistry,\n  split: 'search' | 'holdout',\n): Promise<FailureClusterInsight | undefined> {\n  const failed = runs.filter((r) => compositeOf(r, split) < 0.5 || r.failureMode !== undefined)\n  if (failed.length === 0) return { clusters: [], totalFailures: 0 }\n\n  const clusters = new Map<string, { exemplars: string[]; share: number }>()\n  for (const run of failed) {\n    try {\n      const result = await analyst.run(run.runId, {\n        kind: 'run-record',\n        run,\n      } as Parameters<typeof analyst.run>[1])\n      for (const finding of result.findings as AnalystFinding[]) {\n        const key = finding.area || finding.analyst_id || 'unclassified'\n        const c = clusters.get(key) ?? { exemplars: [], share: 0 }\n        if (c.exemplars.length < 5) c.exemplars.push(run.runId)\n        clusters.set(key, c)\n      }\n    } catch {\n      const c = clusters.get('analyst-error') ?? { exemplars: [], share: 0 }\n      if (c.exemplars.length < 5) c.exemplars.push(run.runId)\n      clusters.set('analyst-error', c)\n    }\n  }\n  const clusterList = [...clusters.entries()].map(([id, c]) => ({\n    id,\n    name: id,\n    share: c.exemplars.length / failed.length,\n    exemplars: c.exemplars,\n  }))\n  clusterList.sort((a, b) => b.share - a.share)\n  return { clusters: clusterList, totalFailures: failed.length }\n}\n\n// ── Contamination ──────────────────────────────────────────────────\n\nfunction computeContamination(\n  runs: RunRecord[],\n  canaries: DatasetScenario[],\n): InsightReport['contamination'] {\n  let leaks = 0\n  const details: Array<{ runId: string; canary: string; matched: string }> = []\n  for (const run of runs) {\n    const output = stringifyOutput(run)\n    if (!output) continue\n    const leaksHere = checkCanaries(output, canaries)\n    for (const leak of leaksHere) {\n      leaks++\n      details.push({ runId: run.runId, canary: leak.canary, matched: leak.evidence })\n    }\n  }\n  return { leaks, holdoutAuditPassed: leaks === 0, details }\n}\n\nfunction stringifyOutput(run: RunRecord): string | undefined {\n  // RunRecord doesn't fix where \"the agent's output\" lives — different\n  // consumers stash it differently. We probe the common shapes: the\n  // outcome.raw map (numeric only by design — unlikely to contain text),\n  // and any string-valued fields tucked under metadata via type casting.\n  // Consumers with bespoke shapes pass canaryScenarios only when they\n  // know their runs carry a stringifiable surface.\n  const metadata = (run as unknown as { metadata?: Record<string, unknown> }).metadata\n  if (typeof metadata?.output === 'string') return metadata.output\n  if (typeof metadata?.text === 'string') return metadata.text\n  return undefined\n}\n\n// ── Outcome correlation + linear reward model ──────────────────────\n\nfunction computeOutcomeCorrelation(\n  runs: RunRecord[],\n  outcome: { metric: string; valueByRunId: Record<string, number> },\n  split: 'search' | 'holdout',\n): OutcomeCorrelationInsight | undefined {\n  const xs: number[] = []\n  const ys: number[] = []\n  for (const run of runs) {\n    const y = outcome.valueByRunId[run.runId]\n    if (y === undefined || !Number.isFinite(y)) continue\n    const x = compositeOf(run, split)\n    if (!Number.isFinite(x)) continue\n    xs.push(x)\n    ys.push(y)\n  }\n  if (xs.length < 3) return undefined\n\n  const p = pearson(xs, ys)\n  const s = spearman(xs, ys)\n  const meanX = mean(xs)\n  const meanY = mean(ys)\n  let num = 0\n  let denom = 0\n  for (let i = 0; i < xs.length; i++) {\n    num += (xs[i]! - meanX) * (ys[i]! - meanY)\n    denom += (xs[i]! - meanX) ** 2\n  }\n  const slope = denom === 0 ? 0 : num / denom\n  const intercept = meanY - slope * meanX\n  const ssTot = ys.reduce((a, y) => a + (y - meanY) ** 2, 0)\n  const ssRes = ys.reduce((a, y, i) => a + (y - (intercept + slope * xs[i]!)) ** 2, 0)\n  const r2 = ssTot === 0 ? 0 : 1 - ssRes / ssTot\n\n  return {\n    metric: outcome.metric,\n    n: xs.length,\n    pearson: p,\n    spearman: s,\n    rewardModel: { intercept, slope, r2 },\n  }\n}\n\nfunction spearman(a: number[], b: number[]): number {\n  if (a.length !== b.length || a.length === 0) return 0\n  return pearson(rank(a), rank(b))\n}\n\nfunction rank(arr: number[]): number[] {\n  const indexed = arr.map((v, i) => ({ v, i }))\n  indexed.sort((x, y) => x.v - y.v)\n  const ranks = new Array(arr.length).fill(0)\n  let i = 0\n  while (i < indexed.length) {\n    let j = i\n    while (j + 1 < indexed.length && indexed[j + 1]!.v === indexed[i]!.v) j++\n    const avg = (i + j + 2) / 2\n    for (let k = i; k <= j; k++) ranks[indexed[k]!.i] = avg\n    i = j + 1\n  }\n  return ranks\n}\n\n// ── Release confidence scorecard ───────────────────────────────────\n\nfunction buildReleaseScorecard(\n  composite: ScalarDistribution,\n  lift: LiftInsight | undefined,\n  contamination: InsightReport['contamination'],\n): InsightReport['release'] {\n  // Synthesise a minimal scorecard from the rolled-up signal. The\n  // substrate's `evaluateReleaseConfidence` primitive consumes a richer\n  // input shape that callers can produce by wiring SLO definitions; the\n  // shape here is the contract `selfImprove`/`analyzeRuns` consumers\n  // receive automatically. They can call `evaluateReleaseConfidence`\n  // directly when they want SLO-based axis evaluation.\n  const axes: InsightReport['release']['axes'] = []\n  const liftPass =\n    lift === undefined || lift.ci95[0] > 0\n      ? ('pass' as const)\n      : lift.delta > 0\n        ? ('warn' as const)\n        : ('fail' as const)\n  axes.push({\n    name: 'quality-lift',\n    status: liftPass,\n    detail: lift\n      ? `delta=${lift.delta.toFixed(3)}, CI95=[${lift.ci95[0].toFixed(3)}, ${lift.ci95[1].toFixed(3)}], n=${lift.n}`\n      : 'no baseline/candidate pair available',\n  })\n  const contamPass =\n    contamination === undefined || contamination.leaks === 0 ? ('pass' as const) : ('fail' as const)\n  axes.push({\n    name: 'contamination',\n    status: contamPass,\n    detail: contamination ? `${contamination.leaks} canary leak(s)` : 'no canaries supplied',\n  })\n  axes.push({\n    name: 'composite-distribution',\n    status: composite.mean >= 0.5 ? 'pass' : composite.mean >= 0.3 ? 'warn' : 'fail',\n    detail: `mean=${composite.mean.toFixed(3)}, p50=${composite.p50.toFixed(3)}, p95=${composite.p95.toFixed(3)} over n=${composite.n}`,\n  })\n  const status = axes.some((a) => a.status === 'fail')\n    ? 'fail'\n    : axes.some((a) => a.status === 'warn')\n      ? 'warn'\n      : 'pass'\n  return {\n    status,\n    axes,\n    issues: [],\n  }\n}\n\n// ── Recommendations engine ─────────────────────────────────────────\n\ninterface RecommendationContext {\n  composite: ScalarDistribution\n  judges: Record<string, JudgeInsight>\n  interRater?: InterRaterInsight\n  lift?: LiftInsight\n  failureClusters?: FailureClusterInsight\n  contamination?: InsightReport['contamination']\n  outcomeCorrelation?: OutcomeCorrelationInsight\n  threshold: number\n}\n\nfunction buildRecommendations(ctx: RecommendationContext): Recommendation[] {\n  const out: Recommendation[] = []\n\n  // Composite-distribution branch. Fires when the overall quality signal is\n  // poor regardless of lift / contamination / clusters — the customer needs\n  // to know they have a problem AND which specific runs to inspect.\n  if (ctx.composite.n > 0) {\n    if (ctx.composite.mean < 0.3) {\n      const tail = ctx.composite.tailRuns ?? []\n      const names = tail\n        .slice(0, 5)\n        .map((t) => `${t.runId}=${t.score.toFixed(3)}`)\n        .join(', ')\n      out.push({\n        priority: 'critical',\n        kind: 'investigate',\n        title: `Composite mean ${ctx.composite.mean.toFixed(3)} is below the 0.3 floor — the agent is broken on this corpus`,\n        detail:\n          tail.length > 0\n            ? `Worst ${tail.length} run${tail.length === 1 ? '' : 's'} to inspect first: ${names}. Histogram p50=${ctx.composite.p50.toFixed(3)}, p95=${ctx.composite.p95.toFixed(3)}.`\n            : `Histogram p50=${ctx.composite.p50.toFixed(3)}, p95=${ctx.composite.p95.toFixed(3)}.`,\n        evidencePath: 'composite.tailRuns',\n      })\n    } else if (ctx.composite.mean < 0.5) {\n      const tail = ctx.composite.tailRuns ?? []\n      const names = tail\n        .slice(0, 3)\n        .map((t) => `${t.runId}=${t.score.toFixed(3)}`)\n        .join(', ')\n      out.push({\n        priority: 'high',\n        kind: 'investigate',\n        title: `Composite mean ${ctx.composite.mean.toFixed(3)} is below 0.5 — investigate the lower tail before claiming the agent is healthy`,\n        detail:\n          tail.length > 0\n            ? `Worst ${tail.length} run${tail.length === 1 ? '' : 's'}: ${names}. Histogram p50=${ctx.composite.p50.toFixed(3)}, p95=${ctx.composite.p95.toFixed(3)}.`\n            : `Histogram p50=${ctx.composite.p50.toFixed(3)}, p95=${ctx.composite.p95.toFixed(3)}.`,\n        evidencePath: 'composite.tailRuns',\n      })\n    }\n  }\n\n  // Missing-judges branch. The report can't surface per-dimension or\n  // calibration signal when `outcome.judgeScores` is empty across the\n  // corpus. Tell the customer how to enrich.\n  if (Object.keys(ctx.judges).length === 0 && ctx.composite.n > 0) {\n    out.push({\n      priority: 'medium',\n      kind: 'expand-corpus',\n      title: 'No judge scores recorded — per-dimension + calibration insights unavailable',\n      detail:\n        'Records have no `outcome.judgeScores`. To unlock perDimension, judges, and calibration, attach a Judge run during your eval pass and populate `outcome.judgeScores.perJudge[judgeName][dimension] = score`. See `docs/insight-report.md` for the expected shape.',\n      evidencePath: 'judges',\n    })\n  }\n\n  if (ctx.lift) {\n    const decisive = ctx.lift.ci95[0] > ctx.threshold\n    const inconclusive = ctx.lift.ci95[0] <= ctx.threshold && ctx.lift.ci95[1] > ctx.threshold\n    if (decisive) {\n      out.push({\n        priority: 'critical',\n        kind: 'ship',\n        title: `Ship — lift ${ctx.lift.delta.toFixed(3)} (95% CI ${ctx.lift.ci95[0].toFixed(3)}..${ctx.lift.ci95[1].toFixed(3)})`,\n        detail: `Holdout lift exceeds threshold ${ctx.threshold} with 95% bootstrap confidence (n=${ctx.lift.n}, p=${ctx.lift.pValue.toFixed(4)}, d=${ctx.lift.cohensD.toFixed(2)}).`,\n        evidencePath: 'lift',\n      })\n    } else if (inconclusive) {\n      out.push({\n        priority: 'high',\n        kind: 'expand-corpus',\n        title: `Inconclusive — need ~${ctx.lift.requiredN} paired runs (have ${ctx.lift.n}) at current effect size`,\n        detail: `CI straddles threshold. Current MDE at 80% power is ${ctx.lift.mde.toFixed(3)}; observed delta is ${ctx.lift.delta.toFixed(3)}.`,\n        evidencePath: 'lift',\n      })\n    } else {\n      out.push({\n        priority: 'critical',\n        kind: 'hold',\n        title: `Hold — lift CI lower bound ${ctx.lift.ci95[0].toFixed(3)} is at or below threshold ${ctx.threshold}`,\n        detail: `Bootstrap CI provides no statistical evidence the candidate is better. Consider tightening the mutation or expanding the holdout.`,\n        evidencePath: 'lift',\n      })\n    }\n  }\n\n  if (ctx.contamination && ctx.contamination.leaks > 0) {\n    out.push({\n      priority: 'critical',\n      kind: 'fix',\n      title: `${ctx.contamination.leaks} canary leak${ctx.contamination.leaks === 1 ? '' : 's'} detected`,\n      detail: `Holdout integrity is compromised. The lift number is unreliable until you investigate.`,\n      evidencePath: 'contamination',\n    })\n  }\n\n  if (ctx.interRater && ctx.interRater.kappa < 0.5) {\n    out.push({\n      priority: 'high',\n      kind: 'recalibrate',\n      title: `Inter-rater agreement κ=${ctx.interRater.kappa.toFixed(2)} is below 0.5`,\n      detail: `Raters disagree on what 'good' looks like. Top disagreement cases listed in interRater.disagreementCases — consider a triage meeting or refining the rubric.`,\n      evidencePath: 'interRater',\n    })\n  }\n\n  if (ctx.failureClusters && ctx.failureClusters.clusters.length > 0) {\n    const top = ctx.failureClusters.clusters[0]!\n    out.push({\n      priority: 'high',\n      kind: 'investigate',\n      title: `Top failure cluster: ${top.name} (${(top.share * 100).toFixed(0)}% of failures)`,\n      detail: `${ctx.failureClusters.totalFailures} runs failed. The largest cluster groups ${top.exemplars.length} exemplars under '${top.name}'.`,\n      evidencePath: 'failureClusters.clusters[0]',\n    })\n  }\n\n  if (ctx.outcomeCorrelation && Math.abs(ctx.outcomeCorrelation.spearman) < 0.3) {\n    out.push({\n      priority: 'medium',\n      kind: 'recalibrate',\n      title: `Judge scores decoupled from ${ctx.outcomeCorrelation.metric} (Spearman ρ=${ctx.outcomeCorrelation.spearman.toFixed(2)})`,\n      detail: `Your judges score what they were trained to score, but it isn't predicting downstream ${ctx.outcomeCorrelation.metric}. Consider retraining the judge against ${ctx.outcomeCorrelation.metric} as the gold signal.`,\n      evidencePath: 'outcomeCorrelation',\n    })\n  }\n\n  return out\n}\n\n// ── Re-export pareto figure spec for hosted-side rendering ─────────\n\nexport type { ParetoFigureSpec }\n","/**\n * # `selfImprove()` — the LAND-tier one-shot.\n *\n * The cheapest possible call site to run a real closed-loop self-\n * improvement over your agent. Wraps `runImprovementLoop` with smart\n * defaults and a budget-shaped options API; every escape hatch the\n * substrate exposes is reachable from here without losing the\n * one-function feel.\n *\n * Defaults picked to match the LAND-tier story:\n *   - In-memory storage (no filesystem touch).\n *   - `gepaDriver` reflective mutation with copywriting-flavored primitives\n *     (override `driver` or `mutationPrimitives` for any domain).\n *   - `defaultProductionGate` with `deltaThreshold: 0.05`.\n *   - Held-out split = 25% of scenarios, deterministic by id hash.\n *   - 3 generations × population 2 (raise via `budget` for more search).\n *   - `autoOnPromote: 'none'` (we don't open PRs unless you ask).\n *\n * Want one-click? Provide `agent` + `scenarios` + `judge`. Done.\n * Want distributed? Pass `cellPlacement` + an `httpDispatch`-backed\n * agent. Want a code-tier surface? Pass a `MutableSurface` + your own\n * `driver`. Same function.\n */\n\nimport { gepaDriver } from '../campaign/drivers/gepa'\nimport { defaultProductionGate } from '../campaign/gates/default-production-gate'\nimport {\n  type RunImprovementLoopResult,\n  runImprovementLoop,\n} from '../campaign/presets/run-improvement-loop'\nimport { type CampaignStorage, inMemoryCampaignStorage } from '../campaign/storage'\nimport type {\n  CampaignCellResult,\n  DispatchContext,\n  Gate,\n  ImprovementDriver,\n  JudgeConfig,\n  MutableSurface,\n  Scenario,\n} from '../campaign/types'\nimport { createHostedClient, type HostedTenant } from '../hosted/client'\nimport type { EvalRunCellScore, EvalRunEvent, EvalRunGenerationSnapshot } from '../hosted/types'\nimport type { JudgeScoresRecord, RunRecord } from '../run-record'\nimport { analyzeRuns } from './analyze-runs'\nimport type { InsightReport } from './insight-report'\n\nexport interface SelfImproveBudget {\n  /** Hard $ ceiling across all cells in baseline + every generation. Cells\n   *  beyond the ceiling are skipped (cost-aware, not aborted). */\n  dollars?: number\n  /** How many improvement generations to explore. Default 3. Set 0 to\n   *  skip improvement entirely (selfImprove becomes a baseline-only run). */\n  generations?: number\n  /** Candidates the driver proposes per generation. Default 2. */\n  populationSize?: number\n  /** Max concurrent cells across the loop. Default 2. */\n  maxConcurrency?: number\n  /** Fraction of `scenarios` held out from training, used for the gate.\n   *  Default 0.25. Ignored when `holdoutScenarios` is set explicitly. */\n  holdoutFraction?: number\n  /** Explicit held-out scenarios; overrides `holdoutFraction`. */\n  holdoutScenarios?: Scenario[]\n}\n\nexport interface SelfImproveLlm {\n  /** Endpoint base URL. Default Tangle Router. */\n  baseUrl?: string\n  /** Bearer token. Default `process.env.OPENAI_API_KEY`. */\n  apiKey?: string\n  /** Model id used by `gepaDriver` reflection. Default\n   *  `anthropic/claude-sonnet-4.6`. */\n  model?: string\n}\n\nexport type SelfImproveProgressEvent =\n  | { kind: 'baseline.started'; scenarios: number }\n  | { kind: 'baseline.completed'; compositeMean: number; durationMs: number }\n  | { kind: 'generation.started'; index: number; populationSize: number }\n  | { kind: 'generation.completed'; index: number; bestComposite: number; durationMs: number }\n  | { kind: 'gate.decided'; decision: string; lift: number }\n\nexport interface SelfImproveOptions<TScenario extends Scenario, TArtifact> {\n  /**\n   * Your agent — a function that takes the current `MutableSurface`\n   * (typically a system prompt the loop is optimizing) plus the\n   * scenario + cell ctx, and returns the artifact your judge scores.\n   *\n   * Same shape as `RunOptimizationOptions.dispatchWithSurface`. Wrap a\n   * plain `Dispatch` if you don't have a surface seam:\n   *\n   *   agent: (_surface, scenario, ctx) => yourPlainDispatch(scenario, ctx)\n   *\n   * That mode evaluates without mutating any surface — useful as a\n   * baseline-only run (set `budget.generations = 0`).\n   */\n  agent: (surface: MutableSurface, scenario: TScenario, ctx: DispatchContext) => Promise<TArtifact>\n\n  /** Scenarios to evaluate against. Train/holdout split is computed from\n   *  these unless `budget.holdoutScenarios` is set explicitly. */\n  scenarios: TScenario[]\n\n  /** Judge that scores artifacts. Bring your own; use `langchainJudge`\n   *  from `/adapters/langchain` for a Runnable-shaped one. */\n  judge: JudgeConfig<TArtifact, TScenario>\n\n  /** Starting surface — system prompt, JSON config, anything `MutableSurface`\n   *  accepts. The driver mutates this each generation. */\n  baselineSurface: MutableSurface\n\n  /** Budget + loop shape. All fields optional; defaults pick the LAND-tier\n   *  story. */\n  budget?: SelfImproveBudget\n\n  /** Custom driver. Default is `gepaDriver` configured from `llm` +\n   *  `mutationPrimitives`. */\n  driver?: ImprovementDriver\n\n  /** Default-driver overrides — used when `driver` is unset. */\n  mutationPrimitives?: string[]\n  driverTarget?: string\n\n  /** Custom gate. Default is `defaultProductionGate` with\n   *  `deltaThreshold: 0.05` on the held-out split. */\n  gate?: Gate<TArtifact, TScenario>\n\n  /** LLM config consumed by the default `gepaDriver`. Ignored if you pass\n   *  your own `driver`. */\n  llm?: SelfImproveLlm\n\n  /** Storage backend. Default `inMemoryCampaignStorage()` — nothing\n   *  persists past the call. Pass `fsCampaignStorage()` to write to disk. */\n  storage?: CampaignStorage\n\n  /** Run directory (logical for in-memory storage, real path for fs).\n   *  Default `mem://selfImprove-<timestamp>`. */\n  runDir?: string\n\n  /** Distributed-driver seam — same as `RunCampaignOptions.cellPlacement`.\n   *  Returns an opaque placement key the substrate forwards to your agent\n   *  as `ctx.placement`. Combined with `httpDispatch` from\n   *  `/adapters/http`, fans cells across regions. */\n  cellPlacement?: (input: {\n    scenario: TScenario\n    rep: number\n    generation?: number\n  }) => string | undefined\n\n  /** Streaming hook — fires on baseline + each generation + gate decision.\n   *  Consumer routes events wherever (UI, dashboard, logs). */\n  onProgress?: (event: SelfImproveProgressEvent) => void\n\n  /** Auto-promotion behavior on a ship decision. Default `'none'` — we\n   *  return the winner; you ship it however you ship. `'pr'` opens a\n   *  GitHub PR via `openAutoPr`; requires `ghOwner` + `ghRepo`. */\n  autoOnPromote?: 'pr' | 'none'\n  ghOwner?: string\n  ghRepo?: string\n\n  /**\n   * Opt-in: ship eval-run events to a hosted orchestrator (ours, your\n   * self-hosted one, or any compatible implementation of the\n   * `docs/hosted-ingest-spec.md` wire format). When set, the substrate\n   * POSTs the final `EvalRunEvent` to `${endpoint}/v1/ingest/eval-runs`\n   * after the loop completes. Failures are logged but do not fail the\n   * loop — local result is always returned.\n   *\n   * For our orchestrator: `{ endpoint: 'https://orchestrator.tangle.tools/v1', apiKey, tenantId }`.\n   *\n   * For your self-hosted: any URL serving the wire format. See\n   * `examples/hosted-ingest-server/` for the reference receiver.\n   */\n  hostedTenant?: HostedTenant\n\n  /** Free-form labels attached to the hosted event (env, branch, model id,\n   *  etc.). Ignored when `hostedTenant` is unset. */\n  hostedLabels?: Record<string, string>\n}\n\nexport interface SelfImproveResult<TScenario extends Scenario, TArtifact> {\n  /** Composite mean across all scenarios, baseline run. */\n  baseline: {\n    compositeMean: number\n    perScenario: Record<string, number>\n  }\n  /** Composite mean on the held-out set, winner run. */\n  winner: {\n    compositeMean: number\n    perScenario: Record<string, number>\n    surface: MutableSurface\n  }\n  /** `winner.compositeMean - baselineOnHoldout.compositeMean`. Positive\n   *  means the gate observed improvement. */\n  lift: number\n  /** `defaultProductionGate.decide()` result. */\n  gateDecision: 'ship' | 'hold' | 'need_more_work' | 'model_ceiling' | 'arch_ceiling'\n  /** Number of generations actually explored (may be less than the\n   *  budget if the driver gave up early). */\n  generationsExplored: number\n  /** Wall-clock total. */\n  durationMs: number\n  /** Total cost across baseline + every generation. */\n  totalCostUsd: number\n  /**\n   * Rigor packet: distributional summary, paired-bootstrap lift CI,\n   * judge stats, contamination check, recommendations. Wired through\n   * `analyzeRuns()` on the baseline + winner cells of the campaign.\n   * Hosted-tier dashboards render this as the v3-vs-v4 decision view.\n   */\n  insight: InsightReport\n  /**\n   * Raw substrate result for advanced inspection — full per-generation\n   * candidates, full campaign artifacts, all judge scores. Useful for\n   * debugging or reporting beyond the summary.\n   */\n  raw: RunImprovementLoopResult<TArtifact, TScenario>\n}\n\n/**\n * Deterministic train/holdout split by a stable hash of `scenario.id`,\n * so the same scenario set always splits the same way across runs.\n */\nfunction splitTrainHoldout<TScenario extends Scenario>(\n  scenarios: TScenario[],\n  fraction: number,\n): { train: TScenario[]; holdout: TScenario[] } {\n  // Stable fnv-1a-ish hash of the id for ordering.\n  function hash(s: string): number {\n    let h = 2166136261 >>> 0\n    for (let i = 0; i < s.length; i++) {\n      h ^= s.charCodeAt(i)\n      h = Math.imul(h, 16777619) >>> 0\n    }\n    return h\n  }\n  const sorted = [...scenarios].sort((a, b) => hash(a.id) - hash(b.id))\n  const nHoldout = Math.max(1, Math.min(sorted.length - 1, Math.round(sorted.length * fraction)))\n  return {\n    holdout: sorted.slice(0, nHoldout),\n    train: sorted.slice(nHoldout),\n  }\n}\n\nfunction meanComposite(byScenario: Record<string, { meanComposite: number }>): {\n  compositeMean: number\n  perScenario: Record<string, number>\n} {\n  const perScenario: Record<string, number> = {}\n  const values: number[] = []\n  for (const [id, agg] of Object.entries(byScenario)) {\n    perScenario[id] = agg.meanComposite\n    values.push(agg.meanComposite)\n  }\n  return {\n    compositeMean: values.length === 0 ? 0 : values.reduce((s, v) => s + v, 0) / values.length,\n    perScenario,\n  }\n}\n\nconst DEFAULT_MUTATION_PRIMITIVES = [\n  'Tighten the hook: lead with the specific user outcome.',\n  'Replace generic adjectives with specific verbs or proof numbers.',\n  \"Anchor every claim in something the scenario's brief literally supports.\",\n  'Honor the surface-shape constraint (length, register, audience vocabulary).',\n]\n\n/**\n * One-shot self-improvement loop. See module docstring for defaults +\n * extension points.\n *\n * @example Minimum (LAND tier):\n *\n *   const result = await selfImprove({\n *     agent: (surface, scenario, ctx) => myAgent(surface, scenario, ctx.signal),\n *     scenarios,\n *     judge,\n *     baselineSurface: DEFAULT_PROMPT,\n *   })\n *   console.log(`lift: ${result.lift.toFixed(3)} (${result.gateDecision})`)\n *\n * @example Distributed (workers in three regions):\n *\n *   await selfImprove({\n *     agent: httpDispatch({ resolveUrl: ({ placement }) => REGION_URLS[placement!] }),\n *     scenarios,\n *     judge,\n *     baselineSurface: DEFAULT_PROMPT,\n *     cellPlacement: ({ scenario }) => scenario.region,\n *     budget: { maxConcurrency: 12 },\n *   })\n */\nexport async function selfImprove<TScenario extends Scenario, TArtifact>(\n  opts: SelfImproveOptions<TScenario, TArtifact>,\n): Promise<SelfImproveResult<TScenario, TArtifact>> {\n  const startedAt = Date.now()\n\n  const budget = opts.budget ?? {}\n  const generations = budget.generations ?? 3\n  const populationSize = budget.populationSize ?? 2\n  const maxConcurrency = budget.maxConcurrency ?? 2\n  const holdoutFraction = budget.holdoutFraction ?? 0.25\n  const costCeiling = budget.dollars\n\n  const explicitHoldout = budget.holdoutScenarios\n  const { train, holdout } = explicitHoldout\n    ? {\n        train: opts.scenarios.filter((s) => !explicitHoldout.some((h) => h.id === s.id)),\n        holdout: explicitHoldout as TScenario[],\n      }\n    : splitTrainHoldout(opts.scenarios, holdoutFraction)\n\n  if (train.length === 0) {\n    throw new Error(\n      'selfImprove: train split is empty. Reduce holdoutFraction or pass more scenarios.',\n    )\n  }\n  if (holdout.length === 0) {\n    throw new Error('selfImprove: holdout split is empty. Pass more scenarios.')\n  }\n\n  const driver: ImprovementDriver =\n    opts.driver ??\n    gepaDriver({\n      llm: {\n        baseUrl: opts.llm?.baseUrl ?? 'https://router.tangle.tools/v1',\n        apiKey: opts.llm?.apiKey ?? process.env.OPENAI_API_KEY ?? '',\n      },\n      model: opts.llm?.model ?? 'anthropic/claude-sonnet-4.6',\n      target:\n        opts.driverTarget ??\n        'agent surface (system prompt or config) being optimized by selfImprove',\n      mutationPrimitives: opts.mutationPrimitives ?? DEFAULT_MUTATION_PRIMITIVES,\n    })\n\n  const gate: Gate<TArtifact, TScenario> =\n    opts.gate ??\n    defaultProductionGate<TArtifact, TScenario>({\n      holdoutScenarios: holdout,\n      deltaThreshold: 0.05,\n    })\n\n  const storage = opts.storage ?? inMemoryCampaignStorage()\n  const runDir = opts.runDir ?? `mem://selfImprove-${startedAt}`\n\n  if (opts.onProgress) {\n    opts.onProgress({ kind: 'baseline.started', scenarios: opts.scenarios.length })\n  }\n\n  const result = await runImprovementLoop<TScenario, TArtifact>({\n    scenarios: train,\n    baselineSurface: opts.baselineSurface,\n    dispatchWithSurface: opts.agent,\n    driver,\n    judges: [opts.judge],\n    populationSize,\n    maxGenerations: generations,\n    holdoutScenarios: holdout,\n    gate,\n    autoOnPromote: opts.autoOnPromote ?? 'none',\n    ghOwner: opts.ghOwner,\n    ghRepo: opts.ghRepo,\n    storage,\n    runDir,\n    maxConcurrency,\n    cellPlacement: opts.cellPlacement,\n    costCeiling,\n  })\n\n  const baseline = meanComposite(result.baselineOnHoldout.aggregates.byScenario)\n  const winnerStats = meanComposite(result.winnerOnHoldout.aggregates.byScenario)\n\n  if (opts.onProgress) {\n    opts.onProgress({\n      kind: 'baseline.completed',\n      compositeMean: baseline.compositeMean,\n      durationMs: Date.now() - startedAt,\n    })\n    opts.onProgress({\n      kind: 'gate.decided',\n      decision: result.gateResult.decision,\n      lift: winnerStats.compositeMean - baseline.compositeMean,\n    })\n  }\n\n  const totalCost =\n    result.baselineCampaign.aggregates.totalCostUsd +\n    result.generations.reduce(\n      (sum, gen) =>\n        sum + gen.surfaces.reduce((s, sf) => s + sf.campaign.aggregates.totalCostUsd, 0),\n      0,\n    )\n\n  // Rigor packet: feed baseline + winner cells through analyzeRuns().\n  // The two candidates (`baseline` / `winner`) give the lift section a\n  // clean paired comparison; per-judge / per-dimension / cost-quality\n  // sections populate from the cells' judgeScores.\n  const insight = await analyzeRuns({\n    runs: [\n      ...cellsToRunRecords(result.baselineCampaign.cells, 'baseline', runDir),\n      ...cellsToRunRecords(result.winnerOnHoldout.cells, 'winner', runDir),\n    ],\n    baselineCandidateId: 'baseline',\n    candidateCandidateId: 'winner',\n  })\n\n  const summary: SelfImproveResult<TScenario, TArtifact> = {\n    baseline,\n    winner: {\n      ...winnerStats,\n      surface: result.winnerSurface,\n    },\n    lift: winnerStats.compositeMean - baseline.compositeMean,\n    gateDecision: result.gateResult.decision,\n    generationsExplored: result.generations.length,\n    durationMs: Date.now() - startedAt,\n    totalCostUsd: totalCost,\n    insight,\n    raw: result,\n  }\n\n  // Opt-in hosted ingest. Failures logged but never fail the loop — the\n  // local result is always returned. This matches the wedge-doc invariant\n  // that LAND-tier never blocks on EXPAND-tier infra.\n  if (opts.hostedTenant) {\n    try {\n      await shipEvalRunToHosted(opts.hostedTenant, opts, summary, result, runDir)\n    } catch (err) {\n      const msg = err instanceof Error ? err.message : String(err)\n      // eslint-disable-next-line no-console -- intentional: hosted-ingest is best-effort\n      console.warn(`[agent-eval] hosted ingest failed (continuing): ${msg}`)\n    }\n  }\n\n  return summary\n}\n\nasync function shipEvalRunToHosted<TScenario extends Scenario, TArtifact>(\n  tenant: HostedTenant,\n  opts: SelfImproveOptions<TScenario, TArtifact>,\n  summary: SelfImproveResult<TScenario, TArtifact>,\n  raw: RunImprovementLoopResult<TArtifact, TScenario>,\n  runDir: string,\n): Promise<void> {\n  const client = createHostedClient(tenant)\n\n  function snapshotFromCampaign(\n    index: number,\n    surface: MutableSurface | undefined,\n    campaign: RunImprovementLoopResult<TArtifact, TScenario>['baselineCampaign'],\n    durationMs: number,\n  ): EvalRunGenerationSnapshot {\n    const cells: EvalRunCellScore[] = campaign.cells.map((cell) => {\n      const judgeScores = Object.values(cell.judgeScores)\n      const composite =\n        judgeScores.length === 0\n          ? 0\n          : judgeScores.reduce((s, j) => s + j.composite, 0) / judgeScores.length\n      return {\n        scenarioId: cell.scenarioId,\n        rep: cell.rep,\n        compositeMean: composite,\n        dimensions: Object.fromEntries(\n          Object.entries(cell.judgeScores).map(([name, score]) => [name, score.dimensions]),\n        ),\n        errorMessage: cell.error ?? undefined,\n      }\n    })\n    const compositeMean =\n      cells.length === 0 ? 0 : cells.reduce((s, c) => s + c.compositeMean, 0) / cells.length\n    return {\n      index,\n      surfaceHash:\n        typeof surface === 'string'\n          ? hashString(surface)\n          : hashString(JSON.stringify(surface ?? '')),\n      surface,\n      cells,\n      compositeMean,\n      costUsd: campaign.aggregates.totalCostUsd,\n      durationMs,\n    }\n  }\n\n  const generations: EvalRunGenerationSnapshot[] = []\n  // Baseline as generation 0.\n  generations.push(snapshotFromCampaign(0, opts.baselineSurface, raw.baselineCampaign, 0))\n  // Improvement generations as 1..N. Substrate stores per-surface campaigns\n  // per generation — we summarize the WINNING surface per generation here.\n  for (const gen of raw.generations) {\n    const winner = gen.surfaces.reduce(\n      (best, s) =>\n        s.campaign.aggregates.cellsExecuted > 0 &&\n        (best === undefined || averageComposite(s.campaign) > averageComposite(best.campaign))\n          ? s\n          : best,\n      gen.surfaces[0],\n    )\n    if (!winner) continue\n    generations.push(\n      snapshotFromCampaign(gen.record.generationIndex + 1, winner.surface, winner.campaign, 0),\n    )\n  }\n\n  const event: EvalRunEvent = {\n    runId: `${runDir}#${Date.now()}`,\n    runDir,\n    timestamp: new Date().toISOString(),\n    status: 'finished',\n    labels: opts.hostedLabels ?? {},\n    baseline: generations[0],\n    generations,\n    gateDecision: summary.gateDecision,\n    holdoutLift: summary.lift,\n    totalCostUsd: summary.totalCostUsd,\n    totalDurationMs: summary.durationMs,\n    insightReport: summary.insight,\n  }\n\n  await client.ingestEvalRun(event)\n}\n\nfunction averageComposite(\n  campaign: RunImprovementLoopResult<unknown, Scenario>['baselineCampaign'],\n): number {\n  const aggs = Object.values(campaign.aggregates.byScenario)\n  return aggs.length === 0 ? 0 : aggs.reduce((s, a) => s + a.meanComposite, 0) / aggs.length\n}\n\nfunction hashString(s: string): string {\n  let h = 2166136261 >>> 0\n  for (let i = 0; i < s.length; i++) {\n    h ^= s.charCodeAt(i)\n    h = Math.imul(h, 16777619) >>> 0\n  }\n  return h.toString(16).padStart(8, '0')\n}\n\n/**\n * Adapt campaign cells into the `RunRecord` shape `analyzeRuns()` consumes.\n * Each cell becomes one run; `candidateId` is the caller-supplied label so\n * baseline + winner pair cleanly on `(experimentId, seed)`.\n */\nfunction cellsToRunRecords<TArtifact>(\n  cells: ReadonlyArray<CampaignCellResult<TArtifact>>,\n  candidateId: 'baseline' | 'winner',\n  runId: string,\n): RunRecord[] {\n  return cells.map((cell) => {\n    const perJudge: Record<string, Record<string, number>> = {}\n    const perDimMeanAccum: Record<string, { sum: number; n: number }> = {}\n    let compositeSum = 0\n    let compositeCount = 0\n    for (const [judgeId, score] of Object.entries(cell.judgeScores)) {\n      perJudge[judgeId] = { ...score.dimensions }\n      for (const [dim, value] of Object.entries(score.dimensions)) {\n        if (!Number.isFinite(value)) continue\n        const accum = perDimMeanAccum[dim] ?? { sum: 0, n: 0 }\n        accum.sum += value\n        accum.n += 1\n        perDimMeanAccum[dim] = accum\n      }\n      if (Number.isFinite(score.composite)) {\n        compositeSum += score.composite\n        compositeCount += 1\n      }\n    }\n    const perDimMean: Record<string, number> = {}\n    for (const [dim, { sum, n }] of Object.entries(perDimMeanAccum)) {\n      perDimMean[dim] = n === 0 ? 0 : sum / n\n    }\n    const composite = compositeCount === 0 ? 0 : compositeSum / compositeCount\n    const judgeScores: JudgeScoresRecord = {\n      perJudge,\n      perDimMean,\n      composite,\n    }\n    return {\n      runId: `${runId}::${candidateId}::${cell.cellId}`,\n      experimentId: runId,\n      candidateId,\n      // Pair on (scenarioId, rep) — analyzeRuns pairs on (experimentId, seed).\n      // Synthesize a stable seed for that pairing.\n      seed:\n        cell.rep * 1_000_000 +\n        hashString(cell.scenarioId)\n          .slice(0, 6)\n          .split('')\n          .reduce((a, c) => (a * 31 + c.charCodeAt(0)) >>> 0, 0),\n      model: 'campaign-cell',\n      promptHash: 'sha256:cell',\n      configHash: 'sha256:cell',\n      commitSha: 'cell',\n      wallMs: cell.durationMs,\n      costUsd: cell.costUsd,\n      tokenUsage: { input: 0, output: 0 },\n      outcome: {\n        holdoutScore: composite,\n        raw: {},\n        judgeScores,\n      },\n      splitTag: 'holdout',\n      ...(cell.error ? { failureMode: cell.error } : {}),\n    } satisfies RunRecord\n  })\n}\n","/**\n * # `intake/feedback-table` — multi-rater approve/reject corpus → `RunRecord[]`.\n *\n * The generic shape behind Obsidian's `#approved` / `#rejected` tags, a\n * Google Sheet, a Postgres `feedback` table, or any CSV with ratings.\n *\n * Caller supplies one row per (run, rater) tuple plus per-run metadata; the\n * adapter rolls them up into the substrate-canonical `RunRecord` shape so\n * `analyzeRuns({ runs, raterScores })` can produce inter-rater agreement,\n * disagreement triage, and downstream recommendations.\n *\n * Per-run `RunRecord.outcome.searchScore` is the rater-mean rating\n * (normalised to 0..1 when scale is supplied); `outcome.raw` carries the\n * per-rater scores keyed by rater id for downstream attribution.\n */\n\nimport type { JudgeScoresRecord, RunOutcome, RunRecord, RunSplitTag } from '../../run-record'\n\nexport interface FeedbackTableRow {\n  /** Stable id for this run — the unit a rater scored. Drives pairing\n   *  across analysis primitives. */\n  runId: string\n  /** Identifier of the rater that produced this rating. */\n  rater: string\n  /** The rating itself. Accepts boolean (approve/reject), 0..1 scalar,\n   *  or any numeric scale — see `scale`. */\n  rating: number | boolean\n  /** Optional metadata carried through to `RunRecord.outcome.raw` and the\n   *  custom-shape metadata bag. */\n  metadata?: Record<string, unknown>\n}\n\nexport interface FeedbackTableMeta {\n  runId: string\n  /** When omitted, defaults to `'feedback-corpus'`. Used to group related\n   *  runs in `analyzeRuns()` lift analysis. */\n  experimentId?: string\n  /** When omitted, defaults to `runId` — each run is its own candidate. */\n  candidateId?: string\n  /** Cost in USD, when available. Set to 0 when unknown — the consumer's\n   *  cost analysis sections will collapse gracefully. */\n  costUsd?: number\n  /** Wall-clock ms, when available. Defaults to 0. */\n  wallMs?: number\n  /** Model identifier including snapshot. Default `unknown@unknown`. */\n  model?: string\n  /** Optional sha256 of the prompt; default `'sha256:unknown'`. */\n  promptHash?: string\n  /** Default `'sha256:unknown'`. */\n  configHash?: string\n  /** Default `'unknown'`. */\n  commitSha?: string\n  /** Default `'holdout'` — feedback corpora are by nature the holdout\n   *  signal a closed-loop improvement aims at. */\n  splitTag?: RunSplitTag\n  /** Free-form metadata available to consumers via the cast-out path on\n   *  the resulting RunRecord. */\n  extras?: Record<string, unknown>\n}\n\nexport interface FromFeedbackTableOptions {\n  /** Per-(run, rater) ratings. */\n  ratings: FeedbackTableRow[]\n  /** Per-run metadata. When a runId appears in `ratings` but not here, the\n   *  adapter synthesises minimal metadata with defaults documented above. */\n  meta?: FeedbackTableMeta[]\n  /** Rating scale. Provide `{ min, max }` for non-0..1 numeric scales.\n   *  Booleans are normalised: true → 1, false → 0. Default: assumes\n   *  ratings are already 0..1. */\n  scale?: { min: number; max: number }\n  /** When true, the rater scores are emitted into `raterScores` (a sibling\n   *  array `analyzeRuns()` accepts) instead of being averaged into the\n   *  run's `outcome.searchScore`. Default `true` — preserves rater-level\n   *  signal for inter-rater analysis. */\n  emitRaterScores?: boolean\n}\n\nexport interface FromFeedbackTableResult {\n  runs: RunRecord[]\n  /** Rater-level scores ready to pass into `analyzeRuns({ raterScores })`\n   *  for inter-rater agreement + disagreement triage. */\n  raterScores: Array<{ runId: string; rater: string; score: number }>\n}\n\nexport function fromFeedbackTable(opts: FromFeedbackTableOptions): FromFeedbackTableResult {\n  const { ratings, meta = [], scale, emitRaterScores = true } = opts\n  const metaByRun = new Map(meta.map((m) => [m.runId, m]))\n\n  // Normalise per-rating to a 0..1 score.\n  const normalise = (rating: number | boolean): number => {\n    if (typeof rating === 'boolean') return rating ? 1 : 0\n    if (!Number.isFinite(rating)) return Number.NaN\n    if (!scale) return rating\n    const { min, max } = scale\n    if (max === min) return rating\n    return (rating - min) / (max - min)\n  }\n\n  // Group ratings by runId.\n  const byRun = new Map<string, FeedbackTableRow[]>()\n  for (const row of ratings) {\n    const list = byRun.get(row.runId) ?? []\n    list.push(row)\n    byRun.set(row.runId, list)\n  }\n\n  const runs: RunRecord[] = []\n  const raterScores: FromFeedbackTableResult['raterScores'] = []\n\n  for (const [runId, rowsForRun] of byRun) {\n    const normalised = rowsForRun\n      .map((r) => ({ rater: r.rater, score: normalise(r.rating) }))\n      .filter((r) => Number.isFinite(r.score))\n    if (normalised.length === 0) continue\n\n    const meanScore = normalised.reduce((s, r) => s + r.score, 0) / normalised.length\n\n    const runMeta = metaByRun.get(runId) ?? ({ runId } as FeedbackTableMeta)\n\n    const judgeScores: JudgeScoresRecord = {\n      perJudge: Object.fromEntries(normalised.map((r) => [r.rater, { rating: r.score }])),\n      perDimMean: { rating: meanScore },\n      composite: meanScore,\n    }\n\n    const outcome: RunOutcome = {\n      // Feedback corpora ARE the holdout signal — score lands on\n      // `holdoutScore` so downstream substrate primitives (`paretoChart`,\n      // promotion gates) read it correctly by default.\n      holdoutScore: meanScore,\n      raw: Object.fromEntries(normalised.map((r) => [`rater:${r.rater}`, r.score])),\n      judgeScores,\n    }\n\n    runs.push({\n      runId,\n      experimentId: runMeta.experimentId ?? 'feedback-corpus',\n      candidateId: runMeta.candidateId ?? runId,\n      seed: 0,\n      model: runMeta.model ?? 'unknown@unknown',\n      promptHash: runMeta.promptHash ?? 'sha256:unknown',\n      configHash: runMeta.configHash ?? 'sha256:unknown',\n      commitSha: runMeta.commitSha ?? 'unknown',\n      wallMs: runMeta.wallMs ?? 0,\n      costUsd: runMeta.costUsd ?? 0,\n      tokenUsage: { input: 0, output: 0 },\n      outcome,\n      splitTag: runMeta.splitTag ?? 'holdout',\n    } as RunRecord)\n\n    if (emitRaterScores) {\n      for (const r of normalised) raterScores.push({ runId, rater: r.rater, score: r.score })\n    }\n  }\n\n  return { runs, raterScores }\n}\n","/**\n * # `intake/otel-spans` — OTel `TraceSpanEvent[]` → `RunRecord[]`.\n *\n * Turns an existing observability stream into the substrate-canonical\n * `RunRecord` shape so consumers with logs but no eval discipline can\n * call `analyzeRuns()` against their production traffic immediately.\n *\n * Pivot rule: spans are grouped by `tangle.runId` (the same attribute the\n * hosted-tier wire format uses) or, when absent, by `traceId`. One group\n * becomes one `RunRecord`. The root span (no `parentSpanId`) supplies:\n *\n *   - `runId` (the group key)\n *   - `wallMs` from `endTimeUnixNano - startTimeUnixNano`\n *   - `model` from `gen_ai.request.model` / `llm.model` / `tangle.model`\n *   - cost from `cost.usd` / `gen_ai.usage.cost_usd` / `tangle.cost.usd`\n *   - token usage from `gen_ai.usage.{input,output}_tokens`\n *   - `outcome.searchScore` from `tangle.score` / `eval.score` when\n *     present; `outcome.raw` collects every numeric attribute.\n *\n * Spans that ERRORed (`status.code === 'ERROR'`) populate `failureMode`\n * with their `name` so `analyzeRuns()`'s failure clustering sees them.\n */\n\nimport type { TraceSpanEvent } from '../../hosted/types'\nimport type {\n  JudgeScoresRecord,\n  RunOutcome,\n  RunRecord,\n  RunSplitTag,\n  RunTokenUsage,\n} from '../../run-record'\n\nconst SCORE_KEYS = ['tangle.score', 'eval.score', 'score']\nconst MODEL_KEYS = ['tangle.model', 'gen_ai.request.model', 'llm.model', 'model']\nconst COST_KEYS = ['tangle.cost.usd', 'gen_ai.usage.cost_usd', 'cost.usd', 'cost']\nconst INPUT_TOKEN_KEYS = ['gen_ai.usage.input_tokens', 'tangle.tokens.in', 'tokens.in']\nconst OUTPUT_TOKEN_KEYS = ['gen_ai.usage.output_tokens', 'tangle.tokens.out', 'tokens.out']\nconst PROMPT_HASH_KEYS = ['tangle.prompt_hash', 'prompt.hash']\nconst CONFIG_HASH_KEYS = ['tangle.config_hash', 'config.hash']\n\nexport interface FromOtelSpansOptions {\n  spans: TraceSpanEvent[]\n  /** Default split tag for synthesized records. Defaults to `'holdout'`. */\n  defaultSplit?: RunSplitTag\n  /** Default `experimentId` when not present on any span. */\n  experimentId?: string\n}\n\nexport function fromOtelSpans(opts: FromOtelSpansOptions): RunRecord[] {\n  const { spans, defaultSplit = 'holdout', experimentId = 'otel-corpus' } = opts\n  const grouped = groupSpans(spans)\n\n  const runs: RunRecord[] = []\n  for (const [groupKey, groupSpans] of grouped) {\n    const root = findRoot(groupSpans)\n    if (!root) continue\n\n    const wallMs = Math.max(0, (root.endTimeUnixNano - root.startTimeUnixNano) / 1_000_000)\n    const model = readAttrString(groupSpans, MODEL_KEYS) ?? 'unknown@unknown'\n    const costUsd = readAttrNumber(groupSpans, COST_KEYS) ?? 0\n    const inputTokens = readAttrNumber(groupSpans, INPUT_TOKEN_KEYS) ?? 0\n    const outputTokens = readAttrNumber(groupSpans, OUTPUT_TOKEN_KEYS) ?? 0\n    const promptHash = readAttrString(groupSpans, PROMPT_HASH_KEYS) ?? 'sha256:unknown'\n    const configHash = readAttrString(groupSpans, CONFIG_HASH_KEYS) ?? 'sha256:unknown'\n    const score = readAttrNumber(groupSpans, SCORE_KEYS)\n\n    const rawNumeric = collectNumericAttrs(groupSpans)\n    const tokenUsage: RunTokenUsage = {\n      input: inputTokens,\n      output: outputTokens,\n    }\n\n    const judgeScores: JudgeScoresRecord | undefined =\n      score !== undefined\n        ? {\n            perJudge: { 'otel-derived': { score } },\n            perDimMean: { score },\n            composite: score,\n          }\n        : undefined\n\n    const errorSpan = groupSpans.find((s) => s.status?.code === 'ERROR')\n    const outcome: RunOutcome = {\n      ...(opts.defaultSplit === 'search' ? { searchScore: score } : { holdoutScore: score }),\n      raw: rawNumeric,\n      ...(judgeScores ? { judgeScores } : {}),\n    }\n\n    runs.push({\n      runId: groupKey,\n      experimentId,\n      candidateId: (root.attributes['tangle.candidateId'] as string | undefined) ?? 'otel-default',\n      seed: 0,\n      model,\n      promptHash,\n      configHash,\n      commitSha: (root.attributes['tangle.commit_sha'] as string | undefined) ?? 'unknown',\n      wallMs,\n      costUsd,\n      tokenUsage,\n      outcome,\n      splitTag: defaultSplit,\n      ...(errorSpan ? { failureMode: errorSpan.name } : {}),\n    } as RunRecord)\n  }\n  return runs\n}\n\n// ── Internal helpers ────────────────────────────────────────────────\n\nfunction groupSpans(spans: TraceSpanEvent[]): Map<string, TraceSpanEvent[]> {\n  const m = new Map<string, TraceSpanEvent[]>()\n  for (const span of spans) {\n    const key = (span['tangle.runId'] as string | undefined) ?? span.traceId\n    const list = m.get(key) ?? []\n    list.push(span)\n    m.set(key, list)\n  }\n  return m\n}\n\nfunction findRoot(group: TraceSpanEvent[]): TraceSpanEvent | undefined {\n  return group.find((s) => !s.parentSpanId) ?? group[0]\n}\n\nfunction readAttrString(spans: TraceSpanEvent[], keys: string[]): string | undefined {\n  for (const span of spans) {\n    for (const key of keys) {\n      const v = span.attributes[key]\n      if (typeof v === 'string' && v.length > 0) return v\n    }\n  }\n  return undefined\n}\n\nfunction readAttrNumber(spans: TraceSpanEvent[], keys: string[]): number | undefined {\n  for (const span of spans) {\n    for (const key of keys) {\n      const v = span.attributes[key]\n      if (typeof v === 'number' && Number.isFinite(v)) return v\n      if (typeof v === 'string') {\n        const parsed = Number(v)\n        if (Number.isFinite(parsed)) return parsed\n      }\n    }\n  }\n  return undefined\n}\n\nfunction collectNumericAttrs(spans: TraceSpanEvent[]): Record<string, number> {\n  const raw: Record<string, number> = {}\n  for (const span of spans) {\n    for (const [k, v] of Object.entries(span.attributes)) {\n      if (typeof v === 'number' && Number.isFinite(v)) raw[k] = v\n    }\n  }\n  return raw\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AA+EA,eAAsB,YAAY,MAAkD;AAClF,QAAM,OAAO,KAAK;AAClB,QAAM,OAAO,KAAK,iBAAiB;AACnC,QAAM,YAAY,KAAK,qBAAqB;AAC5C,QAAM,QAAQ,aAAa,MAAM,KAAK,SAAS,MAAM;AAErD,QAAM,mBAAmB,KACtB,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,OAAO,OAAO,YAAY,GAAG,KAAK,EAAE,EAAE,EAC7D,OAAO,CAAC,MAAM,OAAO,SAAS,EAAE,KAAK,CAAC;AACzC,QAAM,YAAY;AAAA,IAChB,iBAAiB,IAAI,CAAC,MAAM,EAAE,KAAK;AAAA,IACnC;AAAA,IACA;AAAA,EACF;AAEA,QAAM,eAAe,oBAAoB,MAAM,IAAI;AAEnD,QAAM,QAAQ,KAAK,IAAI,CAAC,MAAM,EAAE,OAAO,EAAE,OAAO,OAAO,QAAQ;AAC/D,QAAM,WAAW,eAAe,OAAO,IAAI;AAC3C,QAAM,SAAS,YAAY,MAAM,EAAE,MAAM,CAAC;AAC1C,QAAM,WAA+C,CAAC;AACtD,MAAI,MAAM,WAAW,KAAK,MAAM,MAAM,CAAC,MAAM,MAAM,CAAC,GAAG;AACrD,aAAS,OAAO;AAAA,EAClB;AACA,MAAI,OAAO,OAAO,SAAS,GAAG;AAC5B,aAAS,SACP,OAAO,OAAO,WAAW,IACrB,4CACA;AAAA,EACR;AACA,QAAM,cAAc;AAAA,IAClB,MAAM;AAAA,IACN;AAAA,IACA,GAAI,SAAS,QAAQ,SAAS,SAAS,EAAE,SAAS,IAAI,CAAC;AAAA,EACzD;AAEA,QAAM,SAAS,qBAAqB,IAAI;AAExC,QAAM,aAAa,KAAK,cAAc,kBAAkB,KAAK,WAAW,IAAI;AAE5E,QAAM,OAAO,YAAY,MAAM,KAAK,qBAAqB,KAAK,sBAAsB,KAAK;AAEzF,QAAM,kBAAkB,KAAK,UACzB,MAAM,uBAAuB,MAAM,KAAK,SAAS,KAAK,IACtD;AAEJ,QAAM,gBAAgB,KAAK,kBACvB,qBAAqB,MAAM,KAAK,eAAe,IAC/C;AAEJ,QAAM,qBAAqB,KAAK,gBAC5B,0BAA0B,MAAM,KAAK,eAAe,KAAK,IACzD;AAEJ,QAAM,UAAU,sBAAsB,WAAW,MAAM,aAAa;AAEpE,QAAM,kBAAkB,qBAAqB;AAAA,IAC3C;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF,CAAC;AAED,SAAO;AAAA,IACL,GAAG,KAAK;AAAA,IACR;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AACF;AAIA,SAAS,aACP,MACA,MACsB;AACtB,MAAI,SAAS,OAAQ,QAAO;AAC5B,QAAM,aAAa,KAAK,KAAK,CAAC,MAAM,OAAO,SAAS,EAAE,QAAQ,YAAY,CAAC;AAC3E,SAAO,aAAa,YAAY;AAClC;AAEA,SAAS,YAAY,KAAgB,OAAqC;AACxE,QAAM,UAAU,UAAU,YAAY,IAAI,QAAQ,eAAe,IAAI,QAAQ;AAC7E,MAAI,OAAO,SAAS,OAAO,EAAG,QAAO;AAIrC,QAAM,MAAM,UAAU,YAAY,IAAI,QAAQ,cAAc,IAAI,QAAQ;AACxE,SAAO,OAAO,SAAS,GAAG,IAAK,MAAiB,OAAO;AACzD;AAIA,SAAS,eACP,QACA,MACA,SACoB;AACpB,MAAI,OAAO,WAAW,GAAG;AACvB,WAAO;AAAA,MACL,GAAG;AAAA,MACH,MAAM;AAAA,MACN,KAAK;AAAA,MACL,KAAK;AAAA,MACL,QAAQ;AAAA,MACR,KAAK;AAAA,MACL,KAAK;AAAA,MACL,WAAW,CAAC;AAAA,IACd;AAAA,EACF;AACA,QAAM,SAAS,CAAC,GAAG,MAAM,EAAE,KAAK,CAAC,GAAG,MAAM,IAAI,CAAC;AAC/C,QAAM,IAAI,OAAO;AACjB,QAAMA,QAAO,OAAO,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI;AACjD,QAAM,WAAW,OAAO,OAAO,CAAC,GAAG,MAAM,KAAK,IAAIA,UAAS,GAAG,CAAC,IAAI;AACnE,QAAM,SAAS,KAAK,KAAK,QAAQ;AACjC,QAAM,WAAW,UACb,CAAC,GAAG,OAAO,EAAE,KAAK,CAAC,GAAG,MAAM,EAAE,QAAQ,EAAE,KAAK,EAAE,MAAM,GAAG,KAAK,IAAI,GAAG,QAAQ,MAAM,CAAC,IACnF;AACJ,SAAO;AAAA,IACL;AAAA,IACA,MAAAA;AAAA,IACA,KAAK,WAAW,QAAQ,GAAG;AAAA,IAC3B,KAAK,WAAW,QAAQ,IAAI;AAAA,IAC5B;AAAA,IACA,KAAK,OAAO,CAAC;AAAA,IACb,KAAK,OAAO,IAAI,CAAC;AAAA,IACjB,WAAW,UAAU,QAAQ,IAAI;AAAA,IACjC,GAAI,WAAW,EAAE,SAAS,IAAI,CAAC;AAAA,EACjC;AACF;AAEA,SAAS,WAAW,QAAkB,GAAmB;AACvD,MAAI,OAAO,WAAW,EAAG,QAAO;AAChC,MAAI,OAAO,WAAW,EAAG,QAAO,OAAO,CAAC;AACxC,QAAM,OAAO,OAAO,SAAS,KAAK;AAClC,QAAM,KAAK,KAAK,MAAM,GAAG;AACzB,QAAM,KAAK,KAAK,KAAK,GAAG;AACxB,MAAI,OAAO,GAAI,QAAO,OAAO,EAAE;AAC/B,QAAM,IAAI,MAAM;AAChB,SAAO,OAAO,EAAE,KAAM,IAAI,KAAK,OAAO,EAAE,IAAK;AAC/C;AAKA,SAAS,UAAU,QAAkB,MAA+C;AAClF,MAAI,OAAO,WAAW,KAAK,OAAO,EAAG,QAAO,CAAC;AAC7C,QAAM,MAAM,OAAO,CAAC;AACpB,QAAM,MAAM,OAAO,OAAO,SAAS,CAAC;AACpC,MAAI,QAAQ,IAAK,QAAO,CAAC,EAAE,IAAI,KAAK,IAAI,KAAK,OAAO,OAAO,OAAO,CAAC;AACnE,QAAM,SAAS,MAAM,OAAO;AAC5B,QAAM,MAAuC,CAAC;AAC9C,WAAS,IAAI,GAAG,IAAI,MAAM,KAAK;AAC7B,UAAM,KAAK,MAAM,IAAI;AACrB,UAAM,KAAK,MAAM,OAAO,IAAI,MAAM,KAAK;AACvC,QAAI,KAAK,EAAE,IAAI,IAAI,OAAO,EAAE,CAAC;AAAA,EAC/B;AACA,aAAW,KAAK,QAAQ;AACtB,UAAM,MAAM,KAAK,IAAI,OAAO,GAAG,KAAK,OAAO,IAAI,OAAO,KAAK,CAAC;AAC5D,QAAI,GAAG,EAAG;AAAA,EACZ;AACA,SAAO;AACT;AAEA,SAAS,oBAAoB,MAAmB,MAAkD;AAKhG,QAAM,QAAQ,oBAAI,IAAsB;AACxC,aAAW,OAAO,MAAM;AACtB,UAAM,SAAS,IAAI,QAAQ;AAC3B,QAAI,CAAC,OAAQ;AACb,eAAW,CAAC,KAAK,KAAK,KAAK,OAAO,QAAQ,OAAO,cAAc,CAAC,CAAC,GAAG;AAClE,UAAI,CAAC,OAAO,SAAS,KAAK,EAAG;AAC7B,YAAM,MAAM,MAAM,IAAI,GAAG,KAAK,CAAC;AAC/B,UAAI,KAAK,KAAK;AACd,YAAM,IAAI,KAAK,GAAG;AAAA,IACpB;AAAA,EACF;AACA,QAAM,MAA0C,CAAC;AACjD,aAAW,CAAC,KAAK,MAAM,KAAK,MAAO,KAAI,GAAG,IAAI,eAAe,QAAQ,IAAI;AACzE,SAAO;AACT;AAIA,SAAS,qBAAqB,MAAiD;AAI7E,QAAM,MAAoC,CAAC;AAC3C,QAAM,UAAU,oBAAI,IAAsB;AAC1C,aAAW,OAAO,MAAM;AACtB,UAAM,SAAS,IAAI,QAAQ;AAC3B,QAAI,CAAC,QAAQ,SAAU;AACvB,eAAW,CAAC,SAAS,IAAI,KAAK,OAAO,QAAQ,OAAO,QAAQ,GAAG;AAC7D,YAAM,YAAY,OAAO,OAAO,IAAI,EAAE,OAAO,OAAO,QAAQ;AAC5D,UAAI,UAAU,WAAW,EAAG;AAC5B,YAAM,YAAY,UAAU,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,UAAU;AACnE,YAAM,MAAM,QAAQ,IAAI,OAAO,KAAK,CAAC;AACrC,UAAI,KAAK,SAAS;AAClB,cAAQ,IAAI,SAAS,GAAG;AAAA,IAC1B;AAAA,EACF;AACA,aAAW,CAAC,SAAS,MAAM,KAAK,SAAS;AACvC,QAAI,OAAO,IAAI;AAAA,MACb,GAAG,OAAO;AAAA,MACV,WAAW,OAAO,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,OAAO;AAAA,IACxD;AAAA,EACF;AACA,SAAO;AACT;AAIA,SAAS,kBACP,SAC+B;AAC/B,QAAM,QAAQ,oBAAI,IAAqD;AACvE,aAAW,KAAK,SAAS;AACvB,QAAI,CAAC,OAAO,SAAS,EAAE,KAAK,EAAG;AAC/B,UAAM,OAAO,MAAM,IAAI,EAAE,KAAK,KAAK,CAAC;AACpC,SAAK,KAAK,EAAE,OAAO,EAAE,OAAO,OAAO,EAAE,MAAM,CAAC;AAC5C,UAAM,IAAI,EAAE,OAAO,IAAI;AAAA,EACzB;AACA,QAAM,SAAS,IAAI,IAAI,QAAQ,IAAI,CAAC,MAAM,EAAE,KAAK,CAAC;AAClD,QAAM,eAAyB,CAAC;AAChC,aAAW,CAAC,OAAO,YAAY,KAAK,OAAO;AACzC,UAAM,OAAO,IAAI,IAAI,aAAa,IAAI,CAAC,MAAM,EAAE,KAAK,CAAC;AACrD,QAAI,MAAM;AACV,eAAW,KAAK,OAAQ,KAAI,CAAC,KAAK,IAAI,CAAC,EAAG,OAAM;AAChD,QAAI,IAAK,cAAa,KAAK,KAAK;AAAA,EAClC;AACA,MAAI,OAAO,OAAO,KAAK,aAAa,WAAW,EAAG,QAAO;AAEzD,QAAM,YAAY,CAAC,GAAG,MAAM,EAAE,KAAK;AACnC,QAAM,UAAkC,CAAC;AACzC,WAAS,IAAI,GAAG,IAAI,UAAU,QAAQ,KAAK;AACzC,aAAS,IAAI,IAAI,GAAG,IAAI,UAAU,QAAQ,KAAK;AAC7C,YAAM,IAAI,UAAU,CAAC;AACrB,YAAM,IAAI,UAAU,CAAC;AACrB,YAAM,UAAoB,CAAC;AAC3B,YAAM,UAAoB,CAAC;AAC3B,iBAAW,SAAS,cAAc;AAChC,cAAM,eAAe,MAAM,IAAI,KAAK;AACpC,cAAM,KAAK,aAAa,KAAK,CAAC,MAAM,EAAE,UAAU,CAAC,GAAG;AACpD,cAAM,KAAK,aAAa,KAAK,CAAC,MAAM,EAAE,UAAU,CAAC,GAAG;AACpD,YAAI,OAAO,UAAa,OAAO,QAAW;AACxC,kBAAQ,KAAK,EAAE;AACf,kBAAQ,KAAK,EAAE;AAAA,QACjB;AAAA,MACF;AACA,cAAQ,GAAG,CAAC,KAAK,CAAC,EAAE,IAAI,QAAQ,SAAS,OAAO;AAAA,IAClD;AAAA,EACF;AACA,QAAM,aAAa,OAAO,OAAO,OAAO;AACxC,QAAM,QACJ,WAAW,WAAW,IAAI,IAAI,WAAW,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,WAAW;AAEnF,QAAM,oBAAoB,aACvB,IAAI,CAAC,UAAU;AACd,UAAM,eAAe,MAAM,IAAI,KAAK;AACpC,UAAM,SAAS,aAAa,IAAI,CAAC,MAAM,EAAE,KAAK;AAC9C,UAAM,QAAQ,KAAK,IAAI,GAAG,MAAM,IAAI,KAAK,IAAI,GAAG,MAAM;AACtD,WAAO,EAAE,OAAO,SAAS,cAAc,MAAM;AAAA,EAC/C,CAAC,EACA,KAAK,CAAC,GAAG,MAAM,EAAE,QAAQ,EAAE,KAAK,EAChC,MAAM,GAAG,EAAE;AAEd,SAAO;AAAA,IACL,QAAQ,OAAO;AAAA,IACf,cAAc,aAAa;AAAA,IAC3B;AAAA,IACA;AAAA,IACA;AAAA,EACF;AACF;AAEA,SAAS,QAAQ,GAAa,GAAqB;AACjD,MAAI,EAAE,WAAW,EAAE,UAAU,EAAE,WAAW,EAAG,QAAO;AACpD,QAAM,IAAI,EAAE;AACZ,QAAM,QAAQ,EAAE,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI;AAC7C,QAAM,QAAQ,EAAE,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI;AAC7C,MAAI,MAAM;AACV,MAAI,SAAS;AACb,MAAI,SAAS;AACb,WAAS,IAAI,GAAG,IAAI,GAAG,KAAK;AAC1B,UAAM,KAAK,EAAE,CAAC,IAAK;AACnB,UAAM,KAAK,EAAE,CAAC,IAAK;AACnB,WAAO,KAAK;AACZ,cAAU,KAAK;AACf,cAAU,KAAK;AAAA,EACjB;AACA,QAAM,QAAQ,KAAK,KAAK,SAAS,MAAM;AACvC,SAAO,UAAU,IAAI,IAAI,MAAM;AACjC;AAIA,SAAS,YACP,MACA,YACA,aACA,OACyB;AACzB,MAAI,MAAM;AACV,MAAI,MAAM;AACV,MAAI,CAAC,OAAO,CAAC,KAAK;AAGhB,UAAM,MAAM,CAAC,GAAG,IAAI,IAAI,KAAK,IAAI,CAAC,MAAM,EAAE,WAAW,CAAC,CAAC;AACvD,QAAI,IAAI,WAAW,EAAG,QAAO;AAC7B,UAAM,CAAC,KAAK,GAAG,IAAI;AACnB,UAAM,QAAQ,KAAK,KAAK,OAAO,CAAC,MAAM,EAAE,gBAAgB,GAAG,EAAE,IAAI,CAAC,MAAM,YAAY,GAAG,KAAK,CAAC,CAAC;AAC9F,UAAM,QAAQ,KAAK,KAAK,OAAO,CAAC,MAAM,EAAE,gBAAgB,GAAG,EAAE,IAAI,CAAC,MAAM,YAAY,GAAG,KAAK,CAAC,CAAC;AAC9F,UAAM,SAAS,QAAQ,MAAM;AAC7B,UAAM,SAAS,QAAQ,MAAM;AAAA,EAC/B;AAEA,QAAM,WAAW,KAAK,OAAO,CAAC,MAAM,EAAE,gBAAgB,GAAG;AACzD,QAAM,YAAY,KAAK,OAAO,CAAC,MAAM,EAAE,gBAAgB,GAAG;AAC1D,MAAI,SAAS,WAAW,KAAK,UAAU,WAAW,EAAG,QAAO;AAI5D,QAAM,gBAAgB,IAAI,IAAI,SAAS,IAAI,CAAC,MAAM,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC;AACrE,QAAM,iBAA2B,CAAC;AAClC,QAAM,kBAA4B,CAAC;AACnC,MAAI,iBAAiB;AACrB,aAAW,QAAQ,WAAW;AAC5B,UAAM,IAAI,cAAc,IAAI,WAAW,IAAI,CAAC;AAC5C,QAAI,GAAG;AACL,YAAM,KAAK,YAAY,GAAG,KAAK;AAC/B,YAAM,KAAK,YAAY,MAAM,KAAK;AAClC,UAAI,OAAO,SAAS,EAAE,KAAK,OAAO,SAAS,EAAE,GAAG;AAC9C,uBAAe,KAAK,EAAE;AACtB,wBAAgB,KAAK,EAAE;AACvB,yBAAiB;AAAA,MACnB;AAAA,IACF;AAAA,EACF;AACA,MAAI,CAAC,gBAAgB;AACnB,UAAM,IAAI,KAAK,IAAI,SAAS,QAAQ,UAAU,MAAM;AACpD,aAAS,IAAI,GAAG,IAAI,GAAG,KAAK;AAC1B,YAAM,KAAK,YAAY,SAAS,CAAC,GAAI,KAAK;AAC1C,YAAM,KAAK,YAAY,UAAU,CAAC,GAAI,KAAK;AAC3C,UAAI,OAAO,SAAS,EAAE,KAAK,OAAO,SAAS,EAAE,GAAG;AAC9C,uBAAe,KAAK,EAAE;AACtB,wBAAgB,KAAK,EAAE;AAAA,MACzB;AAAA,IACF;AAAA,EACF;AACA,MAAI,eAAe,WAAW,EAAG,QAAO;AAExC,QAAM,eAAe,KAAK,cAAc;AACxC,QAAM,gBAAgB,KAAK,eAAe;AAC1C,QAAM,QAAQ,gBAAgB;AAE9B,QAAM,YAAY,gBAAgB,gBAAgB,iBAAiB;AAAA,IACjE,YAAY;AAAA,IACZ,WAAW;AAAA,IACX,WAAW;AAAA,EACb,CAAC;AACD,QAAM,QAAQ,YAAY,gBAAgB,eAAe;AACzD,QAAM,IAAI,QAAQ,gBAAgB,eAAe;AACjD,QAAM,MAAM,UAAU,EAAE,SAAS,eAAe,QAAQ,OAAO,KAAK,OAAO,KAAK,CAAC;AACjF,QAAM,YAAY,mBAAmB;AAAA,IACnC,QAAQ,KAAK,IAAI,KAAK,IAAI,KAAK,GAAG,IAAI;AAAA,IACtC,OAAO;AAAA,IACP,OAAO;AAAA,EACT,CAAC;AAED,SAAO;AAAA,IACL;AAAA,IACA;AAAA,IACA;AAAA,IACA,MAAM,CAAC,UAAU,KAAK,UAAU,IAAI;AAAA,IACpC,QAAQ,MAAM;AAAA,IACd,GAAG,eAAe;AAAA,IAClB,SAAS;AAAA,IACT;AAAA,IACA;AAAA,EACF;AACF;AAEA,SAAS,WAAW,GAAsB;AACxC,SAAO,GAAG,EAAE,YAAY,KAAK,EAAE,IAAI;AACrC;AAEA,SAAS,KAAK,KAAuB;AACnC,SAAO,IAAI,WAAW,IAAI,IAAI,IAAI,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,IAAI;AACrE;AAIA,eAAe,uBACb,MACA,SACA,OAC4C;AAC5C,QAAM,SAAS,KAAK,OAAO,CAAC,MAAM,YAAY,GAAG,KAAK,IAAI,OAAO,EAAE,gBAAgB,MAAS;AAC5F,MAAI,OAAO,WAAW,EAAG,QAAO,EAAE,UAAU,CAAC,GAAG,eAAe,EAAE;AAEjE,QAAM,WAAW,oBAAI,IAAoD;AACzE,aAAW,OAAO,QAAQ;AACxB,QAAI;AACF,YAAM,SAAS,MAAM,QAAQ,IAAI,IAAI,OAAO;AAAA,QAC1C,MAAM;AAAA,QACN;AAAA,MACF,CAAsC;AACtC,iBAAW,WAAW,OAAO,UAA8B;AACzD,cAAM,MAAM,QAAQ,QAAQ,QAAQ,cAAc;AAClD,cAAM,IAAI,SAAS,IAAI,GAAG,KAAK,EAAE,WAAW,CAAC,GAAG,OAAO,EAAE;AACzD,YAAI,EAAE,UAAU,SAAS,EAAG,GAAE,UAAU,KAAK,IAAI,KAAK;AACtD,iBAAS,IAAI,KAAK,CAAC;AAAA,MACrB;AAAA,IACF,QAAQ;AACN,YAAM,IAAI,SAAS,IAAI,eAAe,KAAK,EAAE,WAAW,CAAC,GAAG,OAAO,EAAE;AACrE,UAAI,EAAE,UAAU,SAAS,EAAG,GAAE,UAAU,KAAK,IAAI,KAAK;AACtD,eAAS,IAAI,iBAAiB,CAAC;AAAA,IACjC;AAAA,EACF;AACA,QAAM,cAAc,CAAC,GAAG,SAAS,QAAQ,CAAC,EAAE,IAAI,CAAC,CAAC,IAAI,CAAC,OAAO;AAAA,IAC5D;AAAA,IACA,MAAM;AAAA,IACN,OAAO,EAAE,UAAU,SAAS,OAAO;AAAA,IACnC,WAAW,EAAE;AAAA,EACf,EAAE;AACF,cAAY,KAAK,CAAC,GAAG,MAAM,EAAE,QAAQ,EAAE,KAAK;AAC5C,SAAO,EAAE,UAAU,aAAa,eAAe,OAAO,OAAO;AAC/D;AAIA,SAAS,qBACP,MACA,UACgC;AAChC,MAAI,QAAQ;AACZ,QAAM,UAAqE,CAAC;AAC5E,aAAW,OAAO,MAAM;AACtB,UAAM,SAAS,gBAAgB,GAAG;AAClC,QAAI,CAAC,OAAQ;AACb,UAAM,YAAY,cAAc,QAAQ,QAAQ;AAChD,eAAW,QAAQ,WAAW;AAC5B;AACA,cAAQ,KAAK,EAAE,OAAO,IAAI,OAAO,QAAQ,KAAK,QAAQ,SAAS,KAAK,SAAS,CAAC;AAAA,IAChF;AAAA,EACF;AACA,SAAO,EAAE,OAAO,oBAAoB,UAAU,GAAG,QAAQ;AAC3D;AAEA,SAAS,gBAAgB,KAAoC;AAO3D,QAAM,WAAY,IAA0D;AAC5E,MAAI,OAAO,UAAU,WAAW,SAAU,QAAO,SAAS;AAC1D,MAAI,OAAO,UAAU,SAAS,SAAU,QAAO,SAAS;AACxD,SAAO;AACT;AAIA,SAAS,0BACP,MACA,SACA,OACuC;AACvC,QAAM,KAAe,CAAC;AACtB,QAAM,KAAe,CAAC;AACtB,aAAW,OAAO,MAAM;AACtB,UAAM,IAAI,QAAQ,aAAa,IAAI,KAAK;AACxC,QAAI,MAAM,UAAa,CAAC,OAAO,SAAS,CAAC,EAAG;AAC5C,UAAM,IAAI,YAAY,KAAK,KAAK;AAChC,QAAI,CAAC,OAAO,SAAS,CAAC,EAAG;AACzB,OAAG,KAAK,CAAC;AACT,OAAG,KAAK,CAAC;AAAA,EACX;AACA,MAAI,GAAG,SAAS,EAAG,QAAO;AAE1B,QAAM,IAAI,QAAQ,IAAI,EAAE;AACxB,QAAM,IAAI,SAAS,IAAI,EAAE;AACzB,QAAM,QAAQ,KAAK,EAAE;AACrB,QAAM,QAAQ,KAAK,EAAE;AACrB,MAAI,MAAM;AACV,MAAI,QAAQ;AACZ,WAAS,IAAI,GAAG,IAAI,GAAG,QAAQ,KAAK;AAClC,YAAQ,GAAG,CAAC,IAAK,UAAU,GAAG,CAAC,IAAK;AACpC,cAAU,GAAG,CAAC,IAAK,UAAU;AAAA,EAC/B;AACA,QAAM,QAAQ,UAAU,IAAI,IAAI,MAAM;AACtC,QAAM,YAAY,QAAQ,QAAQ;AAClC,QAAM,QAAQ,GAAG,OAAO,CAAC,GAAG,MAAM,KAAK,IAAI,UAAU,GAAG,CAAC;AACzD,QAAM,QAAQ,GAAG,OAAO,CAAC,GAAG,GAAG,MAAM,KAAK,KAAK,YAAY,QAAQ,GAAG,CAAC,OAAQ,GAAG,CAAC;AACnF,QAAM,KAAK,UAAU,IAAI,IAAI,IAAI,QAAQ;AAEzC,SAAO;AAAA,IACL,QAAQ,QAAQ;AAAA,IAChB,GAAG,GAAG;AAAA,IACN,SAAS;AAAA,IACT,UAAU;AAAA,IACV,aAAa,EAAE,WAAW,OAAO,GAAG;AAAA,EACtC;AACF;AAEA,SAAS,SAAS,GAAa,GAAqB;AAClD,MAAI,EAAE,WAAW,EAAE,UAAU,EAAE,WAAW,EAAG,QAAO;AACpD,SAAO,QAAQ,KAAK,CAAC,GAAG,KAAK,CAAC,CAAC;AACjC;AAEA,SAAS,KAAK,KAAyB;AACrC,QAAM,UAAU,IAAI,IAAI,CAAC,GAAGC,QAAO,EAAE,GAAG,GAAAA,GAAE,EAAE;AAC5C,UAAQ,KAAK,CAAC,GAAG,MAAM,EAAE,IAAI,EAAE,CAAC;AAChC,QAAM,QAAQ,IAAI,MAAM,IAAI,MAAM,EAAE,KAAK,CAAC;AAC1C,MAAI,IAAI;AACR,SAAO,IAAI,QAAQ,QAAQ;AACzB,QAAI,IAAI;AACR,WAAO,IAAI,IAAI,QAAQ,UAAU,QAAQ,IAAI,CAAC,EAAG,MAAM,QAAQ,CAAC,EAAG,EAAG;AACtE,UAAM,OAAO,IAAI,IAAI,KAAK;AAC1B,aAAS,IAAI,GAAG,KAAK,GAAG,IAAK,OAAM,QAAQ,CAAC,EAAG,CAAC,IAAI;AACpD,QAAI,IAAI;AAAA,EACV;AACA,SAAO;AACT;AAIA,SAAS,sBACP,WACA,MACA,eAC0B;AAO1B,QAAM,OAAyC,CAAC;AAChD,QAAM,WACJ,SAAS,UAAa,KAAK,KAAK,CAAC,IAAI,IAChC,SACD,KAAK,QAAQ,IACV,SACA;AACT,OAAK,KAAK;AAAA,IACR,MAAM;AAAA,IACN,QAAQ;AAAA,IACR,QAAQ,OACJ,SAAS,KAAK,MAAM,QAAQ,CAAC,CAAC,WAAW,KAAK,KAAK,CAAC,EAAE,QAAQ,CAAC,CAAC,KAAK,KAAK,KAAK,CAAC,EAAE,QAAQ,CAAC,CAAC,QAAQ,KAAK,CAAC,KAC1G;AAAA,EACN,CAAC;AACD,QAAM,aACJ,kBAAkB,UAAa,cAAc,UAAU,IAAK,SAAoB;AAClF,OAAK,KAAK;AAAA,IACR,MAAM;AAAA,IACN,QAAQ;AAAA,IACR,QAAQ,gBAAgB,GAAG,cAAc,KAAK,oBAAoB;AAAA,EACpE,CAAC;AACD,OAAK,KAAK;AAAA,IACR,MAAM;AAAA,IACN,QAAQ,UAAU,QAAQ,MAAM,SAAS,UAAU,QAAQ,MAAM,SAAS;AAAA,IAC1E,QAAQ,QAAQ,UAAU,KAAK,QAAQ,CAAC,CAAC,SAAS,UAAU,IAAI,QAAQ,CAAC,CAAC,SAAS,UAAU,IAAI,QAAQ,CAAC,CAAC,WAAW,UAAU,CAAC;AAAA,EACnI,CAAC;AACD,QAAM,SAAS,KAAK,KAAK,CAAC,MAAM,EAAE,WAAW,MAAM,IAC/C,SACA,KAAK,KAAK,CAAC,MAAM,EAAE,WAAW,MAAM,IAClC,SACA;AACN,SAAO;AAAA,IACL;AAAA,IACA;AAAA,IACA,QAAQ,CAAC;AAAA,EACX;AACF;AAeA,SAAS,qBAAqB,KAA8C;AAC1E,QAAM,MAAwB,CAAC;AAK/B,MAAI,IAAI,UAAU,IAAI,GAAG;AACvB,QAAI,IAAI,UAAU,OAAO,KAAK;AAC5B,YAAM,OAAO,IAAI,UAAU,YAAY,CAAC;AACxC,YAAM,QAAQ,KACX,MAAM,GAAG,CAAC,EACV,IAAI,CAAC,MAAM,GAAG,EAAE,KAAK,IAAI,EAAE,MAAM,QAAQ,CAAC,CAAC,EAAE,EAC7C,KAAK,IAAI;AACZ,UAAI,KAAK;AAAA,QACP,UAAU;AAAA,QACV,MAAM;AAAA,QACN,OAAO,kBAAkB,IAAI,UAAU,KAAK,QAAQ,CAAC,CAAC;AAAA,QACtD,QACE,KAAK,SAAS,IACV,SAAS,KAAK,MAAM,OAAO,KAAK,WAAW,IAAI,KAAK,GAAG,sBAAsB,KAAK,mBAAmB,IAAI,UAAU,IAAI,QAAQ,CAAC,CAAC,SAAS,IAAI,UAAU,IAAI,QAAQ,CAAC,CAAC,MACtK,iBAAiB,IAAI,UAAU,IAAI,QAAQ,CAAC,CAAC,SAAS,IAAI,UAAU,IAAI,QAAQ,CAAC,CAAC;AAAA,QACxF,cAAc;AAAA,MAChB,CAAC;AAAA,IACH,WAAW,IAAI,UAAU,OAAO,KAAK;AACnC,YAAM,OAAO,IAAI,UAAU,YAAY,CAAC;AACxC,YAAM,QAAQ,KACX,MAAM,GAAG,CAAC,EACV,IAAI,CAAC,MAAM,GAAG,EAAE,KAAK,IAAI,EAAE,MAAM,QAAQ,CAAC,CAAC,EAAE,EAC7C,KAAK,IAAI;AACZ,UAAI,KAAK;AAAA,QACP,UAAU;AAAA,QACV,MAAM;AAAA,QACN,OAAO,kBAAkB,IAAI,UAAU,KAAK,QAAQ,CAAC,CAAC;AAAA,QACtD,QACE,KAAK,SAAS,IACV,SAAS,KAAK,MAAM,OAAO,KAAK,WAAW,IAAI,KAAK,GAAG,KAAK,KAAK,mBAAmB,IAAI,UAAU,IAAI,QAAQ,CAAC,CAAC,SAAS,IAAI,UAAU,IAAI,QAAQ,CAAC,CAAC,MACrJ,iBAAiB,IAAI,UAAU,IAAI,QAAQ,CAAC,CAAC,SAAS,IAAI,UAAU,IAAI,QAAQ,CAAC,CAAC;AAAA,QACxF,cAAc;AAAA,MAChB,CAAC;AAAA,IACH;AAAA,EACF;AAKA,MAAI,OAAO,KAAK,IAAI,MAAM,EAAE,WAAW,KAAK,IAAI,UAAU,IAAI,GAAG;AAC/D,QAAI,KAAK;AAAA,MACP,UAAU;AAAA,MACV,MAAM;AAAA,MACN,OAAO;AAAA,MACP,QACE;AAAA,MACF,cAAc;AAAA,IAChB,CAAC;AAAA,EACH;AAEA,MAAI,IAAI,MAAM;AACZ,UAAM,WAAW,IAAI,KAAK,KAAK,CAAC,IAAI,IAAI;AACxC,UAAM,eAAe,IAAI,KAAK,KAAK,CAAC,KAAK,IAAI,aAAa,IAAI,KAAK,KAAK,CAAC,IAAI,IAAI;AACjF,QAAI,UAAU;AACZ,UAAI,KAAK;AAAA,QACP,UAAU;AAAA,QACV,MAAM;AAAA,QACN,OAAO,oBAAe,IAAI,KAAK,MAAM,QAAQ,CAAC,CAAC,YAAY,IAAI,KAAK,KAAK,CAAC,EAAE,QAAQ,CAAC,CAAC,KAAK,IAAI,KAAK,KAAK,CAAC,EAAE,QAAQ,CAAC,CAAC;AAAA,QACtH,QAAQ,kCAAkC,IAAI,SAAS,qCAAqC,IAAI,KAAK,CAAC,OAAO,IAAI,KAAK,OAAO,QAAQ,CAAC,CAAC,OAAO,IAAI,KAAK,QAAQ,QAAQ,CAAC,CAAC;AAAA,QACzK,cAAc;AAAA,MAChB,CAAC;AAAA,IACH,WAAW,cAAc;AACvB,UAAI,KAAK;AAAA,QACP,UAAU;AAAA,QACV,MAAM;AAAA,QACN,OAAO,6BAAwB,IAAI,KAAK,SAAS,sBAAsB,IAAI,KAAK,CAAC;AAAA,QACjF,QAAQ,uDAAuD,IAAI,KAAK,IAAI,QAAQ,CAAC,CAAC,uBAAuB,IAAI,KAAK,MAAM,QAAQ,CAAC,CAAC;AAAA,QACtI,cAAc;AAAA,MAChB,CAAC;AAAA,IACH,OAAO;AACL,UAAI,KAAK;AAAA,QACP,UAAU;AAAA,QACV,MAAM;AAAA,QACN,OAAO,mCAA8B,IAAI,KAAK,KAAK,CAAC,EAAE,QAAQ,CAAC,CAAC,6BAA6B,IAAI,SAAS;AAAA,QAC1G,QAAQ;AAAA,QACR,cAAc;AAAA,MAChB,CAAC;AAAA,IACH;AAAA,EACF;AAEA,MAAI,IAAI,iBAAiB,IAAI,cAAc,QAAQ,GAAG;AACpD,QAAI,KAAK;AAAA,MACP,UAAU;AAAA,MACV,MAAM;AAAA,MACN,OAAO,GAAG,IAAI,cAAc,KAAK,eAAe,IAAI,cAAc,UAAU,IAAI,KAAK,GAAG;AAAA,MACxF,QAAQ;AAAA,MACR,cAAc;AAAA,IAChB,CAAC;AAAA,EACH;AAEA,MAAI,IAAI,cAAc,IAAI,WAAW,QAAQ,KAAK;AAChD,QAAI,KAAK;AAAA,MACP,UAAU;AAAA,MACV,MAAM;AAAA,MACN,OAAO,gCAA2B,IAAI,WAAW,MAAM,QAAQ,CAAC,CAAC;AAAA,MACjE,QAAQ;AAAA,MACR,cAAc;AAAA,IAChB,CAAC;AAAA,EACH;AAEA,MAAI,IAAI,mBAAmB,IAAI,gBAAgB,SAAS,SAAS,GAAG;AAClE,UAAM,MAAM,IAAI,gBAAgB,SAAS,CAAC;AAC1C,QAAI,KAAK;AAAA,MACP,UAAU;AAAA,MACV,MAAM;AAAA,MACN,OAAO,wBAAwB,IAAI,IAAI,MAAM,IAAI,QAAQ,KAAK,QAAQ,CAAC,CAAC;AAAA,MACxE,QAAQ,GAAG,IAAI,gBAAgB,aAAa,4CAA4C,IAAI,UAAU,MAAM,qBAAqB,IAAI,IAAI;AAAA,MACzI,cAAc;AAAA,IAChB,CAAC;AAAA,EACH;AAEA,MAAI,IAAI,sBAAsB,KAAK,IAAI,IAAI,mBAAmB,QAAQ,IAAI,KAAK;AAC7E,QAAI,KAAK;AAAA,MACP,UAAU;AAAA,MACV,MAAM;AAAA,MACN,OAAO,+BAA+B,IAAI,mBAAmB,MAAM,qBAAgB,IAAI,mBAAmB,SAAS,QAAQ,CAAC,CAAC;AAAA,MAC7H,QAAQ,yFAAyF,IAAI,mBAAmB,MAAM,2CAA2C,IAAI,mBAAmB,MAAM;AAAA,MACtM,cAAc;AAAA,IAChB,CAAC;AAAA,EACH;AAEA,SAAO;AACT;;;ACjlBA,SAAS,kBACP,WACA,UAC8C;AAE9C,WAAS,KAAK,GAAmB;AAC/B,QAAI,IAAI,eAAe;AACvB,aAAS,IAAI,GAAG,IAAI,EAAE,QAAQ,KAAK;AACjC,WAAK,EAAE,WAAW,CAAC;AACnB,UAAI,KAAK,KAAK,GAAG,QAAQ,MAAM;AAAA,IACjC;AACA,WAAO;AAAA,EACT;AACA,QAAM,SAAS,CAAC,GAAG,SAAS,EAAE,KAAK,CAAC,GAAG,MAAM,KAAK,EAAE,EAAE,IAAI,KAAK,EAAE,EAAE,CAAC;AACpE,QAAM,WAAW,KAAK,IAAI,GAAG,KAAK,IAAI,OAAO,SAAS,GAAG,KAAK,MAAM,OAAO,SAAS,QAAQ,CAAC,CAAC;AAC9F,SAAO;AAAA,IACL,SAAS,OAAO,MAAM,GAAG,QAAQ;AAAA,IACjC,OAAO,OAAO,MAAM,QAAQ;AAAA,EAC9B;AACF;AAEA,SAAS,cAAc,YAGrB;AACA,QAAM,cAAsC,CAAC;AAC7C,QAAM,SAAmB,CAAC;AAC1B,aAAW,CAAC,IAAI,GAAG,KAAK,OAAO,QAAQ,UAAU,GAAG;AAClD,gBAAY,EAAE,IAAI,IAAI;AACtB,WAAO,KAAK,IAAI,aAAa;AAAA,EAC/B;AACA,SAAO;AAAA,IACL,eAAe,OAAO,WAAW,IAAI,IAAI,OAAO,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,OAAO;AAAA,IACpF;AAAA,EACF;AACF;AAEA,IAAM,8BAA8B;AAAA,EAClC;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF;AA2BA,eAAsB,YACpB,MACkD;AAClD,QAAM,YAAY,KAAK,IAAI;AAE3B,QAAM,SAAS,KAAK,UAAU,CAAC;AAC/B,QAAM,cAAc,OAAO,eAAe;AAC1C,QAAM,iBAAiB,OAAO,kBAAkB;AAChD,QAAM,iBAAiB,OAAO,kBAAkB;AAChD,QAAM,kBAAkB,OAAO,mBAAmB;AAClD,QAAM,cAAc,OAAO;AAE3B,QAAM,kBAAkB,OAAO;AAC/B,QAAM,EAAE,OAAO,QAAQ,IAAI,kBACvB;AAAA,IACE,OAAO,KAAK,UAAU,OAAO,CAAC,MAAM,CAAC,gBAAgB,KAAK,CAAC,MAAM,EAAE,OAAO,EAAE,EAAE,CAAC;AAAA,IAC/E,SAAS;AAAA,EACX,IACA,kBAAkB,KAAK,WAAW,eAAe;AAErD,MAAI,MAAM,WAAW,GAAG;AACtB,UAAM,IAAI;AAAA,MACR;AAAA,IACF;AAAA,EACF;AACA,MAAI,QAAQ,WAAW,GAAG;AACxB,UAAM,IAAI,MAAM,2DAA2D;AAAA,EAC7E;AAEA,QAAM,SACJ,KAAK,UACL,WAAW;AAAA,IACT,KAAK;AAAA,MACH,SAAS,KAAK,KAAK,WAAW;AAAA,MAC9B,QAAQ,KAAK,KAAK,UAAU,QAAQ,IAAI,kBAAkB;AAAA,IAC5D;AAAA,IACA,OAAO,KAAK,KAAK,SAAS;AAAA,IAC1B,QACE,KAAK,gBACL;AAAA,IACF,oBAAoB,KAAK,sBAAsB;AAAA,EACjD,CAAC;AAEH,QAAM,OACJ,KAAK,QACL,sBAA4C;AAAA,IAC1C,kBAAkB;AAAA,IAClB,gBAAgB;AAAA,EAClB,CAAC;AAEH,QAAM,UAAU,KAAK,WAAW,wBAAwB;AACxD,QAAM,SAAS,KAAK,UAAU,qBAAqB,SAAS;AAE5D,MAAI,KAAK,YAAY;AACnB,SAAK,WAAW,EAAE,MAAM,oBAAoB,WAAW,KAAK,UAAU,OAAO,CAAC;AAAA,EAChF;AAEA,QAAM,SAAS,MAAM,mBAAyC;AAAA,IAC5D,WAAW;AAAA,IACX,iBAAiB,KAAK;AAAA,IACtB,qBAAqB,KAAK;AAAA,IAC1B;AAAA,IACA,QAAQ,CAAC,KAAK,KAAK;AAAA,IACnB;AAAA,IACA,gBAAgB;AAAA,IAChB,kBAAkB;AAAA,IAClB;AAAA,IACA,eAAe,KAAK,iBAAiB;AAAA,IACrC,SAAS,KAAK;AAAA,IACd,QAAQ,KAAK;AAAA,IACb;AAAA,IACA;AAAA,IACA;AAAA,IACA,eAAe,KAAK;AAAA,IACpB;AAAA,EACF,CAAC;AAED,QAAM,WAAW,cAAc,OAAO,kBAAkB,WAAW,UAAU;AAC7E,QAAM,cAAc,cAAc,OAAO,gBAAgB,WAAW,UAAU;AAE9E,MAAI,KAAK,YAAY;AACnB,SAAK,WAAW;AAAA,MACd,MAAM;AAAA,MACN,eAAe,SAAS;AAAA,MACxB,YAAY,KAAK,IAAI,IAAI;AAAA,IAC3B,CAAC;AACD,SAAK,WAAW;AAAA,MACd,MAAM;AAAA,MACN,UAAU,OAAO,WAAW;AAAA,MAC5B,MAAM,YAAY,gBAAgB,SAAS;AAAA,IAC7C,CAAC;AAAA,EACH;AAEA,QAAM,YACJ,OAAO,iBAAiB,WAAW,eACnC,OAAO,YAAY;AAAA,IACjB,CAAC,KAAK,QACJ,MAAM,IAAI,SAAS,OAAO,CAAC,GAAG,OAAO,IAAI,GAAG,SAAS,WAAW,cAAc,CAAC;AAAA,IACjF;AAAA,EACF;AAMF,QAAM,UAAU,MAAM,YAAY;AAAA,IAChC,MAAM;AAAA,MACJ,GAAG,kBAAkB,OAAO,iBAAiB,OAAO,YAAY,MAAM;AAAA,MACtE,GAAG,kBAAkB,OAAO,gBAAgB,OAAO,UAAU,MAAM;AAAA,IACrE;AAAA,IACA,qBAAqB;AAAA,IACrB,sBAAsB;AAAA,EACxB,CAAC;AAED,QAAM,UAAmD;AAAA,IACvD;AAAA,IACA,QAAQ;AAAA,MACN,GAAG;AAAA,MACH,SAAS,OAAO;AAAA,IAClB;AAAA,IACA,MAAM,YAAY,gBAAgB,SAAS;AAAA,IAC3C,cAAc,OAAO,WAAW;AAAA,IAChC,qBAAqB,OAAO,YAAY;AAAA,IACxC,YAAY,KAAK,IAAI,IAAI;AAAA,IACzB,cAAc;AAAA,IACd;AAAA,IACA,KAAK;AAAA,EACP;AAKA,MAAI,KAAK,cAAc;AACrB,QAAI;AACF,YAAM,oBAAoB,KAAK,cAAc,MAAM,SAAS,QAAQ,MAAM;AAAA,IAC5E,SAAS,KAAK;AACZ,YAAM,MAAM,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AAE3D,cAAQ,KAAK,mDAAmD,GAAG,EAAE;AAAA,IACvE;AAAA,EACF;AAEA,SAAO;AACT;AAEA,eAAe,oBACb,QACA,MACA,SACA,KACA,QACe;AACf,QAAM,SAAS,mBAAmB,MAAM;AAExC,WAAS,qBACP,OACA,SACA,UACA,YAC2B;AAC3B,UAAM,QAA4B,SAAS,MAAM,IAAI,CAAC,SAAS;AAC7D,YAAM,cAAc,OAAO,OAAO,KAAK,WAAW;AAClD,YAAM,YACJ,YAAY,WAAW,IACnB,IACA,YAAY,OAAO,CAAC,GAAG,MAAM,IAAI,EAAE,WAAW,CAAC,IAAI,YAAY;AACrE,aAAO;AAAA,QACL,YAAY,KAAK;AAAA,QACjB,KAAK,KAAK;AAAA,QACV,eAAe;AAAA,QACf,YAAY,OAAO;AAAA,UACjB,OAAO,QAAQ,KAAK,WAAW,EAAE,IAAI,CAAC,CAAC,MAAM,KAAK,MAAM,CAAC,MAAM,MAAM,UAAU,CAAC;AAAA,QAClF;AAAA,QACA,cAAc,KAAK,SAAS;AAAA,MAC9B;AAAA,IACF,CAAC;AACD,UAAM,gBACJ,MAAM,WAAW,IAAI,IAAI,MAAM,OAAO,CAAC,GAAG,MAAM,IAAI,EAAE,eAAe,CAAC,IAAI,MAAM;AAClF,WAAO;AAAA,MACL;AAAA,MACA,aACE,OAAO,YAAY,WACf,WAAW,OAAO,IAClB,WAAW,KAAK,UAAU,WAAW,EAAE,CAAC;AAAA,MAC9C;AAAA,MACA;AAAA,MACA;AAAA,MACA,SAAS,SAAS,WAAW;AAAA,MAC7B;AAAA,IACF;AAAA,EACF;AAEA,QAAM,cAA2C,CAAC;AAElD,cAAY,KAAK,qBAAqB,GAAG,KAAK,iBAAiB,IAAI,kBAAkB,CAAC,CAAC;AAGvF,aAAW,OAAO,IAAI,aAAa;AACjC,UAAM,SAAS,IAAI,SAAS;AAAA,MAC1B,CAAC,MAAM,MACL,EAAE,SAAS,WAAW,gBAAgB,MACrC,SAAS,UAAa,iBAAiB,EAAE,QAAQ,IAAI,iBAAiB,KAAK,QAAQ,KAChF,IACA;AAAA,MACN,IAAI,SAAS,CAAC;AAAA,IAChB;AACA,QAAI,CAAC,OAAQ;AACb,gBAAY;AAAA,MACV,qBAAqB,IAAI,OAAO,kBAAkB,GAAG,OAAO,SAAS,OAAO,UAAU,CAAC;AAAA,IACzF;AAAA,EACF;AAEA,QAAM,QAAsB;AAAA,IAC1B,OAAO,GAAG,MAAM,IAAI,KAAK,IAAI,CAAC;AAAA,IAC9B;AAAA,IACA,YAAW,oBAAI,KAAK,GAAE,YAAY;AAAA,IAClC,QAAQ;AAAA,IACR,QAAQ,KAAK,gBAAgB,CAAC;AAAA,IAC9B,UAAU,YAAY,CAAC;AAAA,IACvB;AAAA,IACA,cAAc,QAAQ;AAAA,IACtB,aAAa,QAAQ;AAAA,IACrB,cAAc,QAAQ;AAAA,IACtB,iBAAiB,QAAQ;AAAA,IACzB,eAAe,QAAQ;AAAA,EACzB;AAEA,QAAM,OAAO,cAAc,KAAK;AAClC;AAEA,SAAS,iBACP,UACQ;AACR,QAAM,OAAO,OAAO,OAAO,SAAS,WAAW,UAAU;AACzD,SAAO,KAAK,WAAW,IAAI,IAAI,KAAK,OAAO,CAAC,GAAG,MAAM,IAAI,EAAE,eAAe,CAAC,IAAI,KAAK;AACtF;AAEA,SAAS,WAAW,GAAmB;AACrC,MAAI,IAAI,eAAe;AACvB,WAAS,IAAI,GAAG,IAAI,EAAE,QAAQ,KAAK;AACjC,SAAK,EAAE,WAAW,CAAC;AACnB,QAAI,KAAK,KAAK,GAAG,QAAQ,MAAM;AAAA,EACjC;AACA,SAAO,EAAE,SAAS,EAAE,EAAE,SAAS,GAAG,GAAG;AACvC;AAOA,SAAS,kBACP,OACA,aACA,OACa;AACb,SAAO,MAAM,IAAI,CAAC,SAAS;AACzB,UAAM,WAAmD,CAAC;AAC1D,UAAM,kBAA8D,CAAC;AACrE,QAAI,eAAe;AACnB,QAAI,iBAAiB;AACrB,eAAW,CAAC,SAAS,KAAK,KAAK,OAAO,QAAQ,KAAK,WAAW,GAAG;AAC/D,eAAS,OAAO,IAAI,EAAE,GAAG,MAAM,WAAW;AAC1C,iBAAW,CAAC,KAAK,KAAK,KAAK,OAAO,QAAQ,MAAM,UAAU,GAAG;AAC3D,YAAI,CAAC,OAAO,SAAS,KAAK,EAAG;AAC7B,cAAM,QAAQ,gBAAgB,GAAG,KAAK,EAAE,KAAK,GAAG,GAAG,EAAE;AACrD,cAAM,OAAO;AACb,cAAM,KAAK;AACX,wBAAgB,GAAG,IAAI;AAAA,MACzB;AACA,UAAI,OAAO,SAAS,MAAM,SAAS,GAAG;AACpC,wBAAgB,MAAM;AACtB,0BAAkB;AAAA,MACpB;AAAA,IACF;AACA,UAAM,aAAqC,CAAC;AAC5C,eAAW,CAAC,KAAK,EAAE,KAAK,EAAE,CAAC,KAAK,OAAO,QAAQ,eAAe,GAAG;AAC/D,iBAAW,GAAG,IAAI,MAAM,IAAI,IAAI,MAAM;AAAA,IACxC;AACA,UAAM,YAAY,mBAAmB,IAAI,IAAI,eAAe;AAC5D,UAAM,cAAiC;AAAA,MACrC;AAAA,MACA;AAAA,MACA;AAAA,IACF;AACA,WAAO;AAAA,MACL,OAAO,GAAG,KAAK,KAAK,WAAW,KAAK,KAAK,MAAM;AAAA,MAC/C,cAAc;AAAA,MACd;AAAA;AAAA;AAAA,MAGA,MACE,KAAK,MAAM,MACX,WAAW,KAAK,UAAU,EACvB,MAAM,GAAG,CAAC,EACV,MAAM,EAAE,EACR,OAAO,CAAC,GAAG,MAAO,IAAI,KAAK,EAAE,WAAW,CAAC,MAAO,GAAG,CAAC;AAAA,MACzD,OAAO;AAAA,MACP,YAAY;AAAA,MACZ,YAAY;AAAA,MACZ,WAAW;AAAA,MACX,QAAQ,KAAK;AAAA,MACb,SAAS,KAAK;AAAA,MACd,YAAY,EAAE,OAAO,GAAG,QAAQ,EAAE;AAAA,MAClC,SAAS;AAAA,QACP,cAAc;AAAA,QACd,KAAK,CAAC;AAAA,QACN;AAAA,MACF;AAAA,MACA,UAAU;AAAA,MACV,GAAI,KAAK,QAAQ,EAAE,aAAa,KAAK,MAAM,IAAI,CAAC;AAAA,IAClD;AAAA,EACF,CAAC;AACH;;;ACvgBO,SAAS,kBAAkB,MAAyD;AACzF,QAAM,EAAE,SAAS,OAAO,CAAC,GAAG,OAAO,kBAAkB,KAAK,IAAI;AAC9D,QAAM,YAAY,IAAI,IAAI,KAAK,IAAI,CAAC,MAAM,CAAC,EAAE,OAAO,CAAC,CAAC,CAAC;AAGvD,QAAM,YAAY,CAAC,WAAqC;AACtD,QAAI,OAAO,WAAW,UAAW,QAAO,SAAS,IAAI;AACrD,QAAI,CAAC,OAAO,SAAS,MAAM,EAAG,QAAO,OAAO;AAC5C,QAAI,CAAC,MAAO,QAAO;AACnB,UAAM,EAAE,KAAK,IAAI,IAAI;AACrB,QAAI,QAAQ,IAAK,QAAO;AACxB,YAAQ,SAAS,QAAQ,MAAM;AAAA,EACjC;AAGA,QAAM,QAAQ,oBAAI,IAAgC;AAClD,aAAW,OAAO,SAAS;AACzB,UAAM,OAAO,MAAM,IAAI,IAAI,KAAK,KAAK,CAAC;AACtC,SAAK,KAAK,GAAG;AACb,UAAM,IAAI,IAAI,OAAO,IAAI;AAAA,EAC3B;AAEA,QAAM,OAAoB,CAAC;AAC3B,QAAM,cAAsD,CAAC;AAE7D,aAAW,CAAC,OAAO,UAAU,KAAK,OAAO;AACvC,UAAM,aAAa,WAChB,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,OAAO,OAAO,UAAU,EAAE,MAAM,EAAE,EAAE,EAC3D,OAAO,CAAC,MAAM,OAAO,SAAS,EAAE,KAAK,CAAC;AACzC,QAAI,WAAW,WAAW,EAAG;AAE7B,UAAM,YAAY,WAAW,OAAO,CAAC,GAAG,MAAM,IAAI,EAAE,OAAO,CAAC,IAAI,WAAW;AAE3E,UAAM,UAAU,UAAU,IAAI,KAAK,KAAM,EAAE,MAAM;AAEjD,UAAM,cAAiC;AAAA,MACrC,UAAU,OAAO,YAAY,WAAW,IAAI,CAAC,MAAM,CAAC,EAAE,OAAO,EAAE,QAAQ,EAAE,MAAM,CAAC,CAAC,CAAC;AAAA,MAClF,YAAY,EAAE,QAAQ,UAAU;AAAA,MAChC,WAAW;AAAA,IACb;AAEA,UAAM,UAAsB;AAAA;AAAA;AAAA;AAAA,MAI1B,cAAc;AAAA,MACd,KAAK,OAAO,YAAY,WAAW,IAAI,CAAC,MAAM,CAAC,SAAS,EAAE,KAAK,IAAI,EAAE,KAAK,CAAC,CAAC;AAAA,MAC5E;AAAA,IACF;AAEA,SAAK,KAAK;AAAA,MACR;AAAA,MACA,cAAc,QAAQ,gBAAgB;AAAA,MACtC,aAAa,QAAQ,eAAe;AAAA,MACpC,MAAM;AAAA,MACN,OAAO,QAAQ,SAAS;AAAA,MACxB,YAAY,QAAQ,cAAc;AAAA,MAClC,YAAY,QAAQ,cAAc;AAAA,MAClC,WAAW,QAAQ,aAAa;AAAA,MAChC,QAAQ,QAAQ,UAAU;AAAA,MAC1B,SAAS,QAAQ,WAAW;AAAA,MAC5B,YAAY,EAAE,OAAO,GAAG,QAAQ,EAAE;AAAA,MAClC;AAAA,MACA,UAAU,QAAQ,YAAY;AAAA,IAChC,CAAc;AAEd,QAAI,iBAAiB;AACnB,iBAAW,KAAK,WAAY,aAAY,KAAK,EAAE,OAAO,OAAO,EAAE,OAAO,OAAO,EAAE,MAAM,CAAC;AAAA,IACxF;AAAA,EACF;AAEA,SAAO,EAAE,MAAM,YAAY;AAC7B;;;AC5HA,IAAM,aAAa,CAAC,gBAAgB,cAAc,OAAO;AACzD,IAAM,aAAa,CAAC,gBAAgB,wBAAwB,aAAa,OAAO;AAChF,IAAM,YAAY,CAAC,mBAAmB,yBAAyB,YAAY,MAAM;AACjF,IAAM,mBAAmB,CAAC,6BAA6B,oBAAoB,WAAW;AACtF,IAAM,oBAAoB,CAAC,8BAA8B,qBAAqB,YAAY;AAC1F,IAAM,mBAAmB,CAAC,sBAAsB,aAAa;AAC7D,IAAM,mBAAmB,CAAC,sBAAsB,aAAa;AAUtD,SAAS,cAAc,MAAyC;AACrE,QAAM,EAAE,OAAO,eAAe,WAAW,eAAe,cAAc,IAAI;AAC1E,QAAM,UAAU,WAAW,KAAK;AAEhC,QAAM,OAAoB,CAAC;AAC3B,aAAW,CAAC,UAAUC,WAAU,KAAK,SAAS;AAC5C,UAAM,OAAO,SAASA,WAAU;AAChC,QAAI,CAAC,KAAM;AAEX,UAAM,SAAS,KAAK,IAAI,IAAI,KAAK,kBAAkB,KAAK,qBAAqB,GAAS;AACtF,UAAM,QAAQ,eAAeA,aAAY,UAAU,KAAK;AACxD,UAAM,UAAU,eAAeA,aAAY,SAAS,KAAK;AACzD,UAAM,cAAc,eAAeA,aAAY,gBAAgB,KAAK;AACpE,UAAM,eAAe,eAAeA,aAAY,iBAAiB,KAAK;AACtE,UAAM,aAAa,eAAeA,aAAY,gBAAgB,KAAK;AACnE,UAAM,aAAa,eAAeA,aAAY,gBAAgB,KAAK;AACnE,UAAM,QAAQ,eAAeA,aAAY,UAAU;AAEnD,UAAM,aAAa,oBAAoBA,WAAU;AACjD,UAAM,aAA4B;AAAA,MAChC,OAAO;AAAA,MACP,QAAQ;AAAA,IACV;AAEA,UAAM,cACJ,UAAU,SACN;AAAA,MACE,UAAU,EAAE,gBAAgB,EAAE,MAAM,EAAE;AAAA,MACtC,YAAY,EAAE,MAAM;AAAA,MACpB,WAAW;AAAA,IACb,IACA;AAEN,UAAM,YAAYA,YAAW,KAAK,CAAC,MAAM,EAAE,QAAQ,SAAS,OAAO;AACnE,UAAM,UAAsB;AAAA,MAC1B,GAAI,KAAK,iBAAiB,WAAW,EAAE,aAAa,MAAM,IAAI,EAAE,cAAc,MAAM;AAAA,MACpF,KAAK;AAAA,MACL,GAAI,cAAc,EAAE,YAAY,IAAI,CAAC;AAAA,IACvC;AAEA,SAAK,KAAK;AAAA,MACR,OAAO;AAAA,MACP;AAAA,MACA,aAAc,KAAK,WAAW,oBAAoB,KAA4B;AAAA,MAC9E,MAAM;AAAA,MACN;AAAA,MACA;AAAA,MACA;AAAA,MACA,WAAY,KAAK,WAAW,mBAAmB,KAA4B;AAAA,MAC3E;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA,UAAU;AAAA,MACV,GAAI,YAAY,EAAE,aAAa,UAAU,KAAK,IAAI,CAAC;AAAA,IACrD,CAAc;AAAA,EAChB;AACA,SAAO;AACT;AAIA,SAAS,WAAW,OAAwD;AAC1E,QAAM,IAAI,oBAAI,IAA8B;AAC5C,aAAW,QAAQ,OAAO;AACxB,UAAM,MAAO,KAAK,cAAc,KAA4B,KAAK;AACjE,UAAM,OAAO,EAAE,IAAI,GAAG,KAAK,CAAC;AAC5B,SAAK,KAAK,IAAI;AACd,MAAE,IAAI,KAAK,IAAI;AAAA,EACjB;AACA,SAAO;AACT;AAEA,SAAS,SAAS,OAAqD;AACrE,SAAO,MAAM,KAAK,CAAC,MAAM,CAAC,EAAE,YAAY,KAAK,MAAM,CAAC;AACtD;AAEA,SAAS,eAAe,OAAyB,MAAoC;AACnF,aAAW,QAAQ,OAAO;AACxB,eAAW,OAAO,MAAM;AACtB,YAAM,IAAI,KAAK,WAAW,GAAG;AAC7B,UAAI,OAAO,MAAM,YAAY,EAAE,SAAS,EAAG,QAAO;AAAA,IACpD;AAAA,EACF;AACA,SAAO;AACT;AAEA,SAAS,eAAe,OAAyB,MAAoC;AACnF,aAAW,QAAQ,OAAO;AACxB,eAAW,OAAO,MAAM;AACtB,YAAM,IAAI,KAAK,WAAW,GAAG;AAC7B,UAAI,OAAO,MAAM,YAAY,OAAO,SAAS,CAAC,EAAG,QAAO;AACxD,UAAI,OAAO,MAAM,UAAU;AACzB,cAAM,SAAS,OAAO,CAAC;AACvB,YAAI,OAAO,SAAS,MAAM,EAAG,QAAO;AAAA,MACtC;AAAA,IACF;AAAA,EACF;AACA,SAAO;AACT;AAEA,SAAS,oBAAoB,OAAiD;AAC5E,QAAM,MAA8B,CAAC;AACrC,aAAW,QAAQ,OAAO;AACxB,eAAW,CAAC,GAAG,CAAC,KAAK,OAAO,QAAQ,KAAK,UAAU,GAAG;AACpD,UAAI,OAAO,MAAM,YAAY,OAAO,SAAS,CAAC,EAAG,KAAI,CAAC,IAAI;AAAA,IAC5D;AAAA,EACF;AACA,SAAO;AACT;","names":["mean","i","groupSpans"]}
+{"version":3,"sources":["../../src/contract/analyze-runs.ts","../../src/contract/self-improve.ts","../../src/contract/intake/feedback-table.ts","../../src/contract/intake/otel-spans.ts"],"sourcesContent":["/**\n * # `analyzeRuns()` — turn a set of agent runs into an actionable decision packet.\n *\n * Wires the substrate's statistical, calibration, clustering, Pareto, and\n * release-confidence primitives into one `InsightReport`. Two top-level\n * entry points use this function:\n *\n *   - `selfImprove()` calls it on the campaign output to attach a packet\n *     to every run.\n *   - Consumers with observed `RunRecord[]` (production traces, gold\n *     corpora, approve/reject tables) call it directly via `analyzeRuns()`\n *     for analysis without a closed loop.\n *\n * Every section is opt-in based on what the input data supports — the\n * function never invents signal. If runs carry no judge scores, `judges`\n * is empty. If there's no baseline/candidate split, `lift` is undefined.\n * If no `analyst` is wired, `failureClusters` is undefined.\n *\n * The `recommendations` array is the human-readable layer; everything\n * else is the evidence backing each recommendation.\n */\n\nimport type { AnalystRegistry } from '../analyst/registry'\nimport type { AnalystFinding } from '../analyst/types'\nimport { checkCanaries } from '../contamination-guard'\nimport type { DatasetScenario } from '../dataset'\nimport type { RunRecord } from '../run-record'\nimport { cohensD, pairedBootstrap, pairedMde, pairedTTest, requiredSampleSize } from '../statistics'\nimport { type ParetoFigureSpec, paretoChart } from '../summary-report'\n\nimport type {\n  FailureClusterInsight,\n  InsightReport,\n  InterRaterInsight,\n  JudgeInsight,\n  LiftInsight,\n  MetricDelta,\n  OutcomeCorrelationInsight,\n  PriorPeriodComparison,\n  Recommendation,\n  ScalarDistribution,\n} from './insight-report'\n\n// ── Public API ───────────────────────────────────────────────────────\n\nexport interface AnalyzeRunsOptions {\n  /** The runs to analyze. */\n  runs: RunRecord[]\n  /** Which split to score against when reading composite from RunOutcome.\n   *  Default: holdout when ANY run has a `holdoutScore`, else search. */\n  split?: 'search' | 'holdout' | 'auto'\n  /** Pairwise analysis configuration. When both `baselineCandidateId` and\n   *  `candidateCandidateId` are present, lift is computed on paired\n   *  (experimentId, seed) tuples shared between the two sides. */\n  baselineCandidateId?: string\n  candidateCandidateId?: string\n  /** Canary scenarios — checked against every run's raw output for\n   *  holdout contamination. */\n  canaryScenarios?: DatasetScenario[]\n  /** Analyst registry for failure clustering. When omitted, the\n   *  `failureClusters` section is left undefined. */\n  analyst?: AnalystRegistry\n  /** Downstream outcome metric per run (e.g. engagement rate, approval\n   *  rate, downstream pass rate). When present, the report includes\n   *  `outcomeCorrelation` + a simple linear reward model fit. */\n  outcomeSignal?: {\n    metric: string\n    valueByRunId: Record<string, number>\n  }\n  /** Multi-rater feedback for inter-rater agreement. Each entry is one\n   *  rater's score for one run. Two or more raters → kappa + disagreement\n   *  triage list. */\n  raterScores?: Array<{ runId: string; rater: string; score: number }>\n  /** Number of histogram bins for distributional summaries. Default 12. */\n  histogramBins?: number\n  /** Decision threshold — the smallest composite lift the caller cares\n   *  about. Used by the recommendations engine to call ship vs hold.\n   *  Default 0.02. */\n  decisionThreshold?: number\n  /** Optional prior-period runs. When set, the report includes\n   *  `priorPeriodComparison` with per-metric Welch-CI deltas and\n   *  recommendations fire on statistically significant regressions.\n   *  The two windows do NOT have to share scenarios — the comparison\n   *  is two-sample unpaired (the substrate's `lift` field uses paired\n   *  bootstrap on shared (experimentId, seed) tuples; this is the\n   *  shape for \"this week vs last week\" rather than \"candidate vs\n   *  baseline within a campaign\"). */\n  baselineRuns?: RunRecord[]\n  /** Human-readable label for the baseline window, e.g. \"vs prior 7\n   *  days\", \"vs v3.1 release\". Surfaces in recommendations + UI. */\n  baselineLabel?: string\n}\n\nexport async function analyzeRuns(opts: AnalyzeRunsOptions): Promise<InsightReport> {\n  const runs = opts.runs\n  const bins = opts.histogramBins ?? 12\n  const threshold = opts.decisionThreshold ?? 0.02\n  const split = resolveSplit(runs, opts.split ?? 'auto')\n\n  const compositeWithIds = runs\n    .map((r) => ({ runId: r.runId, score: compositeOf(r, split) }))\n    .filter((p) => Number.isFinite(p.score))\n  const composite = distributionOf(\n    compositeWithIds.map((p) => p.score),\n    bins,\n    compositeWithIds,\n  )\n\n  const perDimension = computePerDimension(runs, bins)\n\n  const costs = runs.map((r) => r.costUsd).filter(Number.isFinite)\n  const costDist = distributionOf(costs, bins)\n  const pareto = paretoChart(runs, { split })\n  const degraded: { cost?: string; pareto?: string } = {}\n  if (costs.length === 0 || costs.every((c) => c === 0)) {\n    degraded.cost = 'no costUsd values recorded — cost axis carries no signal'\n  }\n  if (pareto.points.length < 2) {\n    degraded.pareto =\n      pareto.points.length === 0\n        ? 'no candidates — Pareto unavailable'\n        : 'single candidate — Pareto is a single point, not a frontier'\n  }\n  const costQuality = {\n    cost: costDist,\n    pareto,\n    ...(degraded.cost || degraded.pareto ? { degraded } : {}),\n  }\n\n  const judges = computeJudgeInsights(runs)\n\n  const interRater = opts.raterScores ? computeInterRater(opts.raterScores) : undefined\n\n  const lift = computeLift(runs, opts.baselineCandidateId, opts.candidateCandidateId, split)\n\n  const failureClusters = opts.analyst\n    ? await computeFailureClusters(runs, opts.analyst, split)\n    : undefined\n\n  const contamination = opts.canaryScenarios\n    ? computeContamination(runs, opts.canaryScenarios)\n    : undefined\n\n  const outcomeCorrelation = opts.outcomeSignal\n    ? computeOutcomeCorrelation(runs, opts.outcomeSignal, split)\n    : undefined\n\n  const release = buildReleaseScorecard(composite, lift, contamination)\n\n  const priorPeriodComparison = opts.baselineRuns\n    ? computePriorPeriodComparison(runs, opts.baselineRuns, split, opts.baselineLabel)\n    : undefined\n\n  const recommendations = buildRecommendations({\n    composite,\n    judges,\n    interRater,\n    lift,\n    failureClusters,\n    contamination,\n    outcomeCorrelation,\n    priorPeriodComparison,\n    threshold,\n  })\n\n  return {\n    n: runs.length,\n    composite,\n    perDimension,\n    costQuality,\n    judges,\n    interRater,\n    lift,\n    failureClusters,\n    contamination,\n    outcomeCorrelation,\n    release,\n    ...(priorPeriodComparison ? { priorPeriodComparison } : {}),\n    recommendations,\n  }\n}\n\n// ── Prior-period comparison ─────────────────────────────────────────\n\n/** Direction of the metric — does \"higher current\" mean better or worse?\n *  Composite + judge dimensions: higher is better. Cost + duration: lower\n *  is better. The recommendations engine flips the sign before judging\n *  regressed vs improved. */\ntype MetricDirection = 'higher-is-better' | 'lower-is-better'\n\nfunction computePriorPeriodComparison(\n  current: RunRecord[],\n  baseline: RunRecord[],\n  split: 'search' | 'holdout',\n  windowLabel: string | undefined,\n): PriorPeriodComparison | undefined {\n  if (current.length === 0 || baseline.length === 0) return undefined\n\n  const metrics: Record<string, MetricDelta> = {}\n  const directions: Record<string, MetricDirection> = {}\n\n  const compositeCurrent = current\n    .map((r) => compositeOf(r, split))\n    .filter(Number.isFinite) as number[]\n  const compositeBaseline = baseline\n    .map((r) => compositeOf(r, split))\n    .filter(Number.isFinite) as number[]\n  if (compositeCurrent.length > 0 && compositeBaseline.length > 0) {\n    metrics.composite = welchCompare(compositeBaseline, compositeCurrent)\n    directions.composite = 'higher-is-better'\n  }\n\n  const costCurrent = current.map((r) => r.costUsd).filter(Number.isFinite)\n  const costBaseline = baseline.map((r) => r.costUsd).filter(Number.isFinite)\n  if (costCurrent.length > 0 && costBaseline.length > 0) {\n    metrics.cost = welchCompare(costBaseline, costCurrent)\n    directions.cost = 'lower-is-better'\n  }\n\n  const durCurrent = current.map((r) => r.wallMs).filter(Number.isFinite)\n  const durBaseline = baseline.map((r) => r.wallMs).filter(Number.isFinite)\n  if (durCurrent.length > 0 && durBaseline.length > 0) {\n    metrics.duration = welchCompare(durBaseline, durCurrent)\n    directions.duration = 'lower-is-better'\n  }\n\n  const tokCurrent = current\n    .map((r) => (r.tokenUsage.input ?? 0) + (r.tokenUsage.output ?? 0))\n    .filter(Number.isFinite)\n  const tokBaseline = baseline\n    .map((r) => (r.tokenUsage.input ?? 0) + (r.tokenUsage.output ?? 0))\n    .filter(Number.isFinite)\n  if (tokCurrent.length > 0 && tokBaseline.length > 0) {\n    metrics.tokenUsage = welchCompare(tokBaseline, tokCurrent)\n    directions.tokenUsage = 'lower-is-better'\n  }\n\n  // Per-dimension judge comparisons — only for dimensions present in BOTH\n  // windows. We use perDimMean since per-judge nesting is finicky for\n  // two-sample comparisons across different judge configurations.\n  const dimsCurrent = collectPerDimension(current)\n  const dimsBaseline = collectPerDimension(baseline)\n  for (const dim of Object.keys(dimsCurrent)) {\n    const b = dimsBaseline[dim]\n    const c = dimsCurrent[dim]\n    if (!b || b.length === 0 || !c || c.length === 0) continue\n    metrics[`dim.${dim}`] = welchCompare(b, c)\n    directions[`dim.${dim}`] = 'higher-is-better'\n  }\n\n  const regressedMetrics: string[] = []\n  const improvedMetrics: string[] = []\n  for (const [name, delta] of Object.entries(metrics)) {\n    if (!delta.significant) continue\n    const dir = directions[name] ?? 'higher-is-better'\n    const better = dir === 'higher-is-better' ? delta.delta > 0 : delta.delta < 0\n    if (better) improvedMetrics.push(name)\n    else regressedMetrics.push(name)\n  }\n\n  return {\n    baselineN: baseline.length,\n    currentN: current.length,\n    ...(windowLabel ? { windowLabel } : {}),\n    metrics,\n    regressedMetrics,\n    improvedMetrics,\n  }\n}\n\n/** Collect per-dimension values across runs (from outcome.judgeScores.perDimMean). */\nfunction collectPerDimension(runs: RunRecord[]): Record<string, number[]> {\n  const out: Record<string, number[]> = {}\n  for (const r of runs) {\n    const perDim = r.outcome.judgeScores?.perDimMean\n    if (!perDim) continue\n    for (const [dim, value] of Object.entries(perDim)) {\n      if (!Number.isFinite(value)) continue\n      if (!out[dim]) out[dim] = []\n      out[dim].push(value as number)\n    }\n  }\n  return out\n}\n\n/** Two-sample Welch comparison: unequal-variance t-test + CI on the delta\n *  + Cohen's d (pooled stddev). Significance = p < 0.05 AND |d| >= 0.2. */\nfunction welchCompare(baseline: number[], current: number[]): MetricDelta {\n  const baselineMean = mean(baseline)\n  const currentMean = mean(current)\n  const baselineVar = sampleVariance(baseline, baselineMean)\n  const currentVar = sampleVariance(current, currentMean)\n  const baselineN = baseline.length\n  const currentN = current.length\n  const delta = currentMean - baselineMean\n\n  // Welch standard error\n  const se = Math.sqrt(baselineVar / baselineN + currentVar / currentN)\n  // For 95% CI we use z=1.96 (large-n approximation). Customers running\n  // analyzeRuns will typically have n >= 30; the t-correction is\n  // negligible vs the practical noise floor.\n  const halfWidth = 1.96 * (se > 0 ? se : 0)\n  const ci95: [number, number] = [delta - halfWidth, delta + halfWidth]\n\n  // p-value via normal approximation to the t-statistic.\n  const t = se > 0 ? delta / se : 0\n  const pValue = se > 0 ? 2 * (1 - standardNormalCdf(Math.abs(t))) : 1\n\n  // Cohen's d — pooled stddev.\n  const pooledStddev = Math.sqrt(\n    ((baselineN - 1) * baselineVar + (currentN - 1) * currentVar) /\n      Math.max(1, baselineN + currentN - 2),\n  )\n  const cohensD = pooledStddev > 0 ? delta / pooledStddev : 0\n\n  // Significance: BOTH p < 0.05 AND |d| >= 0.2 (small-effect threshold).\n  const significant = pValue < 0.05 && Math.abs(cohensD) >= 0.2\n\n  return {\n    current: currentMean,\n    baseline: baselineMean,\n    delta,\n    ci95,\n    pValue,\n    cohensD,\n    baselineN,\n    currentN,\n    significant,\n  }\n}\n\nfunction sampleVariance(xs: number[], xsMean: number): number {\n  if (xs.length < 2) return 0\n  let s = 0\n  for (const x of xs) s += (x - xsMean) ** 2\n  return s / (xs.length - 1)\n}\n\n/** Abramowitz & Stegun approximation to Φ(z). Maximum error ~7.5e-8. */\nfunction standardNormalCdf(z: number): number {\n  const a1 = 0.254829592\n  const a2 = -0.284496736\n  const a3 = 1.421413741\n  const a4 = -1.453152027\n  const a5 = 1.061405429\n  const p = 0.3275911\n  const sign = z < 0 ? -1 : 1\n  const x = Math.abs(z) / Math.SQRT2\n  const t = 1 / (1 + p * x)\n  const y = 1 - ((((a5 * t + a4) * t + a3) * t + a2) * t + a1) * t * Math.exp(-x * x)\n  return 0.5 * (1 + sign * y)\n}\n\n// ── Composite + split selection ─────────────────────────────────────\n\nfunction resolveSplit(\n  runs: RunRecord[],\n  pref: 'search' | 'holdout' | 'auto',\n): 'search' | 'holdout' {\n  if (pref !== 'auto') return pref\n  const hasHoldout = runs.some((r) => Number.isFinite(r.outcome.holdoutScore))\n  return hasHoldout ? 'holdout' : 'search'\n}\n\nfunction compositeOf(run: RunRecord, split: 'search' | 'holdout'): number {\n  const primary = split === 'holdout' ? run.outcome.holdoutScore : run.outcome.searchScore\n  if (Number.isFinite(primary)) return primary as number\n  // Fall through to the other split if the preferred one is missing —\n  // analyzeRuns shouldn't refuse to summarise a run just because the\n  // caller asked for the split that wasn't recorded.\n  const alt = split === 'holdout' ? run.outcome.searchScore : run.outcome.holdoutScore\n  return Number.isFinite(alt) ? (alt as number) : Number.NaN\n}\n\n// ── Distribution helpers ────────────────────────────────────────────\n\nfunction distributionOf(\n  values: number[],\n  bins: number,\n  withIds?: Array<{ runId: string; score: number }>,\n): ScalarDistribution {\n  if (values.length === 0) {\n    return {\n      n: 0,\n      mean: 0,\n      p50: 0,\n      p95: 0,\n      stddev: 0,\n      min: 0,\n      max: 0,\n      histogram: [],\n    }\n  }\n  const sorted = [...values].sort((a, b) => a - b)\n  const n = sorted.length\n  const mean = sorted.reduce((s, v) => s + v, 0) / n\n  const variance = sorted.reduce((s, v) => s + (v - mean) ** 2, 0) / n\n  const stddev = Math.sqrt(variance)\n  const tailRuns = withIds\n    ? [...withIds].sort((a, b) => a.score - b.score).slice(0, Math.min(5, withIds.length))\n    : undefined\n  return {\n    n,\n    mean,\n    p50: percentile(sorted, 0.5),\n    p95: percentile(sorted, 0.95),\n    stddev,\n    min: sorted[0]!,\n    max: sorted[n - 1]!,\n    histogram: histogram(sorted, bins),\n    ...(tailRuns ? { tailRuns } : {}),\n  }\n}\n\nfunction percentile(sorted: number[], q: number): number {\n  if (sorted.length === 0) return 0\n  if (sorted.length === 1) return sorted[0]!\n  const idx = (sorted.length - 1) * q\n  const lo = Math.floor(idx)\n  const hi = Math.ceil(idx)\n  if (lo === hi) return sorted[lo]!\n  const w = idx - lo\n  return sorted[lo]! * (1 - w) + sorted[hi]! * w\n}\n\n/** Even-width histogram over the value range. Returns inclusive-lo /\n *  exclusive-hi bins (closed on right for the last bin) compatible with\n *  the substrate's `GainDistributionBin` shape. */\nfunction histogram(sorted: number[], bins: number): ScalarDistribution['histogram'] {\n  if (sorted.length === 0 || bins < 1) return []\n  const min = sorted[0]!\n  const max = sorted[sorted.length - 1]!\n  if (min === max) return [{ lo: min, hi: max, count: sorted.length }]\n  const width = (max - min) / bins\n  const out: ScalarDistribution['histogram'] = []\n  for (let i = 0; i < bins; i++) {\n    const lo = min + i * width\n    const hi = i === bins - 1 ? max : lo + width\n    out.push({ lo, hi, count: 0 })\n  }\n  for (const v of sorted) {\n    const idx = Math.min(bins - 1, Math.floor((v - min) / width))\n    out[idx]!.count++\n  }\n  return out\n}\n\nfunction computePerDimension(runs: RunRecord[], bins: number): Record<string, ScalarDistribution> {\n  // JudgeScoresRecord pre-aggregates `perDimMean` (mean across judges per\n  // dimension). We collect those means across runs to produce a per-dim\n  // distribution at the corpus level. Consumers who want per-judge\n  // dimension values reach into `perJudge[judgeId][dim]` themselves.\n  const byDim = new Map<string, number[]>()\n  for (const run of runs) {\n    const scores = run.outcome.judgeScores\n    if (!scores) continue\n    for (const [dim, value] of Object.entries(scores.perDimMean ?? {})) {\n      if (!Number.isFinite(value)) continue\n      const arr = byDim.get(dim) ?? []\n      arr.push(value)\n      byDim.set(dim, arr)\n    }\n  }\n  const out: Record<string, ScalarDistribution> = {}\n  for (const [dim, values] of byDim) out[dim] = distributionOf(values, bins)\n  return out\n}\n\n// ── Judge insights ──────────────────────────────────────────────────\n\nfunction computeJudgeInsights(runs: RunRecord[]): Record<string, JudgeInsight> {\n  // Each judge's per-run mean is the average of its per-dimension scores\n  // for that run. We aggregate those means across all runs each judge\n  // scored — giving consumers a \"this judge's typical verdict\" reading.\n  const out: Record<string, JudgeInsight> = {}\n  const byJudge = new Map<string, number[]>()\n  for (const run of runs) {\n    const scores = run.outcome.judgeScores\n    if (!scores?.perJudge) continue\n    for (const [judgeId, dims] of Object.entries(scores.perJudge)) {\n      const dimValues = Object.values(dims).filter(Number.isFinite) as number[]\n      if (dimValues.length === 0) continue\n      const judgeMean = dimValues.reduce((s, v) => s + v, 0) / dimValues.length\n      const arr = byJudge.get(judgeId) ?? []\n      arr.push(judgeMean)\n      byJudge.set(judgeId, arr)\n    }\n  }\n  for (const [judgeId, values] of byJudge) {\n    out[judgeId] = {\n      n: values.length,\n      meanScore: values.reduce((s, v) => s + v, 0) / values.length,\n    }\n  }\n  return out\n}\n\n// ── Inter-rater agreement ───────────────────────────────────────────\n\nfunction computeInterRater(\n  ratings: Array<{ runId: string; rater: string; score: number }>,\n): InterRaterInsight | undefined {\n  const byRun = new Map<string, Array<{ rater: string; score: number }>>()\n  for (const r of ratings) {\n    if (!Number.isFinite(r.score)) continue\n    const list = byRun.get(r.runId) ?? []\n    list.push({ rater: r.rater, score: r.score })\n    byRun.set(r.runId, list)\n  }\n  const raters = new Set(ratings.map((r) => r.rater))\n  const jointlyRated: string[] = []\n  for (const [runId, ratersForRun] of byRun) {\n    const seen = new Set(ratersForRun.map((r) => r.rater))\n    let all = true\n    for (const r of raters) if (!seen.has(r)) all = false\n    if (all) jointlyRated.push(runId)\n  }\n  if (raters.size < 2 || jointlyRated.length === 0) return undefined\n\n  const raterList = [...raters].sort()\n  const perPair: Record<string, number> = {}\n  for (let i = 0; i < raterList.length; i++) {\n    for (let j = i + 1; j < raterList.length; j++) {\n      const a = raterList[i]!\n      const b = raterList[j]!\n      const aScores: number[] = []\n      const bScores: number[] = []\n      for (const runId of jointlyRated) {\n        const ratersForRun = byRun.get(runId)!\n        const sa = ratersForRun.find((r) => r.rater === a)?.score\n        const sb = ratersForRun.find((r) => r.rater === b)?.score\n        if (sa !== undefined && sb !== undefined) {\n          aScores.push(sa)\n          bScores.push(sb)\n        }\n      }\n      perPair[`${a}::${b}`] = pearson(aScores, bScores)\n    }\n  }\n  const pairKappas = Object.values(perPair)\n  const kappa =\n    pairKappas.length === 0 ? 0 : pairKappas.reduce((s, v) => s + v, 0) / pairKappas.length\n\n  const disagreementCases = jointlyRated\n    .map((runId) => {\n      const ratersForRun = byRun.get(runId)!\n      const scores = ratersForRun.map((r) => r.score)\n      const range = Math.max(...scores) - Math.min(...scores)\n      return { runId, ratings: ratersForRun, range }\n    })\n    .sort((a, b) => b.range - a.range)\n    .slice(0, 20)\n\n  return {\n    raters: raters.size,\n    jointlyRated: jointlyRated.length,\n    kappa,\n    perPair,\n    disagreementCases,\n  }\n}\n\nfunction pearson(a: number[], b: number[]): number {\n  if (a.length !== b.length || a.length === 0) return 0\n  const n = a.length\n  const meanA = a.reduce((s, v) => s + v, 0) / n\n  const meanB = b.reduce((s, v) => s + v, 0) / n\n  let num = 0\n  let denomA = 0\n  let denomB = 0\n  for (let i = 0; i < n; i++) {\n    const da = a[i]! - meanA\n    const db = b[i]! - meanB\n    num += da * db\n    denomA += da * da\n    denomB += db * db\n  }\n  const denom = Math.sqrt(denomA * denomB)\n  return denom === 0 ? 0 : num / denom\n}\n\n// ── Lift ────────────────────────────────────────────────────────────\n\nfunction computeLift(\n  runs: RunRecord[],\n  baselineId: string | undefined,\n  candidateId: string | undefined,\n  split: 'search' | 'holdout',\n): LiftInsight | undefined {\n  let bId = baselineId\n  let cId = candidateId\n  if (!bId || !cId) {\n    // Auto-detect: when exactly two distinct candidateIds appear, treat the\n    // lower-mean side as baseline.\n    const ids = [...new Set(runs.map((r) => r.candidateId))]\n    if (ids.length !== 2) return undefined\n    const [idA, idB] = ids as [string, string]\n    const meanA = mean(runs.filter((r) => r.candidateId === idA).map((r) => compositeOf(r, split)))\n    const meanB = mean(runs.filter((r) => r.candidateId === idB).map((r) => compositeOf(r, split)))\n    bId = meanA <= meanB ? idA : idB\n    cId = meanA <= meanB ? idB : idA\n  }\n\n  const baseline = runs.filter((r) => r.candidateId === bId)\n  const candidate = runs.filter((r) => r.candidateId === cId)\n  if (baseline.length === 0 || candidate.length === 0) return undefined\n\n  // Pair on (experimentId, seed). When that key doesn't match, fall back\n  // to ordinal pairing — common for fresh runs from the same scenario list.\n  const baselineByKey = new Map(baseline.map((r) => [pairingKey(r), r]))\n  const pairedBaseline: number[] = []\n  const pairedCandidate: number[] = []\n  let usedKeyPairing = false\n  for (const cand of candidate) {\n    const b = baselineByKey.get(pairingKey(cand))\n    if (b) {\n      const bC = compositeOf(b, split)\n      const cC = compositeOf(cand, split)\n      if (Number.isFinite(bC) && Number.isFinite(cC)) {\n        pairedBaseline.push(bC)\n        pairedCandidate.push(cC)\n        usedKeyPairing = true\n      }\n    }\n  }\n  if (!usedKeyPairing) {\n    const n = Math.min(baseline.length, candidate.length)\n    for (let i = 0; i < n; i++) {\n      const bC = compositeOf(baseline[i]!, split)\n      const cC = compositeOf(candidate[i]!, split)\n      if (Number.isFinite(bC) && Number.isFinite(cC)) {\n        pairedBaseline.push(bC)\n        pairedCandidate.push(cC)\n      }\n    }\n  }\n  if (pairedBaseline.length === 0) return undefined\n\n  const baselineMean = mean(pairedBaseline)\n  const candidateMean = mean(pairedCandidate)\n  const delta = candidateMean - baselineMean\n\n  const bootstrap = pairedBootstrap(pairedBaseline, pairedCandidate, {\n    confidence: 0.95,\n    resamples: 2000,\n    statistic: 'mean',\n  })\n  const tTest = pairedTTest(pairedBaseline, pairedCandidate)\n  const d = cohensD(pairedBaseline, pairedCandidate)\n  const mde = pairedMde({ nPaired: pairedBaseline.length, power: 0.8, alpha: 0.05 })\n  const requiredN = requiredSampleSize({\n    effect: Math.max(Math.abs(delta), 1e-6),\n    power: 0.8,\n    alpha: 0.05,\n  })\n\n  return {\n    baselineMean,\n    candidateMean,\n    delta,\n    ci95: [bootstrap.low, bootstrap.high],\n    pValue: tTest.p,\n    n: pairedBaseline.length,\n    cohensD: d,\n    mde,\n    requiredN,\n  }\n}\n\nfunction pairingKey(r: RunRecord): string {\n  return `${r.experimentId}::${r.seed}`\n}\n\nfunction mean(arr: number[]): number {\n  return arr.length === 0 ? 0 : arr.reduce((s, v) => s + v, 0) / arr.length\n}\n\n// ── Failure clustering ──────────────────────────────────────────────\n\nasync function computeFailureClusters(\n  runs: RunRecord[],\n  analyst: AnalystRegistry,\n  split: 'search' | 'holdout',\n): Promise<FailureClusterInsight | undefined> {\n  const failed = runs.filter((r) => compositeOf(r, split) < 0.5 || r.failureMode !== undefined)\n  if (failed.length === 0) return { clusters: [], totalFailures: 0 }\n\n  const clusters = new Map<string, { exemplars: string[]; share: number }>()\n  for (const run of failed) {\n    try {\n      const result = await analyst.run(run.runId, {\n        kind: 'run-record',\n        run,\n      } as Parameters<typeof analyst.run>[1])\n      for (const finding of result.findings as AnalystFinding[]) {\n        const key = finding.area || finding.analyst_id || 'unclassified'\n        const c = clusters.get(key) ?? { exemplars: [], share: 0 }\n        if (c.exemplars.length < 5) c.exemplars.push(run.runId)\n        clusters.set(key, c)\n      }\n    } catch {\n      const c = clusters.get('analyst-error') ?? { exemplars: [], share: 0 }\n      if (c.exemplars.length < 5) c.exemplars.push(run.runId)\n      clusters.set('analyst-error', c)\n    }\n  }\n  const clusterList = [...clusters.entries()].map(([id, c]) => ({\n    id,\n    name: id,\n    share: c.exemplars.length / failed.length,\n    exemplars: c.exemplars,\n  }))\n  clusterList.sort((a, b) => b.share - a.share)\n  return { clusters: clusterList, totalFailures: failed.length }\n}\n\n// ── Contamination ──────────────────────────────────────────────────\n\nfunction computeContamination(\n  runs: RunRecord[],\n  canaries: DatasetScenario[],\n): InsightReport['contamination'] {\n  let leaks = 0\n  const details: Array<{ runId: string; canary: string; matched: string }> = []\n  for (const run of runs) {\n    const output = stringifyOutput(run)\n    if (!output) continue\n    const leaksHere = checkCanaries(output, canaries)\n    for (const leak of leaksHere) {\n      leaks++\n      details.push({ runId: run.runId, canary: leak.canary, matched: leak.evidence })\n    }\n  }\n  return { leaks, holdoutAuditPassed: leaks === 0, details }\n}\n\nfunction stringifyOutput(run: RunRecord): string | undefined {\n  // RunRecord doesn't fix where \"the agent's output\" lives — different\n  // consumers stash it differently. We probe the common shapes: the\n  // outcome.raw map (numeric only by design — unlikely to contain text),\n  // and any string-valued fields tucked under metadata via type casting.\n  // Consumers with bespoke shapes pass canaryScenarios only when they\n  // know their runs carry a stringifiable surface.\n  const metadata = (run as unknown as { metadata?: Record<string, unknown> }).metadata\n  if (typeof metadata?.output === 'string') return metadata.output\n  if (typeof metadata?.text === 'string') return metadata.text\n  return undefined\n}\n\n// ── Outcome correlation + linear reward model ──────────────────────\n\nfunction computeOutcomeCorrelation(\n  runs: RunRecord[],\n  outcome: { metric: string; valueByRunId: Record<string, number> },\n  split: 'search' | 'holdout',\n): OutcomeCorrelationInsight | undefined {\n  const xs: number[] = []\n  const ys: number[] = []\n  for (const run of runs) {\n    const y = outcome.valueByRunId[run.runId]\n    if (y === undefined || !Number.isFinite(y)) continue\n    const x = compositeOf(run, split)\n    if (!Number.isFinite(x)) continue\n    xs.push(x)\n    ys.push(y)\n  }\n  if (xs.length < 3) return undefined\n\n  const p = pearson(xs, ys)\n  const s = spearman(xs, ys)\n  const meanX = mean(xs)\n  const meanY = mean(ys)\n  let num = 0\n  let denom = 0\n  for (let i = 0; i < xs.length; i++) {\n    num += (xs[i]! - meanX) * (ys[i]! - meanY)\n    denom += (xs[i]! - meanX) ** 2\n  }\n  const slope = denom === 0 ? 0 : num / denom\n  const intercept = meanY - slope * meanX\n  const ssTot = ys.reduce((a, y) => a + (y - meanY) ** 2, 0)\n  const ssRes = ys.reduce((a, y, i) => a + (y - (intercept + slope * xs[i]!)) ** 2, 0)\n  const r2 = ssTot === 0 ? 0 : 1 - ssRes / ssTot\n\n  return {\n    metric: outcome.metric,\n    n: xs.length,\n    pearson: p,\n    spearman: s,\n    rewardModel: { intercept, slope, r2 },\n  }\n}\n\nfunction spearman(a: number[], b: number[]): number {\n  if (a.length !== b.length || a.length === 0) return 0\n  return pearson(rank(a), rank(b))\n}\n\nfunction rank(arr: number[]): number[] {\n  const indexed = arr.map((v, i) => ({ v, i }))\n  indexed.sort((x, y) => x.v - y.v)\n  const ranks = new Array(arr.length).fill(0)\n  let i = 0\n  while (i < indexed.length) {\n    let j = i\n    while (j + 1 < indexed.length && indexed[j + 1]!.v === indexed[i]!.v) j++\n    const avg = (i + j + 2) / 2\n    for (let k = i; k <= j; k++) ranks[indexed[k]!.i] = avg\n    i = j + 1\n  }\n  return ranks\n}\n\n// ── Release confidence scorecard ───────────────────────────────────\n\nfunction buildReleaseScorecard(\n  composite: ScalarDistribution,\n  lift: LiftInsight | undefined,\n  contamination: InsightReport['contamination'],\n): InsightReport['release'] {\n  // Synthesise a minimal scorecard from the rolled-up signal. The\n  // substrate's `evaluateReleaseConfidence` primitive consumes a richer\n  // input shape that callers can produce by wiring SLO definitions; the\n  // shape here is the contract `selfImprove`/`analyzeRuns` consumers\n  // receive automatically. They can call `evaluateReleaseConfidence`\n  // directly when they want SLO-based axis evaluation.\n  const axes: InsightReport['release']['axes'] = []\n  const liftPass =\n    lift === undefined || lift.ci95[0] > 0\n      ? ('pass' as const)\n      : lift.delta > 0\n        ? ('warn' as const)\n        : ('fail' as const)\n  axes.push({\n    name: 'quality-lift',\n    status: liftPass,\n    detail: lift\n      ? `delta=${lift.delta.toFixed(3)}, CI95=[${lift.ci95[0].toFixed(3)}, ${lift.ci95[1].toFixed(3)}], n=${lift.n}`\n      : 'no baseline/candidate pair available',\n  })\n  const contamPass =\n    contamination === undefined || contamination.leaks === 0 ? ('pass' as const) : ('fail' as const)\n  axes.push({\n    name: 'contamination',\n    status: contamPass,\n    detail: contamination ? `${contamination.leaks} canary leak(s)` : 'no canaries supplied',\n  })\n  axes.push({\n    name: 'composite-distribution',\n    status: composite.mean >= 0.5 ? 'pass' : composite.mean >= 0.3 ? 'warn' : 'fail',\n    detail: `mean=${composite.mean.toFixed(3)}, p50=${composite.p50.toFixed(3)}, p95=${composite.p95.toFixed(3)} over n=${composite.n}`,\n  })\n  const status = axes.some((a) => a.status === 'fail')\n    ? 'fail'\n    : axes.some((a) => a.status === 'warn')\n      ? 'warn'\n      : 'pass'\n  return {\n    status,\n    axes,\n    issues: [],\n  }\n}\n\n// ── Recommendations engine ─────────────────────────────────────────\n\ninterface RecommendationContext {\n  composite: ScalarDistribution\n  judges: Record<string, JudgeInsight>\n  interRater?: InterRaterInsight\n  lift?: LiftInsight\n  failureClusters?: FailureClusterInsight\n  contamination?: InsightReport['contamination']\n  outcomeCorrelation?: OutcomeCorrelationInsight\n  priorPeriodComparison?: PriorPeriodComparison\n  threshold: number\n}\n\nfunction buildRecommendations(ctx: RecommendationContext): Recommendation[] {\n  const out: Recommendation[] = []\n\n  // Prior-period regressions — highest customer-impact signal when present.\n  // \"Did my last change help?\" with a falsifiable answer.\n  if (ctx.priorPeriodComparison) {\n    const ppc = ctx.priorPeriodComparison\n    const label = ppc.windowLabel ?? 'baseline period'\n    for (const name of ppc.regressedMetrics) {\n      const d = ppc.metrics[name]\n      if (!d) continue\n      out.push({\n        priority: 'critical',\n        kind: 'investigate',\n        title: `${name} regressed from ${d.baseline.toFixed(3)} → ${d.current.toFixed(3)} vs ${label}`,\n        detail: `Welch CI95 = [${d.ci95[0].toFixed(3)}, ${d.ci95[1].toFixed(3)}], p=${d.pValue.toFixed(4)}, Cohen's d=${d.cohensD.toFixed(2)} (n_current=${d.currentN}, n_baseline=${d.baselineN}). The regression is statistically significant at p<0.05 with at-least-small effect size.`,\n        evidencePath: `priorPeriodComparison.metrics.${name}`,\n      })\n    }\n    for (const name of ppc.improvedMetrics) {\n      const d = ppc.metrics[name]\n      if (!d) continue\n      out.push({\n        priority: 'low',\n        kind: 'ship',\n        title: `${name} improved from ${d.baseline.toFixed(3)} → ${d.current.toFixed(3)} vs ${label}`,\n        detail: `Welch CI95 = [${d.ci95[0].toFixed(3)}, ${d.ci95[1].toFixed(3)}], p=${d.pValue.toFixed(4)}, Cohen's d=${d.cohensD.toFixed(2)} (n_current=${d.currentN}, n_baseline=${d.baselineN}). Statistically significant improvement worth flagging.`,\n        evidencePath: `priorPeriodComparison.metrics.${name}`,\n      })\n    }\n  }\n\n  // Composite-distribution branch. Fires when the overall quality signal is\n  // poor regardless of lift / contamination / clusters — the customer needs\n  // to know they have a problem AND which specific runs to inspect.\n  if (ctx.composite.n > 0) {\n    if (ctx.composite.mean < 0.3) {\n      const tail = ctx.composite.tailRuns ?? []\n      const names = tail\n        .slice(0, 5)\n        .map((t) => `${t.runId}=${t.score.toFixed(3)}`)\n        .join(', ')\n      out.push({\n        priority: 'critical',\n        kind: 'investigate',\n        title: `Composite mean ${ctx.composite.mean.toFixed(3)} is below the 0.3 floor — the agent is broken on this corpus`,\n        detail:\n          tail.length > 0\n            ? `Worst ${tail.length} run${tail.length === 1 ? '' : 's'} to inspect first: ${names}. Histogram p50=${ctx.composite.p50.toFixed(3)}, p95=${ctx.composite.p95.toFixed(3)}.`\n            : `Histogram p50=${ctx.composite.p50.toFixed(3)}, p95=${ctx.composite.p95.toFixed(3)}.`,\n        evidencePath: 'composite.tailRuns',\n      })\n    } else if (ctx.composite.mean < 0.5) {\n      const tail = ctx.composite.tailRuns ?? []\n      const names = tail\n        .slice(0, 3)\n        .map((t) => `${t.runId}=${t.score.toFixed(3)}`)\n        .join(', ')\n      out.push({\n        priority: 'high',\n        kind: 'investigate',\n        title: `Composite mean ${ctx.composite.mean.toFixed(3)} is below 0.5 — investigate the lower tail before claiming the agent is healthy`,\n        detail:\n          tail.length > 0\n            ? `Worst ${tail.length} run${tail.length === 1 ? '' : 's'}: ${names}. Histogram p50=${ctx.composite.p50.toFixed(3)}, p95=${ctx.composite.p95.toFixed(3)}.`\n            : `Histogram p50=${ctx.composite.p50.toFixed(3)}, p95=${ctx.composite.p95.toFixed(3)}.`,\n        evidencePath: 'composite.tailRuns',\n      })\n    }\n  }\n\n  // Missing-judges branch. The report can't surface per-dimension or\n  // calibration signal when `outcome.judgeScores` is empty across the\n  // corpus. Tell the customer how to enrich.\n  if (Object.keys(ctx.judges).length === 0 && ctx.composite.n > 0) {\n    out.push({\n      priority: 'medium',\n      kind: 'expand-corpus',\n      title: 'No judge scores recorded — per-dimension + calibration insights unavailable',\n      detail:\n        'Records have no `outcome.judgeScores`. To unlock perDimension, judges, and calibration, attach a Judge run during your eval pass and populate `outcome.judgeScores.perJudge[judgeName][dimension] = score`. See `docs/insight-report.md` for the expected shape.',\n      evidencePath: 'judges',\n    })\n  }\n\n  if (ctx.lift) {\n    const decisive = ctx.lift.ci95[0] > ctx.threshold\n    const inconclusive = ctx.lift.ci95[0] <= ctx.threshold && ctx.lift.ci95[1] > ctx.threshold\n    if (decisive) {\n      out.push({\n        priority: 'critical',\n        kind: 'ship',\n        title: `Ship — lift ${ctx.lift.delta.toFixed(3)} (95% CI ${ctx.lift.ci95[0].toFixed(3)}..${ctx.lift.ci95[1].toFixed(3)})`,\n        detail: `Holdout lift exceeds threshold ${ctx.threshold} with 95% bootstrap confidence (n=${ctx.lift.n}, p=${ctx.lift.pValue.toFixed(4)}, d=${ctx.lift.cohensD.toFixed(2)}).`,\n        evidencePath: 'lift',\n      })\n    } else if (inconclusive) {\n      out.push({\n        priority: 'high',\n        kind: 'expand-corpus',\n        title: `Inconclusive — need ~${ctx.lift.requiredN} paired runs (have ${ctx.lift.n}) at current effect size`,\n        detail: `CI straddles threshold. Current MDE at 80% power is ${ctx.lift.mde.toFixed(3)}; observed delta is ${ctx.lift.delta.toFixed(3)}.`,\n        evidencePath: 'lift',\n      })\n    } else {\n      out.push({\n        priority: 'critical',\n        kind: 'hold',\n        title: `Hold — lift CI lower bound ${ctx.lift.ci95[0].toFixed(3)} is at or below threshold ${ctx.threshold}`,\n        detail: `Bootstrap CI provides no statistical evidence the candidate is better. Consider tightening the mutation or expanding the holdout.`,\n        evidencePath: 'lift',\n      })\n    }\n  }\n\n  if (ctx.contamination && ctx.contamination.leaks > 0) {\n    out.push({\n      priority: 'critical',\n      kind: 'fix',\n      title: `${ctx.contamination.leaks} canary leak${ctx.contamination.leaks === 1 ? '' : 's'} detected`,\n      detail: `Holdout integrity is compromised. The lift number is unreliable until you investigate.`,\n      evidencePath: 'contamination',\n    })\n  }\n\n  if (ctx.interRater && ctx.interRater.kappa < 0.5) {\n    out.push({\n      priority: 'high',\n      kind: 'recalibrate',\n      title: `Inter-rater agreement κ=${ctx.interRater.kappa.toFixed(2)} is below 0.5`,\n      detail: `Raters disagree on what 'good' looks like. Top disagreement cases listed in interRater.disagreementCases — consider a triage meeting or refining the rubric.`,\n      evidencePath: 'interRater',\n    })\n  }\n\n  if (ctx.failureClusters && ctx.failureClusters.clusters.length > 0) {\n    const top = ctx.failureClusters.clusters[0]!\n    out.push({\n      priority: 'high',\n      kind: 'investigate',\n      title: `Top failure cluster: ${top.name} (${(top.share * 100).toFixed(0)}% of failures)`,\n      detail: `${ctx.failureClusters.totalFailures} runs failed. The largest cluster groups ${top.exemplars.length} exemplars under '${top.name}'.`,\n      evidencePath: 'failureClusters.clusters[0]',\n    })\n  }\n\n  if (ctx.outcomeCorrelation && Math.abs(ctx.outcomeCorrelation.spearman) < 0.3) {\n    out.push({\n      priority: 'medium',\n      kind: 'recalibrate',\n      title: `Judge scores decoupled from ${ctx.outcomeCorrelation.metric} (Spearman ρ=${ctx.outcomeCorrelation.spearman.toFixed(2)})`,\n      detail: `Your judges score what they were trained to score, but it isn't predicting downstream ${ctx.outcomeCorrelation.metric}. Consider retraining the judge against ${ctx.outcomeCorrelation.metric} as the gold signal.`,\n      evidencePath: 'outcomeCorrelation',\n    })\n  }\n\n  return out\n}\n\n// ── Re-export pareto figure spec for hosted-side rendering ─────────\n\nexport type { ParetoFigureSpec }\n","/**\n * # `selfImprove()` — the LAND-tier one-shot.\n *\n * The cheapest possible call site to run a real closed-loop self-\n * improvement over your agent. Wraps `runImprovementLoop` with smart\n * defaults and a budget-shaped options API; every escape hatch the\n * substrate exposes is reachable from here without losing the\n * one-function feel.\n *\n * Defaults picked to match the LAND-tier story:\n *   - In-memory storage (no filesystem touch).\n *   - `gepaDriver` reflective mutation with copywriting-flavored primitives\n *     (override `driver` or `mutationPrimitives` for any domain).\n *   - `defaultProductionGate` with `deltaThreshold: 0.05`.\n *   - Held-out split = 25% of scenarios, deterministic by id hash.\n *   - 3 generations × population 2 (raise via `budget` for more search).\n *   - `autoOnPromote: 'none'` (we don't open PRs unless you ask).\n *\n * Want one-click? Provide `agent` + `scenarios` + `judge`. Done.\n * Want distributed? Pass `cellPlacement` + an `httpDispatch`-backed\n * agent. Want a code-tier surface? Pass a `MutableSurface` + your own\n * `driver`. Same function.\n */\n\nimport { gepaDriver } from '../campaign/drivers/gepa'\nimport { defaultProductionGate } from '../campaign/gates/default-production-gate'\nimport {\n  type RunImprovementLoopResult,\n  runImprovementLoop,\n} from '../campaign/presets/run-improvement-loop'\nimport { type CampaignStorage, inMemoryCampaignStorage } from '../campaign/storage'\nimport type {\n  CampaignCellResult,\n  DispatchContext,\n  Gate,\n  ImprovementDriver,\n  JudgeConfig,\n  MutableSurface,\n  Scenario,\n} from '../campaign/types'\nimport { createHostedClient, type HostedTenant } from '../hosted/client'\nimport type { EvalRunCellScore, EvalRunEvent, EvalRunGenerationSnapshot } from '../hosted/types'\nimport type { JudgeScoresRecord, RunRecord } from '../run-record'\nimport { analyzeRuns } from './analyze-runs'\nimport type { InsightReport } from './insight-report'\n\nexport interface SelfImproveBudget {\n  /** Hard $ ceiling across all cells in baseline + every generation. Cells\n   *  beyond the ceiling are skipped (cost-aware, not aborted). */\n  dollars?: number\n  /** How many improvement generations to explore. Default 3. Set 0 to\n   *  skip improvement entirely (selfImprove becomes a baseline-only run). */\n  generations?: number\n  /** Candidates the driver proposes per generation. Default 2. */\n  populationSize?: number\n  /** Max concurrent cells across the loop. Default 2. */\n  maxConcurrency?: number\n  /** Fraction of `scenarios` held out from training, used for the gate.\n   *  Default 0.25. Ignored when `holdoutScenarios` is set explicitly. */\n  holdoutFraction?: number\n  /** Explicit held-out scenarios; overrides `holdoutFraction`. */\n  holdoutScenarios?: Scenario[]\n}\n\nexport interface SelfImproveLlm {\n  /** Endpoint base URL. Default Tangle Router. */\n  baseUrl?: string\n  /** Bearer token. Default `process.env.OPENAI_API_KEY`. */\n  apiKey?: string\n  /** Model id used by `gepaDriver` reflection. Default\n   *  `anthropic/claude-sonnet-4.6`. */\n  model?: string\n}\n\nexport type SelfImproveProgressEvent =\n  | { kind: 'baseline.started'; scenarios: number }\n  | { kind: 'baseline.completed'; compositeMean: number; durationMs: number }\n  | { kind: 'generation.started'; index: number; populationSize: number }\n  | { kind: 'generation.completed'; index: number; bestComposite: number; durationMs: number }\n  | { kind: 'gate.decided'; decision: string; lift: number }\n\nexport interface SelfImproveOptions<TScenario extends Scenario, TArtifact> {\n  /**\n   * Your agent — a function that takes the current `MutableSurface`\n   * (typically a system prompt the loop is optimizing) plus the\n   * scenario + cell ctx, and returns the artifact your judge scores.\n   *\n   * Same shape as `RunOptimizationOptions.dispatchWithSurface`. Wrap a\n   * plain `Dispatch` if you don't have a surface seam:\n   *\n   *   agent: (_surface, scenario, ctx) => yourPlainDispatch(scenario, ctx)\n   *\n   * That mode evaluates without mutating any surface — useful as a\n   * baseline-only run (set `budget.generations = 0`).\n   */\n  agent: (surface: MutableSurface, scenario: TScenario, ctx: DispatchContext) => Promise<TArtifact>\n\n  /** Scenarios to evaluate against. Train/holdout split is computed from\n   *  these unless `budget.holdoutScenarios` is set explicitly. */\n  scenarios: TScenario[]\n\n  /** Judge that scores artifacts. Bring your own; use `langchainJudge`\n   *  from `/adapters/langchain` for a Runnable-shaped one. */\n  judge: JudgeConfig<TArtifact, TScenario>\n\n  /** Starting surface — system prompt, JSON config, anything `MutableSurface`\n   *  accepts. The driver mutates this each generation. */\n  baselineSurface: MutableSurface\n\n  /** Budget + loop shape. All fields optional; defaults pick the LAND-tier\n   *  story. */\n  budget?: SelfImproveBudget\n\n  /** Custom driver. Default is `gepaDriver` configured from `llm` +\n   *  `mutationPrimitives`. */\n  driver?: ImprovementDriver\n\n  /** Default-driver overrides — used when `driver` is unset. */\n  mutationPrimitives?: string[]\n  driverTarget?: string\n\n  /** Custom gate. Default is `defaultProductionGate` with\n   *  `deltaThreshold: 0.05` on the held-out split. */\n  gate?: Gate<TArtifact, TScenario>\n\n  /** LLM config consumed by the default `gepaDriver`. Ignored if you pass\n   *  your own `driver`. */\n  llm?: SelfImproveLlm\n\n  /** Storage backend. Default `inMemoryCampaignStorage()` — nothing\n   *  persists past the call. Pass `fsCampaignStorage()` to write to disk. */\n  storage?: CampaignStorage\n\n  /** Run directory (logical for in-memory storage, real path for fs).\n   *  Default `mem://selfImprove-<timestamp>`. */\n  runDir?: string\n\n  /** Distributed-driver seam — same as `RunCampaignOptions.cellPlacement`.\n   *  Returns an opaque placement key the substrate forwards to your agent\n   *  as `ctx.placement`. Combined with `httpDispatch` from\n   *  `/adapters/http`, fans cells across regions. */\n  cellPlacement?: (input: {\n    scenario: TScenario\n    rep: number\n    generation?: number\n  }) => string | undefined\n\n  /** Streaming hook — fires on baseline + each generation + gate decision.\n   *  Consumer routes events wherever (UI, dashboard, logs). */\n  onProgress?: (event: SelfImproveProgressEvent) => void\n\n  /** Auto-promotion behavior on a ship decision. Default `'none'` — we\n   *  return the winner; you ship it however you ship. `'pr'` opens a\n   *  GitHub PR via `openAutoPr`; requires `ghOwner` + `ghRepo`. */\n  autoOnPromote?: 'pr' | 'none'\n  ghOwner?: string\n  ghRepo?: string\n\n  /**\n   * Opt-in: ship eval-run events to a hosted orchestrator (ours, your\n   * self-hosted one, or any compatible implementation of the\n   * `docs/hosted-ingest-spec.md` wire format). When set, the substrate\n   * POSTs the final `EvalRunEvent` to `${endpoint}/v1/ingest/eval-runs`\n   * after the loop completes. Failures are logged but do not fail the\n   * loop — local result is always returned.\n   *\n   * For our orchestrator: `{ endpoint: 'https://orchestrator.tangle.tools/v1', apiKey, tenantId }`.\n   *\n   * For your self-hosted: any URL serving the wire format. See\n   * `examples/hosted-ingest-server/` for the reference receiver.\n   */\n  hostedTenant?: HostedTenant\n\n  /** Free-form labels attached to the hosted event (env, branch, model id,\n   *  etc.). Ignored when `hostedTenant` is unset. */\n  hostedLabels?: Record<string, string>\n}\n\nexport interface SelfImproveResult<TScenario extends Scenario, TArtifact> {\n  /** Composite mean across all scenarios, baseline run. */\n  baseline: {\n    compositeMean: number\n    perScenario: Record<string, number>\n  }\n  /** Composite mean on the held-out set, winner run. */\n  winner: {\n    compositeMean: number\n    perScenario: Record<string, number>\n    surface: MutableSurface\n  }\n  /** `winner.compositeMean - baselineOnHoldout.compositeMean`. Positive\n   *  means the gate observed improvement. */\n  lift: number\n  /** `defaultProductionGate.decide()` result. */\n  gateDecision: 'ship' | 'hold' | 'need_more_work' | 'model_ceiling' | 'arch_ceiling'\n  /** Number of generations actually explored (may be less than the\n   *  budget if the driver gave up early). */\n  generationsExplored: number\n  /** Wall-clock total. */\n  durationMs: number\n  /** Total cost across baseline + every generation. */\n  totalCostUsd: number\n  /**\n   * Rigor packet: distributional summary, paired-bootstrap lift CI,\n   * judge stats, contamination check, recommendations. Wired through\n   * `analyzeRuns()` on the baseline + winner cells of the campaign.\n   * Hosted-tier dashboards render this as the v3-vs-v4 decision view.\n   */\n  insight: InsightReport\n  /**\n   * Raw substrate result for advanced inspection — full per-generation\n   * candidates, full campaign artifacts, all judge scores. Useful for\n   * debugging or reporting beyond the summary.\n   */\n  raw: RunImprovementLoopResult<TArtifact, TScenario>\n}\n\n/**\n * Deterministic train/holdout split by a stable hash of `scenario.id`,\n * so the same scenario set always splits the same way across runs.\n */\nfunction splitTrainHoldout<TScenario extends Scenario>(\n  scenarios: TScenario[],\n  fraction: number,\n): { train: TScenario[]; holdout: TScenario[] } {\n  // Stable fnv-1a-ish hash of the id for ordering.\n  function hash(s: string): number {\n    let h = 2166136261 >>> 0\n    for (let i = 0; i < s.length; i++) {\n      h ^= s.charCodeAt(i)\n      h = Math.imul(h, 16777619) >>> 0\n    }\n    return h\n  }\n  const sorted = [...scenarios].sort((a, b) => hash(a.id) - hash(b.id))\n  const nHoldout = Math.max(1, Math.min(sorted.length - 1, Math.round(sorted.length * fraction)))\n  return {\n    holdout: sorted.slice(0, nHoldout),\n    train: sorted.slice(nHoldout),\n  }\n}\n\nfunction meanComposite(byScenario: Record<string, { meanComposite: number }>): {\n  compositeMean: number\n  perScenario: Record<string, number>\n} {\n  const perScenario: Record<string, number> = {}\n  const values: number[] = []\n  for (const [id, agg] of Object.entries(byScenario)) {\n    perScenario[id] = agg.meanComposite\n    values.push(agg.meanComposite)\n  }\n  return {\n    compositeMean: values.length === 0 ? 0 : values.reduce((s, v) => s + v, 0) / values.length,\n    perScenario,\n  }\n}\n\nconst DEFAULT_MUTATION_PRIMITIVES = [\n  'Tighten the hook: lead with the specific user outcome.',\n  'Replace generic adjectives with specific verbs or proof numbers.',\n  \"Anchor every claim in something the scenario's brief literally supports.\",\n  'Honor the surface-shape constraint (length, register, audience vocabulary).',\n]\n\n/**\n * One-shot self-improvement loop. See module docstring for defaults +\n * extension points.\n *\n * @example Minimum (LAND tier):\n *\n *   const result = await selfImprove({\n *     agent: (surface, scenario, ctx) => myAgent(surface, scenario, ctx.signal),\n *     scenarios,\n *     judge,\n *     baselineSurface: DEFAULT_PROMPT,\n *   })\n *   console.log(`lift: ${result.lift.toFixed(3)} (${result.gateDecision})`)\n *\n * @example Distributed (workers in three regions):\n *\n *   await selfImprove({\n *     agent: httpDispatch({ resolveUrl: ({ placement }) => REGION_URLS[placement!] }),\n *     scenarios,\n *     judge,\n *     baselineSurface: DEFAULT_PROMPT,\n *     cellPlacement: ({ scenario }) => scenario.region,\n *     budget: { maxConcurrency: 12 },\n *   })\n */\nexport async function selfImprove<TScenario extends Scenario, TArtifact>(\n  opts: SelfImproveOptions<TScenario, TArtifact>,\n): Promise<SelfImproveResult<TScenario, TArtifact>> {\n  const startedAt = Date.now()\n\n  const budget = opts.budget ?? {}\n  const generations = budget.generations ?? 3\n  const populationSize = budget.populationSize ?? 2\n  const maxConcurrency = budget.maxConcurrency ?? 2\n  const holdoutFraction = budget.holdoutFraction ?? 0.25\n  const costCeiling = budget.dollars\n\n  const explicitHoldout = budget.holdoutScenarios\n  const { train, holdout } = explicitHoldout\n    ? {\n        train: opts.scenarios.filter((s) => !explicitHoldout.some((h) => h.id === s.id)),\n        holdout: explicitHoldout as TScenario[],\n      }\n    : splitTrainHoldout(opts.scenarios, holdoutFraction)\n\n  if (train.length === 0) {\n    throw new Error(\n      'selfImprove: train split is empty. Reduce holdoutFraction or pass more scenarios.',\n    )\n  }\n  if (holdout.length === 0) {\n    throw new Error('selfImprove: holdout split is empty. Pass more scenarios.')\n  }\n\n  const driver: ImprovementDriver =\n    opts.driver ??\n    gepaDriver({\n      llm: {\n        baseUrl: opts.llm?.baseUrl ?? 'https://router.tangle.tools/v1',\n        apiKey: opts.llm?.apiKey ?? process.env.OPENAI_API_KEY ?? '',\n      },\n      model: opts.llm?.model ?? 'anthropic/claude-sonnet-4.6',\n      target:\n        opts.driverTarget ??\n        'agent surface (system prompt or config) being optimized by selfImprove',\n      mutationPrimitives: opts.mutationPrimitives ?? DEFAULT_MUTATION_PRIMITIVES,\n    })\n\n  const gate: Gate<TArtifact, TScenario> =\n    opts.gate ??\n    defaultProductionGate<TArtifact, TScenario>({\n      holdoutScenarios: holdout,\n      deltaThreshold: 0.05,\n    })\n\n  const storage = opts.storage ?? inMemoryCampaignStorage()\n  const runDir = opts.runDir ?? `mem://selfImprove-${startedAt}`\n\n  if (opts.onProgress) {\n    opts.onProgress({ kind: 'baseline.started', scenarios: opts.scenarios.length })\n  }\n\n  const result = await runImprovementLoop<TScenario, TArtifact>({\n    scenarios: train,\n    baselineSurface: opts.baselineSurface,\n    dispatchWithSurface: opts.agent,\n    driver,\n    judges: [opts.judge],\n    populationSize,\n    maxGenerations: generations,\n    holdoutScenarios: holdout,\n    gate,\n    autoOnPromote: opts.autoOnPromote ?? 'none',\n    ghOwner: opts.ghOwner,\n    ghRepo: opts.ghRepo,\n    storage,\n    runDir,\n    maxConcurrency,\n    cellPlacement: opts.cellPlacement,\n    costCeiling,\n  })\n\n  const baseline = meanComposite(result.baselineOnHoldout.aggregates.byScenario)\n  const winnerStats = meanComposite(result.winnerOnHoldout.aggregates.byScenario)\n\n  if (opts.onProgress) {\n    opts.onProgress({\n      kind: 'baseline.completed',\n      compositeMean: baseline.compositeMean,\n      durationMs: Date.now() - startedAt,\n    })\n    opts.onProgress({\n      kind: 'gate.decided',\n      decision: result.gateResult.decision,\n      lift: winnerStats.compositeMean - baseline.compositeMean,\n    })\n  }\n\n  const totalCost =\n    result.baselineCampaign.aggregates.totalCostUsd +\n    result.generations.reduce(\n      (sum, gen) =>\n        sum + gen.surfaces.reduce((s, sf) => s + sf.campaign.aggregates.totalCostUsd, 0),\n      0,\n    )\n\n  // Rigor packet: feed baseline + winner cells through analyzeRuns().\n  // The two candidates (`baseline` / `winner`) give the lift section a\n  // clean paired comparison; per-judge / per-dimension / cost-quality\n  // sections populate from the cells' judgeScores.\n  const insight = await analyzeRuns({\n    runs: [\n      ...cellsToRunRecords(result.baselineCampaign.cells, 'baseline', runDir),\n      ...cellsToRunRecords(result.winnerOnHoldout.cells, 'winner', runDir),\n    ],\n    baselineCandidateId: 'baseline',\n    candidateCandidateId: 'winner',\n  })\n\n  const summary: SelfImproveResult<TScenario, TArtifact> = {\n    baseline,\n    winner: {\n      ...winnerStats,\n      surface: result.winnerSurface,\n    },\n    lift: winnerStats.compositeMean - baseline.compositeMean,\n    gateDecision: result.gateResult.decision,\n    generationsExplored: result.generations.length,\n    durationMs: Date.now() - startedAt,\n    totalCostUsd: totalCost,\n    insight,\n    raw: result,\n  }\n\n  // Opt-in hosted ingest. Failures logged but never fail the loop — the\n  // local result is always returned. This matches the wedge-doc invariant\n  // that LAND-tier never blocks on EXPAND-tier infra.\n  if (opts.hostedTenant) {\n    try {\n      await shipEvalRunToHosted(opts.hostedTenant, opts, summary, result, runDir)\n    } catch (err) {\n      const msg = err instanceof Error ? err.message : String(err)\n      // eslint-disable-next-line no-console -- intentional: hosted-ingest is best-effort\n      console.warn(`[agent-eval] hosted ingest failed (continuing): ${msg}`)\n    }\n  }\n\n  return summary\n}\n\nasync function shipEvalRunToHosted<TScenario extends Scenario, TArtifact>(\n  tenant: HostedTenant,\n  opts: SelfImproveOptions<TScenario, TArtifact>,\n  summary: SelfImproveResult<TScenario, TArtifact>,\n  raw: RunImprovementLoopResult<TArtifact, TScenario>,\n  runDir: string,\n): Promise<void> {\n  const client = createHostedClient(tenant)\n\n  function snapshotFromCampaign(\n    index: number,\n    surface: MutableSurface | undefined,\n    campaign: RunImprovementLoopResult<TArtifact, TScenario>['baselineCampaign'],\n    durationMs: number,\n  ): EvalRunGenerationSnapshot {\n    const cells: EvalRunCellScore[] = campaign.cells.map((cell) => {\n      const judgeScores = Object.values(cell.judgeScores)\n      const composite =\n        judgeScores.length === 0\n          ? 0\n          : judgeScores.reduce((s, j) => s + j.composite, 0) / judgeScores.length\n      return {\n        scenarioId: cell.scenarioId,\n        rep: cell.rep,\n        compositeMean: composite,\n        dimensions: Object.fromEntries(\n          Object.entries(cell.judgeScores).map(([name, score]) => [name, score.dimensions]),\n        ),\n        errorMessage: cell.error ?? undefined,\n      }\n    })\n    const compositeMean =\n      cells.length === 0 ? 0 : cells.reduce((s, c) => s + c.compositeMean, 0) / cells.length\n    return {\n      index,\n      surfaceHash:\n        typeof surface === 'string'\n          ? hashString(surface)\n          : hashString(JSON.stringify(surface ?? '')),\n      surface,\n      cells,\n      compositeMean,\n      costUsd: campaign.aggregates.totalCostUsd,\n      durationMs,\n    }\n  }\n\n  const generations: EvalRunGenerationSnapshot[] = []\n  // Baseline as generation 0.\n  generations.push(snapshotFromCampaign(0, opts.baselineSurface, raw.baselineCampaign, 0))\n  // Improvement generations as 1..N. Substrate stores per-surface campaigns\n  // per generation — we summarize the WINNING surface per generation here.\n  for (const gen of raw.generations) {\n    const winner = gen.surfaces.reduce(\n      (best, s) =>\n        s.campaign.aggregates.cellsExecuted > 0 &&\n        (best === undefined || averageComposite(s.campaign) > averageComposite(best.campaign))\n          ? s\n          : best,\n      gen.surfaces[0],\n    )\n    if (!winner) continue\n    generations.push(\n      snapshotFromCampaign(gen.record.generationIndex + 1, winner.surface, winner.campaign, 0),\n    )\n  }\n\n  const event: EvalRunEvent = {\n    runId: `${runDir}#${Date.now()}`,\n    runDir,\n    timestamp: new Date().toISOString(),\n    status: 'finished',\n    labels: opts.hostedLabels ?? {},\n    baseline: generations[0],\n    generations,\n    gateDecision: summary.gateDecision,\n    holdoutLift: summary.lift,\n    totalCostUsd: summary.totalCostUsd,\n    totalDurationMs: summary.durationMs,\n    insightReport: summary.insight,\n  }\n\n  await client.ingestEvalRun(event)\n}\n\nfunction averageComposite(\n  campaign: RunImprovementLoopResult<unknown, Scenario>['baselineCampaign'],\n): number {\n  const aggs = Object.values(campaign.aggregates.byScenario)\n  return aggs.length === 0 ? 0 : aggs.reduce((s, a) => s + a.meanComposite, 0) / aggs.length\n}\n\nfunction hashString(s: string): string {\n  let h = 2166136261 >>> 0\n  for (let i = 0; i < s.length; i++) {\n    h ^= s.charCodeAt(i)\n    h = Math.imul(h, 16777619) >>> 0\n  }\n  return h.toString(16).padStart(8, '0')\n}\n\n/**\n * Adapt campaign cells into the `RunRecord` shape `analyzeRuns()` consumes.\n * Each cell becomes one run; `candidateId` is the caller-supplied label so\n * baseline + winner pair cleanly on `(experimentId, seed)`.\n */\nfunction cellsToRunRecords<TArtifact>(\n  cells: ReadonlyArray<CampaignCellResult<TArtifact>>,\n  candidateId: 'baseline' | 'winner',\n  runId: string,\n): RunRecord[] {\n  return cells.map((cell) => {\n    const perJudge: Record<string, Record<string, number>> = {}\n    const perDimMeanAccum: Record<string, { sum: number; n: number }> = {}\n    let compositeSum = 0\n    let compositeCount = 0\n    for (const [judgeId, score] of Object.entries(cell.judgeScores)) {\n      perJudge[judgeId] = { ...score.dimensions }\n      for (const [dim, value] of Object.entries(score.dimensions)) {\n        if (!Number.isFinite(value)) continue\n        const accum = perDimMeanAccum[dim] ?? { sum: 0, n: 0 }\n        accum.sum += value\n        accum.n += 1\n        perDimMeanAccum[dim] = accum\n      }\n      if (Number.isFinite(score.composite)) {\n        compositeSum += score.composite\n        compositeCount += 1\n      }\n    }\n    const perDimMean: Record<string, number> = {}\n    for (const [dim, { sum, n }] of Object.entries(perDimMeanAccum)) {\n      perDimMean[dim] = n === 0 ? 0 : sum / n\n    }\n    const composite = compositeCount === 0 ? 0 : compositeSum / compositeCount\n    const judgeScores: JudgeScoresRecord = {\n      perJudge,\n      perDimMean,\n      composite,\n    }\n    return {\n      runId: `${runId}::${candidateId}::${cell.cellId}`,\n      experimentId: runId,\n      candidateId,\n      // Pair on (scenarioId, rep) — analyzeRuns pairs on (experimentId, seed).\n      // Synthesize a stable seed for that pairing.\n      seed:\n        cell.rep * 1_000_000 +\n        hashString(cell.scenarioId)\n          .slice(0, 6)\n          .split('')\n          .reduce((a, c) => (a * 31 + c.charCodeAt(0)) >>> 0, 0),\n      model: 'campaign-cell',\n      promptHash: 'sha256:cell',\n      configHash: 'sha256:cell',\n      commitSha: 'cell',\n      wallMs: cell.durationMs,\n      costUsd: cell.costUsd,\n      tokenUsage: { input: 0, output: 0 },\n      outcome: {\n        holdoutScore: composite,\n        raw: {},\n        judgeScores,\n      },\n      splitTag: 'holdout',\n      ...(cell.error ? { failureMode: cell.error } : {}),\n    } satisfies RunRecord\n  })\n}\n","/**\n * # `intake/feedback-table` — multi-rater approve/reject corpus → `RunRecord[]`.\n *\n * The generic shape behind Obsidian's `#approved` / `#rejected` tags, a\n * Google Sheet, a Postgres `feedback` table, or any CSV with ratings.\n *\n * Caller supplies one row per (run, rater) tuple plus per-run metadata; the\n * adapter rolls them up into the substrate-canonical `RunRecord` shape so\n * `analyzeRuns({ runs, raterScores })` can produce inter-rater agreement,\n * disagreement triage, and downstream recommendations.\n *\n * Per-run `RunRecord.outcome.searchScore` is the rater-mean rating\n * (normalised to 0..1 when scale is supplied); `outcome.raw` carries the\n * per-rater scores keyed by rater id for downstream attribution.\n */\n\nimport type { JudgeScoresRecord, RunOutcome, RunRecord, RunSplitTag } from '../../run-record'\n\nexport interface FeedbackTableRow {\n  /** Stable id for this run — the unit a rater scored. Drives pairing\n   *  across analysis primitives. */\n  runId: string\n  /** Identifier of the rater that produced this rating. */\n  rater: string\n  /** The rating itself. Accepts boolean (approve/reject), 0..1 scalar,\n   *  or any numeric scale — see `scale`. */\n  rating: number | boolean\n  /** Optional metadata carried through to `RunRecord.outcome.raw` and the\n   *  custom-shape metadata bag. */\n  metadata?: Record<string, unknown>\n}\n\nexport interface FeedbackTableMeta {\n  runId: string\n  /** When omitted, defaults to `'feedback-corpus'`. Used to group related\n   *  runs in `analyzeRuns()` lift analysis. */\n  experimentId?: string\n  /** When omitted, defaults to `runId` — each run is its own candidate. */\n  candidateId?: string\n  /** Cost in USD, when available. Set to 0 when unknown — the consumer's\n   *  cost analysis sections will collapse gracefully. */\n  costUsd?: number\n  /** Wall-clock ms, when available. Defaults to 0. */\n  wallMs?: number\n  /** Model identifier including snapshot. Default `unknown@unknown`. */\n  model?: string\n  /** Optional sha256 of the prompt; default `'sha256:unknown'`. */\n  promptHash?: string\n  /** Default `'sha256:unknown'`. */\n  configHash?: string\n  /** Default `'unknown'`. */\n  commitSha?: string\n  /** Default `'holdout'` — feedback corpora are by nature the holdout\n   *  signal a closed-loop improvement aims at. */\n  splitTag?: RunSplitTag\n  /** Free-form metadata available to consumers via the cast-out path on\n   *  the resulting RunRecord. */\n  extras?: Record<string, unknown>\n}\n\nexport interface FromFeedbackTableOptions {\n  /** Per-(run, rater) ratings. */\n  ratings: FeedbackTableRow[]\n  /** Per-run metadata. When a runId appears in `ratings` but not here, the\n   *  adapter synthesises minimal metadata with defaults documented above. */\n  meta?: FeedbackTableMeta[]\n  /** Rating scale. Provide `{ min, max }` for non-0..1 numeric scales.\n   *  Booleans are normalised: true → 1, false → 0. Default: assumes\n   *  ratings are already 0..1. */\n  scale?: { min: number; max: number }\n  /** When true, the rater scores are emitted into `raterScores` (a sibling\n   *  array `analyzeRuns()` accepts) instead of being averaged into the\n   *  run's `outcome.searchScore`. Default `true` — preserves rater-level\n   *  signal for inter-rater analysis. */\n  emitRaterScores?: boolean\n}\n\nexport interface FromFeedbackTableResult {\n  runs: RunRecord[]\n  /** Rater-level scores ready to pass into `analyzeRuns({ raterScores })`\n   *  for inter-rater agreement + disagreement triage. */\n  raterScores: Array<{ runId: string; rater: string; score: number }>\n}\n\nexport function fromFeedbackTable(opts: FromFeedbackTableOptions): FromFeedbackTableResult {\n  const { ratings, meta = [], scale, emitRaterScores = true } = opts\n  const metaByRun = new Map(meta.map((m) => [m.runId, m]))\n\n  // Normalise per-rating to a 0..1 score.\n  const normalise = (rating: number | boolean): number => {\n    if (typeof rating === 'boolean') return rating ? 1 : 0\n    if (!Number.isFinite(rating)) return Number.NaN\n    if (!scale) return rating\n    const { min, max } = scale\n    if (max === min) return rating\n    return (rating - min) / (max - min)\n  }\n\n  // Group ratings by runId.\n  const byRun = new Map<string, FeedbackTableRow[]>()\n  for (const row of ratings) {\n    const list = byRun.get(row.runId) ?? []\n    list.push(row)\n    byRun.set(row.runId, list)\n  }\n\n  const runs: RunRecord[] = []\n  const raterScores: FromFeedbackTableResult['raterScores'] = []\n\n  for (const [runId, rowsForRun] of byRun) {\n    const normalised = rowsForRun\n      .map((r) => ({ rater: r.rater, score: normalise(r.rating) }))\n      .filter((r) => Number.isFinite(r.score))\n    if (normalised.length === 0) continue\n\n    const meanScore = normalised.reduce((s, r) => s + r.score, 0) / normalised.length\n\n    const runMeta = metaByRun.get(runId) ?? ({ runId } as FeedbackTableMeta)\n\n    const judgeScores: JudgeScoresRecord = {\n      perJudge: Object.fromEntries(normalised.map((r) => [r.rater, { rating: r.score }])),\n      perDimMean: { rating: meanScore },\n      composite: meanScore,\n    }\n\n    const outcome: RunOutcome = {\n      // Feedback corpora ARE the holdout signal — score lands on\n      // `holdoutScore` so downstream substrate primitives (`paretoChart`,\n      // promotion gates) read it correctly by default.\n      holdoutScore: meanScore,\n      raw: Object.fromEntries(normalised.map((r) => [`rater:${r.rater}`, r.score])),\n      judgeScores,\n    }\n\n    runs.push({\n      runId,\n      experimentId: runMeta.experimentId ?? 'feedback-corpus',\n      candidateId: runMeta.candidateId ?? runId,\n      seed: 0,\n      model: runMeta.model ?? 'unknown@unknown',\n      promptHash: runMeta.promptHash ?? 'sha256:unknown',\n      configHash: runMeta.configHash ?? 'sha256:unknown',\n      commitSha: runMeta.commitSha ?? 'unknown',\n      wallMs: runMeta.wallMs ?? 0,\n      costUsd: runMeta.costUsd ?? 0,\n      tokenUsage: { input: 0, output: 0 },\n      outcome,\n      splitTag: runMeta.splitTag ?? 'holdout',\n    } as RunRecord)\n\n    if (emitRaterScores) {\n      for (const r of normalised) raterScores.push({ runId, rater: r.rater, score: r.score })\n    }\n  }\n\n  return { runs, raterScores }\n}\n","/**\n * # `intake/otel-spans` — OTel `TraceSpanEvent[]` → `RunRecord[]`.\n *\n * Turns an existing observability stream into the substrate-canonical\n * `RunRecord` shape so consumers with logs but no eval discipline can\n * call `analyzeRuns()` against their production traffic immediately.\n *\n * Pivot rule: spans are grouped by `tangle.runId` (the same attribute the\n * hosted-tier wire format uses) or, when absent, by `traceId`. One group\n * becomes one `RunRecord`. The root span (no `parentSpanId`) supplies:\n *\n *   - `runId` (the group key)\n *   - `wallMs` from `endTimeUnixNano - startTimeUnixNano`\n *   - `model` from `gen_ai.request.model` / `llm.model` / `tangle.model`\n *   - cost from `cost.usd` / `gen_ai.usage.cost_usd` / `tangle.cost.usd`\n *   - token usage from `gen_ai.usage.{input,output}_tokens`\n *   - `outcome.searchScore` from `tangle.score` / `eval.score` when\n *     present; `outcome.raw` collects every numeric attribute.\n *\n * Spans that ERRORed (`status.code === 'ERROR'`) populate `failureMode`\n * with their `name` so `analyzeRuns()`'s failure clustering sees them.\n */\n\nimport type { TraceSpanEvent } from '../../hosted/types'\nimport type {\n  JudgeScoresRecord,\n  RunOutcome,\n  RunRecord,\n  RunSplitTag,\n  RunTokenUsage,\n} from '../../run-record'\n\nconst SCORE_KEYS = ['tangle.score', 'eval.score', 'score']\nconst MODEL_KEYS = ['tangle.model', 'gen_ai.request.model', 'llm.model', 'model']\nconst COST_KEYS = ['tangle.cost.usd', 'gen_ai.usage.cost_usd', 'cost.usd', 'cost']\nconst INPUT_TOKEN_KEYS = ['gen_ai.usage.input_tokens', 'tangle.tokens.in', 'tokens.in']\nconst OUTPUT_TOKEN_KEYS = ['gen_ai.usage.output_tokens', 'tangle.tokens.out', 'tokens.out']\nconst PROMPT_HASH_KEYS = ['tangle.prompt_hash', 'prompt.hash']\nconst CONFIG_HASH_KEYS = ['tangle.config_hash', 'config.hash']\n\nexport interface FromOtelSpansOptions {\n  spans: TraceSpanEvent[]\n  /** Default split tag for synthesized records. Defaults to `'holdout'`. */\n  defaultSplit?: RunSplitTag\n  /** Default `experimentId` when not present on any span. */\n  experimentId?: string\n}\n\nexport function fromOtelSpans(opts: FromOtelSpansOptions): RunRecord[] {\n  const { spans, defaultSplit = 'holdout', experimentId = 'otel-corpus' } = opts\n  const grouped = groupSpans(spans)\n\n  const runs: RunRecord[] = []\n  for (const [groupKey, groupSpans] of grouped) {\n    const root = findRoot(groupSpans)\n    if (!root) continue\n\n    const wallMs = Math.max(0, (root.endTimeUnixNano - root.startTimeUnixNano) / 1_000_000)\n    const model = readAttrString(groupSpans, MODEL_KEYS) ?? 'unknown@unknown'\n    const costUsd = readAttrNumber(groupSpans, COST_KEYS) ?? 0\n    const inputTokens = readAttrNumber(groupSpans, INPUT_TOKEN_KEYS) ?? 0\n    const outputTokens = readAttrNumber(groupSpans, OUTPUT_TOKEN_KEYS) ?? 0\n    const promptHash = readAttrString(groupSpans, PROMPT_HASH_KEYS) ?? 'sha256:unknown'\n    const configHash = readAttrString(groupSpans, CONFIG_HASH_KEYS) ?? 'sha256:unknown'\n    const score = readAttrNumber(groupSpans, SCORE_KEYS)\n\n    const rawNumeric = collectNumericAttrs(groupSpans)\n    const tokenUsage: RunTokenUsage = {\n      input: inputTokens,\n      output: outputTokens,\n    }\n\n    const judgeScores: JudgeScoresRecord | undefined =\n      score !== undefined\n        ? {\n            perJudge: { 'otel-derived': { score } },\n            perDimMean: { score },\n            composite: score,\n          }\n        : undefined\n\n    const errorSpan = groupSpans.find((s) => s.status?.code === 'ERROR')\n    const outcome: RunOutcome = {\n      ...(opts.defaultSplit === 'search' ? { searchScore: score } : { holdoutScore: score }),\n      raw: rawNumeric,\n      ...(judgeScores ? { judgeScores } : {}),\n    }\n\n    runs.push({\n      runId: groupKey,\n      experimentId,\n      candidateId: (root.attributes['tangle.candidateId'] as string | undefined) ?? 'otel-default',\n      seed: 0,\n      model,\n      promptHash,\n      configHash,\n      commitSha: (root.attributes['tangle.commit_sha'] as string | undefined) ?? 'unknown',\n      wallMs,\n      costUsd,\n      tokenUsage,\n      outcome,\n      splitTag: defaultSplit,\n      ...(errorSpan ? { failureMode: errorSpan.name } : {}),\n    } as RunRecord)\n  }\n  return runs\n}\n\n// ── Internal helpers ────────────────────────────────────────────────\n\nfunction groupSpans(spans: TraceSpanEvent[]): Map<string, TraceSpanEvent[]> {\n  const m = new Map<string, TraceSpanEvent[]>()\n  for (const span of spans) {\n    const key = (span['tangle.runId'] as string | undefined) ?? span.traceId\n    const list = m.get(key) ?? []\n    list.push(span)\n    m.set(key, list)\n  }\n  return m\n}\n\nfunction findRoot(group: TraceSpanEvent[]): TraceSpanEvent | undefined {\n  return group.find((s) => !s.parentSpanId) ?? group[0]\n}\n\nfunction readAttrString(spans: TraceSpanEvent[], keys: string[]): string | undefined {\n  for (const span of spans) {\n    for (const key of keys) {\n      const v = span.attributes[key]\n      if (typeof v === 'string' && v.length > 0) return v\n    }\n  }\n  return undefined\n}\n\nfunction readAttrNumber(spans: TraceSpanEvent[], keys: string[]): number | undefined {\n  for (const span of spans) {\n    for (const key of keys) {\n      const v = span.attributes[key]\n      if (typeof v === 'number' && Number.isFinite(v)) return v\n      if (typeof v === 'string') {\n        const parsed = Number(v)\n        if (Number.isFinite(parsed)) return parsed\n      }\n    }\n  }\n  return undefined\n}\n\nfunction collectNumericAttrs(spans: TraceSpanEvent[]): Record<string, number> {\n  const raw: Record<string, number> = {}\n  for (const span of spans) {\n    for (const [k, v] of Object.entries(span.attributes)) {\n      if (typeof v === 'number' && Number.isFinite(v)) raw[k] = v\n    }\n  }\n  return raw\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AA6FA,eAAsB,YAAY,MAAkD;AAClF,QAAM,OAAO,KAAK;AAClB,QAAM,OAAO,KAAK,iBAAiB;AACnC,QAAM,YAAY,KAAK,qBAAqB;AAC5C,QAAM,QAAQ,aAAa,MAAM,KAAK,SAAS,MAAM;AAErD,QAAM,mBAAmB,KACtB,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,OAAO,OAAO,YAAY,GAAG,KAAK,EAAE,EAAE,EAC7D,OAAO,CAAC,MAAM,OAAO,SAAS,EAAE,KAAK,CAAC;AACzC,QAAM,YAAY;AAAA,IAChB,iBAAiB,IAAI,CAAC,MAAM,EAAE,KAAK;AAAA,IACnC;AAAA,IACA;AAAA,EACF;AAEA,QAAM,eAAe,oBAAoB,MAAM,IAAI;AAEnD,QAAM,QAAQ,KAAK,IAAI,CAAC,MAAM,EAAE,OAAO,EAAE,OAAO,OAAO,QAAQ;AAC/D,QAAM,WAAW,eAAe,OAAO,IAAI;AAC3C,QAAM,SAAS,YAAY,MAAM,EAAE,MAAM,CAAC;AAC1C,QAAM,WAA+C,CAAC;AACtD,MAAI,MAAM,WAAW,KAAK,MAAM,MAAM,CAAC,MAAM,MAAM,CAAC,GAAG;AACrD,aAAS,OAAO;AAAA,EAClB;AACA,MAAI,OAAO,OAAO,SAAS,GAAG;AAC5B,aAAS,SACP,OAAO,OAAO,WAAW,IACrB,4CACA;AAAA,EACR;AACA,QAAM,cAAc;AAAA,IAClB,MAAM;AAAA,IACN;AAAA,IACA,GAAI,SAAS,QAAQ,SAAS,SAAS,EAAE,SAAS,IAAI,CAAC;AAAA,EACzD;AAEA,QAAM,SAAS,qBAAqB,IAAI;AAExC,QAAM,aAAa,KAAK,cAAc,kBAAkB,KAAK,WAAW,IAAI;AAE5E,QAAM,OAAO,YAAY,MAAM,KAAK,qBAAqB,KAAK,sBAAsB,KAAK;AAEzF,QAAM,kBAAkB,KAAK,UACzB,MAAM,uBAAuB,MAAM,KAAK,SAAS,KAAK,IACtD;AAEJ,QAAM,gBAAgB,KAAK,kBACvB,qBAAqB,MAAM,KAAK,eAAe,IAC/C;AAEJ,QAAM,qBAAqB,KAAK,gBAC5B,0BAA0B,MAAM,KAAK,eAAe,KAAK,IACzD;AAEJ,QAAM,UAAU,sBAAsB,WAAW,MAAM,aAAa;AAEpE,QAAM,wBAAwB,KAAK,eAC/B,6BAA6B,MAAM,KAAK,cAAc,OAAO,KAAK,aAAa,IAC/E;AAEJ,QAAM,kBAAkB,qBAAqB;AAAA,IAC3C;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF,CAAC;AAED,SAAO;AAAA,IACL,GAAG,KAAK;AAAA,IACR;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA,GAAI,wBAAwB,EAAE,sBAAsB,IAAI,CAAC;AAAA,IACzD;AAAA,EACF;AACF;AAUA,SAAS,6BACP,SACA,UACA,OACA,aACmC;AACnC,MAAI,QAAQ,WAAW,KAAK,SAAS,WAAW,EAAG,QAAO;AAE1D,QAAM,UAAuC,CAAC;AAC9C,QAAM,aAA8C,CAAC;AAErD,QAAM,mBAAmB,QACtB,IAAI,CAAC,MAAM,YAAY,GAAG,KAAK,CAAC,EAChC,OAAO,OAAO,QAAQ;AACzB,QAAM,oBAAoB,SACvB,IAAI,CAAC,MAAM,YAAY,GAAG,KAAK,CAAC,EAChC,OAAO,OAAO,QAAQ;AACzB,MAAI,iBAAiB,SAAS,KAAK,kBAAkB,SAAS,GAAG;AAC/D,YAAQ,YAAY,aAAa,mBAAmB,gBAAgB;AACpE,eAAW,YAAY;AAAA,EACzB;AAEA,QAAM,cAAc,QAAQ,IAAI,CAAC,MAAM,EAAE,OAAO,EAAE,OAAO,OAAO,QAAQ;AACxE,QAAM,eAAe,SAAS,IAAI,CAAC,MAAM,EAAE,OAAO,EAAE,OAAO,OAAO,QAAQ;AAC1E,MAAI,YAAY,SAAS,KAAK,aAAa,SAAS,GAAG;AACrD,YAAQ,OAAO,aAAa,cAAc,WAAW;AACrD,eAAW,OAAO;AAAA,EACpB;AAEA,QAAM,aAAa,QAAQ,IAAI,CAAC,MAAM,EAAE,MAAM,EAAE,OAAO,OAAO,QAAQ;AACtE,QAAM,cAAc,SAAS,IAAI,CAAC,MAAM,EAAE,MAAM,EAAE,OAAO,OAAO,QAAQ;AACxE,MAAI,WAAW,SAAS,KAAK,YAAY,SAAS,GAAG;AACnD,YAAQ,WAAW,aAAa,aAAa,UAAU;AACvD,eAAW,WAAW;AAAA,EACxB;AAEA,QAAM,aAAa,QAChB,IAAI,CAAC,OAAO,EAAE,WAAW,SAAS,MAAM,EAAE,WAAW,UAAU,EAAE,EACjE,OAAO,OAAO,QAAQ;AACzB,QAAM,cAAc,SACjB,IAAI,CAAC,OAAO,EAAE,WAAW,SAAS,MAAM,EAAE,WAAW,UAAU,EAAE,EACjE,OAAO,OAAO,QAAQ;AACzB,MAAI,WAAW,SAAS,KAAK,YAAY,SAAS,GAAG;AACnD,YAAQ,aAAa,aAAa,aAAa,UAAU;AACzD,eAAW,aAAa;AAAA,EAC1B;AAKA,QAAM,cAAc,oBAAoB,OAAO;AAC/C,QAAM,eAAe,oBAAoB,QAAQ;AACjD,aAAW,OAAO,OAAO,KAAK,WAAW,GAAG;AAC1C,UAAM,IAAI,aAAa,GAAG;AAC1B,UAAM,IAAI,YAAY,GAAG;AACzB,QAAI,CAAC,KAAK,EAAE,WAAW,KAAK,CAAC,KAAK,EAAE,WAAW,EAAG;AAClD,YAAQ,OAAO,GAAG,EAAE,IAAI,aAAa,GAAG,CAAC;AACzC,eAAW,OAAO,GAAG,EAAE,IAAI;AAAA,EAC7B;AAEA,QAAM,mBAA6B,CAAC;AACpC,QAAM,kBAA4B,CAAC;AACnC,aAAW,CAAC,MAAM,KAAK,KAAK,OAAO,QAAQ,OAAO,GAAG;AACnD,QAAI,CAAC,MAAM,YAAa;AACxB,UAAM,MAAM,WAAW,IAAI,KAAK;AAChC,UAAM,SAAS,QAAQ,qBAAqB,MAAM,QAAQ,IAAI,MAAM,QAAQ;AAC5E,QAAI,OAAQ,iBAAgB,KAAK,IAAI;AAAA,QAChC,kBAAiB,KAAK,IAAI;AAAA,EACjC;AAEA,SAAO;AAAA,IACL,WAAW,SAAS;AAAA,IACpB,UAAU,QAAQ;AAAA,IAClB,GAAI,cAAc,EAAE,YAAY,IAAI,CAAC;AAAA,IACrC;AAAA,IACA;AAAA,IACA;AAAA,EACF;AACF;AAGA,SAAS,oBAAoB,MAA6C;AACxE,QAAM,MAAgC,CAAC;AACvC,aAAW,KAAK,MAAM;AACpB,UAAM,SAAS,EAAE,QAAQ,aAAa;AACtC,QAAI,CAAC,OAAQ;AACb,eAAW,CAAC,KAAK,KAAK,KAAK,OAAO,QAAQ,MAAM,GAAG;AACjD,UAAI,CAAC,OAAO,SAAS,KAAK,EAAG;AAC7B,UAAI,CAAC,IAAI,GAAG,EAAG,KAAI,GAAG,IAAI,CAAC;AAC3B,UAAI,GAAG,EAAE,KAAK,KAAe;AAAA,IAC/B;AAAA,EACF;AACA,SAAO;AACT;AAIA,SAAS,aAAa,UAAoB,SAAgC;AACxE,QAAM,eAAe,KAAK,QAAQ;AAClC,QAAM,cAAc,KAAK,OAAO;AAChC,QAAM,cAAc,eAAe,UAAU,YAAY;AACzD,QAAM,aAAa,eAAe,SAAS,WAAW;AACtD,QAAM,YAAY,SAAS;AAC3B,QAAM,WAAW,QAAQ;AACzB,QAAM,QAAQ,cAAc;AAG5B,QAAM,KAAK,KAAK,KAAK,cAAc,YAAY,aAAa,QAAQ;AAIpE,QAAM,YAAY,QAAQ,KAAK,IAAI,KAAK;AACxC,QAAM,OAAyB,CAAC,QAAQ,WAAW,QAAQ,SAAS;AAGpE,QAAM,IAAI,KAAK,IAAI,QAAQ,KAAK;AAChC,QAAM,SAAS,KAAK,IAAI,KAAK,IAAI,kBAAkB,KAAK,IAAI,CAAC,CAAC,KAAK;AAGnE,QAAM,eAAe,KAAK;AAAA,MACtB,YAAY,KAAK,eAAe,WAAW,KAAK,cAChD,KAAK,IAAI,GAAG,YAAY,WAAW,CAAC;AAAA,EACxC;AACA,QAAMA,WAAU,eAAe,IAAI,QAAQ,eAAe;AAG1D,QAAM,cAAc,SAAS,QAAQ,KAAK,IAAIA,QAAO,KAAK;AAE1D,SAAO;AAAA,IACL,SAAS;AAAA,IACT,UAAU;AAAA,IACV;AAAA,IACA;AAAA,IACA;AAAA,IACA,SAAAA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AACF;AAEA,SAAS,eAAe,IAAc,QAAwB;AAC5D,MAAI,GAAG,SAAS,EAAG,QAAO;AAC1B,MAAI,IAAI;AACR,aAAW,KAAK,GAAI,OAAM,IAAI,WAAW;AACzC,SAAO,KAAK,GAAG,SAAS;AAC1B;AAGA,SAAS,kBAAkB,GAAmB;AAC5C,QAAM,KAAK;AACX,QAAM,KAAK;AACX,QAAM,KAAK;AACX,QAAM,KAAK;AACX,QAAM,KAAK;AACX,QAAM,IAAI;AACV,QAAM,OAAO,IAAI,IAAI,KAAK;AAC1B,QAAM,IAAI,KAAK,IAAI,CAAC,IAAI,KAAK;AAC7B,QAAM,IAAI,KAAK,IAAI,IAAI;AACvB,QAAM,IAAI,QAAQ,KAAK,IAAI,MAAM,IAAI,MAAM,IAAI,MAAM,IAAI,MAAM,IAAI,KAAK,IAAI,CAAC,IAAI,CAAC;AAClF,SAAO,OAAO,IAAI,OAAO;AAC3B;AAIA,SAAS,aACP,MACA,MACsB;AACtB,MAAI,SAAS,OAAQ,QAAO;AAC5B,QAAM,aAAa,KAAK,KAAK,CAAC,MAAM,OAAO,SAAS,EAAE,QAAQ,YAAY,CAAC;AAC3E,SAAO,aAAa,YAAY;AAClC;AAEA,SAAS,YAAY,KAAgB,OAAqC;AACxE,QAAM,UAAU,UAAU,YAAY,IAAI,QAAQ,eAAe,IAAI,QAAQ;AAC7E,MAAI,OAAO,SAAS,OAAO,EAAG,QAAO;AAIrC,QAAM,MAAM,UAAU,YAAY,IAAI,QAAQ,cAAc,IAAI,QAAQ;AACxE,SAAO,OAAO,SAAS,GAAG,IAAK,MAAiB,OAAO;AACzD;AAIA,SAAS,eACP,QACA,MACA,SACoB;AACpB,MAAI,OAAO,WAAW,GAAG;AACvB,WAAO;AAAA,MACL,GAAG;AAAA,MACH,MAAM;AAAA,MACN,KAAK;AAAA,MACL,KAAK;AAAA,MACL,QAAQ;AAAA,MACR,KAAK;AAAA,MACL,KAAK;AAAA,MACL,WAAW,CAAC;AAAA,IACd;AAAA,EACF;AACA,QAAM,SAAS,CAAC,GAAG,MAAM,EAAE,KAAK,CAAC,GAAG,MAAM,IAAI,CAAC;AAC/C,QAAM,IAAI,OAAO;AACjB,QAAMC,QAAO,OAAO,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI;AACjD,QAAM,WAAW,OAAO,OAAO,CAAC,GAAG,MAAM,KAAK,IAAIA,UAAS,GAAG,CAAC,IAAI;AACnE,QAAM,SAAS,KAAK,KAAK,QAAQ;AACjC,QAAM,WAAW,UACb,CAAC,GAAG,OAAO,EAAE,KAAK,CAAC,GAAG,MAAM,EAAE,QAAQ,EAAE,KAAK,EAAE,MAAM,GAAG,KAAK,IAAI,GAAG,QAAQ,MAAM,CAAC,IACnF;AACJ,SAAO;AAAA,IACL;AAAA,IACA,MAAAA;AAAA,IACA,KAAK,WAAW,QAAQ,GAAG;AAAA,IAC3B,KAAK,WAAW,QAAQ,IAAI;AAAA,IAC5B;AAAA,IACA,KAAK,OAAO,CAAC;AAAA,IACb,KAAK,OAAO,IAAI,CAAC;AAAA,IACjB,WAAW,UAAU,QAAQ,IAAI;AAAA,IACjC,GAAI,WAAW,EAAE,SAAS,IAAI,CAAC;AAAA,EACjC;AACF;AAEA,SAAS,WAAW,QAAkB,GAAmB;AACvD,MAAI,OAAO,WAAW,EAAG,QAAO;AAChC,MAAI,OAAO,WAAW,EAAG,QAAO,OAAO,CAAC;AACxC,QAAM,OAAO,OAAO,SAAS,KAAK;AAClC,QAAM,KAAK,KAAK,MAAM,GAAG;AACzB,QAAM,KAAK,KAAK,KAAK,GAAG;AACxB,MAAI,OAAO,GAAI,QAAO,OAAO,EAAE;AAC/B,QAAM,IAAI,MAAM;AAChB,SAAO,OAAO,EAAE,KAAM,IAAI,KAAK,OAAO,EAAE,IAAK;AAC/C;AAKA,SAAS,UAAU,QAAkB,MAA+C;AAClF,MAAI,OAAO,WAAW,KAAK,OAAO,EAAG,QAAO,CAAC;AAC7C,QAAM,MAAM,OAAO,CAAC;AACpB,QAAM,MAAM,OAAO,OAAO,SAAS,CAAC;AACpC,MAAI,QAAQ,IAAK,QAAO,CAAC,EAAE,IAAI,KAAK,IAAI,KAAK,OAAO,OAAO,OAAO,CAAC;AACnE,QAAM,SAAS,MAAM,OAAO;AAC5B,QAAM,MAAuC,CAAC;AAC9C,WAAS,IAAI,GAAG,IAAI,MAAM,KAAK;AAC7B,UAAM,KAAK,MAAM,IAAI;AACrB,UAAM,KAAK,MAAM,OAAO,IAAI,MAAM,KAAK;AACvC,QAAI,KAAK,EAAE,IAAI,IAAI,OAAO,EAAE,CAAC;AAAA,EAC/B;AACA,aAAW,KAAK,QAAQ;AACtB,UAAM,MAAM,KAAK,IAAI,OAAO,GAAG,KAAK,OAAO,IAAI,OAAO,KAAK,CAAC;AAC5D,QAAI,GAAG,EAAG;AAAA,EACZ;AACA,SAAO;AACT;AAEA,SAAS,oBAAoB,MAAmB,MAAkD;AAKhG,QAAM,QAAQ,oBAAI,IAAsB;AACxC,aAAW,OAAO,MAAM;AACtB,UAAM,SAAS,IAAI,QAAQ;AAC3B,QAAI,CAAC,OAAQ;AACb,eAAW,CAAC,KAAK,KAAK,KAAK,OAAO,QAAQ,OAAO,cAAc,CAAC,CAAC,GAAG;AAClE,UAAI,CAAC,OAAO,SAAS,KAAK,EAAG;AAC7B,YAAM,MAAM,MAAM,IAAI,GAAG,KAAK,CAAC;AAC/B,UAAI,KAAK,KAAK;AACd,YAAM,IAAI,KAAK,GAAG;AAAA,IACpB;AAAA,EACF;AACA,QAAM,MAA0C,CAAC;AACjD,aAAW,CAAC,KAAK,MAAM,KAAK,MAAO,KAAI,GAAG,IAAI,eAAe,QAAQ,IAAI;AACzE,SAAO;AACT;AAIA,SAAS,qBAAqB,MAAiD;AAI7E,QAAM,MAAoC,CAAC;AAC3C,QAAM,UAAU,oBAAI,IAAsB;AAC1C,aAAW,OAAO,MAAM;AACtB,UAAM,SAAS,IAAI,QAAQ;AAC3B,QAAI,CAAC,QAAQ,SAAU;AACvB,eAAW,CAAC,SAAS,IAAI,KAAK,OAAO,QAAQ,OAAO,QAAQ,GAAG;AAC7D,YAAM,YAAY,OAAO,OAAO,IAAI,EAAE,OAAO,OAAO,QAAQ;AAC5D,UAAI,UAAU,WAAW,EAAG;AAC5B,YAAM,YAAY,UAAU,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,UAAU;AACnE,YAAM,MAAM,QAAQ,IAAI,OAAO,KAAK,CAAC;AACrC,UAAI,KAAK,SAAS;AAClB,cAAQ,IAAI,SAAS,GAAG;AAAA,IAC1B;AAAA,EACF;AACA,aAAW,CAAC,SAAS,MAAM,KAAK,SAAS;AACvC,QAAI,OAAO,IAAI;AAAA,MACb,GAAG,OAAO;AAAA,MACV,WAAW,OAAO,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,OAAO;AAAA,IACxD;AAAA,EACF;AACA,SAAO;AACT;AAIA,SAAS,kBACP,SAC+B;AAC/B,QAAM,QAAQ,oBAAI,IAAqD;AACvE,aAAW,KAAK,SAAS;AACvB,QAAI,CAAC,OAAO,SAAS,EAAE,KAAK,EAAG;AAC/B,UAAM,OAAO,MAAM,IAAI,EAAE,KAAK,KAAK,CAAC;AACpC,SAAK,KAAK,EAAE,OAAO,EAAE,OAAO,OAAO,EAAE,MAAM,CAAC;AAC5C,UAAM,IAAI,EAAE,OAAO,IAAI;AAAA,EACzB;AACA,QAAM,SAAS,IAAI,IAAI,QAAQ,IAAI,CAAC,MAAM,EAAE,KAAK,CAAC;AAClD,QAAM,eAAyB,CAAC;AAChC,aAAW,CAAC,OAAO,YAAY,KAAK,OAAO;AACzC,UAAM,OAAO,IAAI,IAAI,aAAa,IAAI,CAAC,MAAM,EAAE,KAAK,CAAC;AACrD,QAAI,MAAM;AACV,eAAW,KAAK,OAAQ,KAAI,CAAC,KAAK,IAAI,CAAC,EAAG,OAAM;AAChD,QAAI,IAAK,cAAa,KAAK,KAAK;AAAA,EAClC;AACA,MAAI,OAAO,OAAO,KAAK,aAAa,WAAW,EAAG,QAAO;AAEzD,QAAM,YAAY,CAAC,GAAG,MAAM,EAAE,KAAK;AACnC,QAAM,UAAkC,CAAC;AACzC,WAAS,IAAI,GAAG,IAAI,UAAU,QAAQ,KAAK;AACzC,aAAS,IAAI,IAAI,GAAG,IAAI,UAAU,QAAQ,KAAK;AAC7C,YAAM,IAAI,UAAU,CAAC;AACrB,YAAM,IAAI,UAAU,CAAC;AACrB,YAAM,UAAoB,CAAC;AAC3B,YAAM,UAAoB,CAAC;AAC3B,iBAAW,SAAS,cAAc;AAChC,cAAM,eAAe,MAAM,IAAI,KAAK;AACpC,cAAM,KAAK,aAAa,KAAK,CAAC,MAAM,EAAE,UAAU,CAAC,GAAG;AACpD,cAAM,KAAK,aAAa,KAAK,CAAC,MAAM,EAAE,UAAU,CAAC,GAAG;AACpD,YAAI,OAAO,UAAa,OAAO,QAAW;AACxC,kBAAQ,KAAK,EAAE;AACf,kBAAQ,KAAK,EAAE;AAAA,QACjB;AAAA,MACF;AACA,cAAQ,GAAG,CAAC,KAAK,CAAC,EAAE,IAAI,QAAQ,SAAS,OAAO;AAAA,IAClD;AAAA,EACF;AACA,QAAM,aAAa,OAAO,OAAO,OAAO;AACxC,QAAM,QACJ,WAAW,WAAW,IAAI,IAAI,WAAW,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,WAAW;AAEnF,QAAM,oBAAoB,aACvB,IAAI,CAAC,UAAU;AACd,UAAM,eAAe,MAAM,IAAI,KAAK;AACpC,UAAM,SAAS,aAAa,IAAI,CAAC,MAAM,EAAE,KAAK;AAC9C,UAAM,QAAQ,KAAK,IAAI,GAAG,MAAM,IAAI,KAAK,IAAI,GAAG,MAAM;AACtD,WAAO,EAAE,OAAO,SAAS,cAAc,MAAM;AAAA,EAC/C,CAAC,EACA,KAAK,CAAC,GAAG,MAAM,EAAE,QAAQ,EAAE,KAAK,EAChC,MAAM,GAAG,EAAE;AAEd,SAAO;AAAA,IACL,QAAQ,OAAO;AAAA,IACf,cAAc,aAAa;AAAA,IAC3B;AAAA,IACA;AAAA,IACA;AAAA,EACF;AACF;AAEA,SAAS,QAAQ,GAAa,GAAqB;AACjD,MAAI,EAAE,WAAW,EAAE,UAAU,EAAE,WAAW,EAAG,QAAO;AACpD,QAAM,IAAI,EAAE;AACZ,QAAM,QAAQ,EAAE,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI;AAC7C,QAAM,QAAQ,EAAE,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI;AAC7C,MAAI,MAAM;AACV,MAAI,SAAS;AACb,MAAI,SAAS;AACb,WAAS,IAAI,GAAG,IAAI,GAAG,KAAK;AAC1B,UAAM,KAAK,EAAE,CAAC,IAAK;AACnB,UAAM,KAAK,EAAE,CAAC,IAAK;AACnB,WAAO,KAAK;AACZ,cAAU,KAAK;AACf,cAAU,KAAK;AAAA,EACjB;AACA,QAAM,QAAQ,KAAK,KAAK,SAAS,MAAM;AACvC,SAAO,UAAU,IAAI,IAAI,MAAM;AACjC;AAIA,SAAS,YACP,MACA,YACA,aACA,OACyB;AACzB,MAAI,MAAM;AACV,MAAI,MAAM;AACV,MAAI,CAAC,OAAO,CAAC,KAAK;AAGhB,UAAM,MAAM,CAAC,GAAG,IAAI,IAAI,KAAK,IAAI,CAAC,MAAM,EAAE,WAAW,CAAC,CAAC;AACvD,QAAI,IAAI,WAAW,EAAG,QAAO;AAC7B,UAAM,CAAC,KAAK,GAAG,IAAI;AACnB,UAAM,QAAQ,KAAK,KAAK,OAAO,CAAC,MAAM,EAAE,gBAAgB,GAAG,EAAE,IAAI,CAAC,MAAM,YAAY,GAAG,KAAK,CAAC,CAAC;AAC9F,UAAM,QAAQ,KAAK,KAAK,OAAO,CAAC,MAAM,EAAE,gBAAgB,GAAG,EAAE,IAAI,CAAC,MAAM,YAAY,GAAG,KAAK,CAAC,CAAC;AAC9F,UAAM,SAAS,QAAQ,MAAM;AAC7B,UAAM,SAAS,QAAQ,MAAM;AAAA,EAC/B;AAEA,QAAM,WAAW,KAAK,OAAO,CAAC,MAAM,EAAE,gBAAgB,GAAG;AACzD,QAAM,YAAY,KAAK,OAAO,CAAC,MAAM,EAAE,gBAAgB,GAAG;AAC1D,MAAI,SAAS,WAAW,KAAK,UAAU,WAAW,EAAG,QAAO;AAI5D,QAAM,gBAAgB,IAAI,IAAI,SAAS,IAAI,CAAC,MAAM,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC;AACrE,QAAM,iBAA2B,CAAC;AAClC,QAAM,kBAA4B,CAAC;AACnC,MAAI,iBAAiB;AACrB,aAAW,QAAQ,WAAW;AAC5B,UAAM,IAAI,cAAc,IAAI,WAAW,IAAI,CAAC;AAC5C,QAAI,GAAG;AACL,YAAM,KAAK,YAAY,GAAG,KAAK;AAC/B,YAAM,KAAK,YAAY,MAAM,KAAK;AAClC,UAAI,OAAO,SAAS,EAAE,KAAK,OAAO,SAAS,EAAE,GAAG;AAC9C,uBAAe,KAAK,EAAE;AACtB,wBAAgB,KAAK,EAAE;AACvB,yBAAiB;AAAA,MACnB;AAAA,IACF;AAAA,EACF;AACA,MAAI,CAAC,gBAAgB;AACnB,UAAM,IAAI,KAAK,IAAI,SAAS,QAAQ,UAAU,MAAM;AACpD,aAAS,IAAI,GAAG,IAAI,GAAG,KAAK;AAC1B,YAAM,KAAK,YAAY,SAAS,CAAC,GAAI,KAAK;AAC1C,YAAM,KAAK,YAAY,UAAU,CAAC,GAAI,KAAK;AAC3C,UAAI,OAAO,SAAS,EAAE,KAAK,OAAO,SAAS,EAAE,GAAG;AAC9C,uBAAe,KAAK,EAAE;AACtB,wBAAgB,KAAK,EAAE;AAAA,MACzB;AAAA,IACF;AAAA,EACF;AACA,MAAI,eAAe,WAAW,EAAG,QAAO;AAExC,QAAM,eAAe,KAAK,cAAc;AACxC,QAAM,gBAAgB,KAAK,eAAe;AAC1C,QAAM,QAAQ,gBAAgB;AAE9B,QAAM,YAAY,gBAAgB,gBAAgB,iBAAiB;AAAA,IACjE,YAAY;AAAA,IACZ,WAAW;AAAA,IACX,WAAW;AAAA,EACb,CAAC;AACD,QAAM,QAAQ,YAAY,gBAAgB,eAAe;AACzD,QAAM,IAAI,QAAQ,gBAAgB,eAAe;AACjD,QAAM,MAAM,UAAU,EAAE,SAAS,eAAe,QAAQ,OAAO,KAAK,OAAO,KAAK,CAAC;AACjF,QAAM,YAAY,mBAAmB;AAAA,IACnC,QAAQ,KAAK,IAAI,KAAK,IAAI,KAAK,GAAG,IAAI;AAAA,IACtC,OAAO;AAAA,IACP,OAAO;AAAA,EACT,CAAC;AAED,SAAO;AAAA,IACL;AAAA,IACA;AAAA,IACA;AAAA,IACA,MAAM,CAAC,UAAU,KAAK,UAAU,IAAI;AAAA,IACpC,QAAQ,MAAM;AAAA,IACd,GAAG,eAAe;AAAA,IAClB,SAAS;AAAA,IACT;AAAA,IACA;AAAA,EACF;AACF;AAEA,SAAS,WAAW,GAAsB;AACxC,SAAO,GAAG,EAAE,YAAY,KAAK,EAAE,IAAI;AACrC;AAEA,SAAS,KAAK,KAAuB;AACnC,SAAO,IAAI,WAAW,IAAI,IAAI,IAAI,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,IAAI;AACrE;AAIA,eAAe,uBACb,MACA,SACA,OAC4C;AAC5C,QAAM,SAAS,KAAK,OAAO,CAAC,MAAM,YAAY,GAAG,KAAK,IAAI,OAAO,EAAE,gBAAgB,MAAS;AAC5F,MAAI,OAAO,WAAW,EAAG,QAAO,EAAE,UAAU,CAAC,GAAG,eAAe,EAAE;AAEjE,QAAM,WAAW,oBAAI,IAAoD;AACzE,aAAW,OAAO,QAAQ;AACxB,QAAI;AACF,YAAM,SAAS,MAAM,QAAQ,IAAI,IAAI,OAAO;AAAA,QAC1C,MAAM;AAAA,QACN;AAAA,MACF,CAAsC;AACtC,iBAAW,WAAW,OAAO,UAA8B;AACzD,cAAM,MAAM,QAAQ,QAAQ,QAAQ,cAAc;AAClD,cAAM,IAAI,SAAS,IAAI,GAAG,KAAK,EAAE,WAAW,CAAC,GAAG,OAAO,EAAE;AACzD,YAAI,EAAE,UAAU,SAAS,EAAG,GAAE,UAAU,KAAK,IAAI,KAAK;AACtD,iBAAS,IAAI,KAAK,CAAC;AAAA,MACrB;AAAA,IACF,QAAQ;AACN,YAAM,IAAI,SAAS,IAAI,eAAe,KAAK,EAAE,WAAW,CAAC,GAAG,OAAO,EAAE;AACrE,UAAI,EAAE,UAAU,SAAS,EAAG,GAAE,UAAU,KAAK,IAAI,KAAK;AACtD,eAAS,IAAI,iBAAiB,CAAC;AAAA,IACjC;AAAA,EACF;AACA,QAAM,cAAc,CAAC,GAAG,SAAS,QAAQ,CAAC,EAAE,IAAI,CAAC,CAAC,IAAI,CAAC,OAAO;AAAA,IAC5D;AAAA,IACA,MAAM;AAAA,IACN,OAAO,EAAE,UAAU,SAAS,OAAO;AAAA,IACnC,WAAW,EAAE;AAAA,EACf,EAAE;AACF,cAAY,KAAK,CAAC,GAAG,MAAM,EAAE,QAAQ,EAAE,KAAK;AAC5C,SAAO,EAAE,UAAU,aAAa,eAAe,OAAO,OAAO;AAC/D;AAIA,SAAS,qBACP,MACA,UACgC;AAChC,MAAI,QAAQ;AACZ,QAAM,UAAqE,CAAC;AAC5E,aAAW,OAAO,MAAM;AACtB,UAAM,SAAS,gBAAgB,GAAG;AAClC,QAAI,CAAC,OAAQ;AACb,UAAM,YAAY,cAAc,QAAQ,QAAQ;AAChD,eAAW,QAAQ,WAAW;AAC5B;AACA,cAAQ,KAAK,EAAE,OAAO,IAAI,OAAO,QAAQ,KAAK,QAAQ,SAAS,KAAK,SAAS,CAAC;AAAA,IAChF;AAAA,EACF;AACA,SAAO,EAAE,OAAO,oBAAoB,UAAU,GAAG,QAAQ;AAC3D;AAEA,SAAS,gBAAgB,KAAoC;AAO3D,QAAM,WAAY,IAA0D;AAC5E,MAAI,OAAO,UAAU,WAAW,SAAU,QAAO,SAAS;AAC1D,MAAI,OAAO,UAAU,SAAS,SAAU,QAAO,SAAS;AACxD,SAAO;AACT;AAIA,SAAS,0BACP,MACA,SACA,OACuC;AACvC,QAAM,KAAe,CAAC;AACtB,QAAM,KAAe,CAAC;AACtB,aAAW,OAAO,MAAM;AACtB,UAAM,IAAI,QAAQ,aAAa,IAAI,KAAK;AACxC,QAAI,MAAM,UAAa,CAAC,OAAO,SAAS,CAAC,EAAG;AAC5C,UAAM,IAAI,YAAY,KAAK,KAAK;AAChC,QAAI,CAAC,OAAO,SAAS,CAAC,EAAG;AACzB,OAAG,KAAK,CAAC;AACT,OAAG,KAAK,CAAC;AAAA,EACX;AACA,MAAI,GAAG,SAAS,EAAG,QAAO;AAE1B,QAAM,IAAI,QAAQ,IAAI,EAAE;AACxB,QAAM,IAAI,SAAS,IAAI,EAAE;AACzB,QAAM,QAAQ,KAAK,EAAE;AACrB,QAAM,QAAQ,KAAK,EAAE;AACrB,MAAI,MAAM;AACV,MAAI,QAAQ;AACZ,WAAS,IAAI,GAAG,IAAI,GAAG,QAAQ,KAAK;AAClC,YAAQ,GAAG,CAAC,IAAK,UAAU,GAAG,CAAC,IAAK;AACpC,cAAU,GAAG,CAAC,IAAK,UAAU;AAAA,EAC/B;AACA,QAAM,QAAQ,UAAU,IAAI,IAAI,MAAM;AACtC,QAAM,YAAY,QAAQ,QAAQ;AAClC,QAAM,QAAQ,GAAG,OAAO,CAAC,GAAG,MAAM,KAAK,IAAI,UAAU,GAAG,CAAC;AACzD,QAAM,QAAQ,GAAG,OAAO,CAAC,GAAG,GAAG,MAAM,KAAK,KAAK,YAAY,QAAQ,GAAG,CAAC,OAAQ,GAAG,CAAC;AACnF,QAAM,KAAK,UAAU,IAAI,IAAI,IAAI,QAAQ;AAEzC,SAAO;AAAA,IACL,QAAQ,QAAQ;AAAA,IAChB,GAAG,GAAG;AAAA,IACN,SAAS;AAAA,IACT,UAAU;AAAA,IACV,aAAa,EAAE,WAAW,OAAO,GAAG;AAAA,EACtC;AACF;AAEA,SAAS,SAAS,GAAa,GAAqB;AAClD,MAAI,EAAE,WAAW,EAAE,UAAU,EAAE,WAAW,EAAG,QAAO;AACpD,SAAO,QAAQ,KAAK,CAAC,GAAG,KAAK,CAAC,CAAC;AACjC;AAEA,SAAS,KAAK,KAAyB;AACrC,QAAM,UAAU,IAAI,IAAI,CAAC,GAAGC,QAAO,EAAE,GAAG,GAAAA,GAAE,EAAE;AAC5C,UAAQ,KAAK,CAAC,GAAG,MAAM,EAAE,IAAI,EAAE,CAAC;AAChC,QAAM,QAAQ,IAAI,MAAM,IAAI,MAAM,EAAE,KAAK,CAAC;AAC1C,MAAI,IAAI;AACR,SAAO,IAAI,QAAQ,QAAQ;AACzB,QAAI,IAAI;AACR,WAAO,IAAI,IAAI,QAAQ,UAAU,QAAQ,IAAI,CAAC,EAAG,MAAM,QAAQ,CAAC,EAAG,EAAG;AACtE,UAAM,OAAO,IAAI,IAAI,KAAK;AAC1B,aAAS,IAAI,GAAG,KAAK,GAAG,IAAK,OAAM,QAAQ,CAAC,EAAG,CAAC,IAAI;AACpD,QAAI,IAAI;AAAA,EACV;AACA,SAAO;AACT;AAIA,SAAS,sBACP,WACA,MACA,eAC0B;AAO1B,QAAM,OAAyC,CAAC;AAChD,QAAM,WACJ,SAAS,UAAa,KAAK,KAAK,CAAC,IAAI,IAChC,SACD,KAAK,QAAQ,IACV,SACA;AACT,OAAK,KAAK;AAAA,IACR,MAAM;AAAA,IACN,QAAQ;AAAA,IACR,QAAQ,OACJ,SAAS,KAAK,MAAM,QAAQ,CAAC,CAAC,WAAW,KAAK,KAAK,CAAC,EAAE,QAAQ,CAAC,CAAC,KAAK,KAAK,KAAK,CAAC,EAAE,QAAQ,CAAC,CAAC,QAAQ,KAAK,CAAC,KAC1G;AAAA,EACN,CAAC;AACD,QAAM,aACJ,kBAAkB,UAAa,cAAc,UAAU,IAAK,SAAoB;AAClF,OAAK,KAAK;AAAA,IACR,MAAM;AAAA,IACN,QAAQ;AAAA,IACR,QAAQ,gBAAgB,GAAG,cAAc,KAAK,oBAAoB;AAAA,EACpE,CAAC;AACD,OAAK,KAAK;AAAA,IACR,MAAM;AAAA,IACN,QAAQ,UAAU,QAAQ,MAAM,SAAS,UAAU,QAAQ,MAAM,SAAS;AAAA,IAC1E,QAAQ,QAAQ,UAAU,KAAK,QAAQ,CAAC,CAAC,SAAS,UAAU,IAAI,QAAQ,CAAC,CAAC,SAAS,UAAU,IAAI,QAAQ,CAAC,CAAC,WAAW,UAAU,CAAC;AAAA,EACnI,CAAC;AACD,QAAM,SAAS,KAAK,KAAK,CAAC,MAAM,EAAE,WAAW,MAAM,IAC/C,SACA,KAAK,KAAK,CAAC,MAAM,EAAE,WAAW,MAAM,IAClC,SACA;AACN,SAAO;AAAA,IACL;AAAA,IACA;AAAA,IACA,QAAQ,CAAC;AAAA,EACX;AACF;AAgBA,SAAS,qBAAqB,KAA8C;AAC1E,QAAM,MAAwB,CAAC;AAI/B,MAAI,IAAI,uBAAuB;AAC7B,UAAM,MAAM,IAAI;AAChB,UAAM,QAAQ,IAAI,eAAe;AACjC,eAAW,QAAQ,IAAI,kBAAkB;AACvC,YAAM,IAAI,IAAI,QAAQ,IAAI;AAC1B,UAAI,CAAC,EAAG;AACR,UAAI,KAAK;AAAA,QACP,UAAU;AAAA,QACV,MAAM;AAAA,QACN,OAAO,GAAG,IAAI,mBAAmB,EAAE,SAAS,QAAQ,CAAC,CAAC,WAAM,EAAE,QAAQ,QAAQ,CAAC,CAAC,OAAO,KAAK;AAAA,QAC5F,QAAQ,iBAAiB,EAAE,KAAK,CAAC,EAAE,QAAQ,CAAC,CAAC,KAAK,EAAE,KAAK,CAAC,EAAE,QAAQ,CAAC,CAAC,QAAQ,EAAE,OAAO,QAAQ,CAAC,CAAC,eAAe,EAAE,QAAQ,QAAQ,CAAC,CAAC,eAAe,EAAE,QAAQ,gBAAgB,EAAE,SAAS;AAAA,QACxL,cAAc,iCAAiC,IAAI;AAAA,MACrD,CAAC;AAAA,IACH;AACA,eAAW,QAAQ,IAAI,iBAAiB;AACtC,YAAM,IAAI,IAAI,QAAQ,IAAI;AAC1B,UAAI,CAAC,EAAG;AACR,UAAI,KAAK;AAAA,QACP,UAAU;AAAA,QACV,MAAM;AAAA,QACN,OAAO,GAAG,IAAI,kBAAkB,EAAE,SAAS,QAAQ,CAAC,CAAC,WAAM,EAAE,QAAQ,QAAQ,CAAC,CAAC,OAAO,KAAK;AAAA,QAC3F,QAAQ,iBAAiB,EAAE,KAAK,CAAC,EAAE,QAAQ,CAAC,CAAC,KAAK,EAAE,KAAK,CAAC,EAAE,QAAQ,CAAC,CAAC,QAAQ,EAAE,OAAO,QAAQ,CAAC,CAAC,eAAe,EAAE,QAAQ,QAAQ,CAAC,CAAC,eAAe,EAAE,QAAQ,gBAAgB,EAAE,SAAS;AAAA,QACxL,cAAc,iCAAiC,IAAI;AAAA,MACrD,CAAC;AAAA,IACH;AAAA,EACF;AAKA,MAAI,IAAI,UAAU,IAAI,GAAG;AACvB,QAAI,IAAI,UAAU,OAAO,KAAK;AAC5B,YAAM,OAAO,IAAI,UAAU,YAAY,CAAC;AACxC,YAAM,QAAQ,KACX,MAAM,GAAG,CAAC,EACV,IAAI,CAAC,MAAM,GAAG,EAAE,KAAK,IAAI,EAAE,MAAM,QAAQ,CAAC,CAAC,EAAE,EAC7C,KAAK,IAAI;AACZ,UAAI,KAAK;AAAA,QACP,UAAU;AAAA,QACV,MAAM;AAAA,QACN,OAAO,kBAAkB,IAAI,UAAU,KAAK,QAAQ,CAAC,CAAC;AAAA,QACtD,QACE,KAAK,SAAS,IACV,SAAS,KAAK,MAAM,OAAO,KAAK,WAAW,IAAI,KAAK,GAAG,sBAAsB,KAAK,mBAAmB,IAAI,UAAU,IAAI,QAAQ,CAAC,CAAC,SAAS,IAAI,UAAU,IAAI,QAAQ,CAAC,CAAC,MACtK,iBAAiB,IAAI,UAAU,IAAI,QAAQ,CAAC,CAAC,SAAS,IAAI,UAAU,IAAI,QAAQ,CAAC,CAAC;AAAA,QACxF,cAAc;AAAA,MAChB,CAAC;AAAA,IACH,WAAW,IAAI,UAAU,OAAO,KAAK;AACnC,YAAM,OAAO,IAAI,UAAU,YAAY,CAAC;AACxC,YAAM,QAAQ,KACX,MAAM,GAAG,CAAC,EACV,IAAI,CAAC,MAAM,GAAG,EAAE,KAAK,IAAI,EAAE,MAAM,QAAQ,CAAC,CAAC,EAAE,EAC7C,KAAK,IAAI;AACZ,UAAI,KAAK;AAAA,QACP,UAAU;AAAA,QACV,MAAM;AAAA,QACN,OAAO,kBAAkB,IAAI,UAAU,KAAK,QAAQ,CAAC,CAAC;AAAA,QACtD,QACE,KAAK,SAAS,IACV,SAAS,KAAK,MAAM,OAAO,KAAK,WAAW,IAAI,KAAK,GAAG,KAAK,KAAK,mBAAmB,IAAI,UAAU,IAAI,QAAQ,CAAC,CAAC,SAAS,IAAI,UAAU,IAAI,QAAQ,CAAC,CAAC,MACrJ,iBAAiB,IAAI,UAAU,IAAI,QAAQ,CAAC,CAAC,SAAS,IAAI,UAAU,IAAI,QAAQ,CAAC,CAAC;AAAA,QACxF,cAAc;AAAA,MAChB,CAAC;AAAA,IACH;AAAA,EACF;AAKA,MAAI,OAAO,KAAK,IAAI,MAAM,EAAE,WAAW,KAAK,IAAI,UAAU,IAAI,GAAG;AAC/D,QAAI,KAAK;AAAA,MACP,UAAU;AAAA,MACV,MAAM;AAAA,MACN,OAAO;AAAA,MACP,QACE;AAAA,MACF,cAAc;AAAA,IAChB,CAAC;AAAA,EACH;AAEA,MAAI,IAAI,MAAM;AACZ,UAAM,WAAW,IAAI,KAAK,KAAK,CAAC,IAAI,IAAI;AACxC,UAAM,eAAe,IAAI,KAAK,KAAK,CAAC,KAAK,IAAI,aAAa,IAAI,KAAK,KAAK,CAAC,IAAI,IAAI;AACjF,QAAI,UAAU;AACZ,UAAI,KAAK;AAAA,QACP,UAAU;AAAA,QACV,MAAM;AAAA,QACN,OAAO,oBAAe,IAAI,KAAK,MAAM,QAAQ,CAAC,CAAC,YAAY,IAAI,KAAK,KAAK,CAAC,EAAE,QAAQ,CAAC,CAAC,KAAK,IAAI,KAAK,KAAK,CAAC,EAAE,QAAQ,CAAC,CAAC;AAAA,QACtH,QAAQ,kCAAkC,IAAI,SAAS,qCAAqC,IAAI,KAAK,CAAC,OAAO,IAAI,KAAK,OAAO,QAAQ,CAAC,CAAC,OAAO,IAAI,KAAK,QAAQ,QAAQ,CAAC,CAAC;AAAA,QACzK,cAAc;AAAA,MAChB,CAAC;AAAA,IACH,WAAW,cAAc;AACvB,UAAI,KAAK;AAAA,QACP,UAAU;AAAA,QACV,MAAM;AAAA,QACN,OAAO,6BAAwB,IAAI,KAAK,SAAS,sBAAsB,IAAI,KAAK,CAAC;AAAA,QACjF,QAAQ,uDAAuD,IAAI,KAAK,IAAI,QAAQ,CAAC,CAAC,uBAAuB,IAAI,KAAK,MAAM,QAAQ,CAAC,CAAC;AAAA,QACtI,cAAc;AAAA,MAChB,CAAC;AAAA,IACH,OAAO;AACL,UAAI,KAAK;AAAA,QACP,UAAU;AAAA,QACV,MAAM;AAAA,QACN,OAAO,mCAA8B,IAAI,KAAK,KAAK,CAAC,EAAE,QAAQ,CAAC,CAAC,6BAA6B,IAAI,SAAS;AAAA,QAC1G,QAAQ;AAAA,QACR,cAAc;AAAA,MAChB,CAAC;AAAA,IACH;AAAA,EACF;AAEA,MAAI,IAAI,iBAAiB,IAAI,cAAc,QAAQ,GAAG;AACpD,QAAI,KAAK;AAAA,MACP,UAAU;AAAA,MACV,MAAM;AAAA,MACN,OAAO,GAAG,IAAI,cAAc,KAAK,eAAe,IAAI,cAAc,UAAU,IAAI,KAAK,GAAG;AAAA,MACxF,QAAQ;AAAA,MACR,cAAc;AAAA,IAChB,CAAC;AAAA,EACH;AAEA,MAAI,IAAI,cAAc,IAAI,WAAW,QAAQ,KAAK;AAChD,QAAI,KAAK;AAAA,MACP,UAAU;AAAA,MACV,MAAM;AAAA,MACN,OAAO,gCAA2B,IAAI,WAAW,MAAM,QAAQ,CAAC,CAAC;AAAA,MACjE,QAAQ;AAAA,MACR,cAAc;AAAA,IAChB,CAAC;AAAA,EACH;AAEA,MAAI,IAAI,mBAAmB,IAAI,gBAAgB,SAAS,SAAS,GAAG;AAClE,UAAM,MAAM,IAAI,gBAAgB,SAAS,CAAC;AAC1C,QAAI,KAAK;AAAA,MACP,UAAU;AAAA,MACV,MAAM;AAAA,MACN,OAAO,wBAAwB,IAAI,IAAI,MAAM,IAAI,QAAQ,KAAK,QAAQ,CAAC,CAAC;AAAA,MACxE,QAAQ,GAAG,IAAI,gBAAgB,aAAa,4CAA4C,IAAI,UAAU,MAAM,qBAAqB,IAAI,IAAI;AAAA,MACzI,cAAc;AAAA,IAChB,CAAC;AAAA,EACH;AAEA,MAAI,IAAI,sBAAsB,KAAK,IAAI,IAAI,mBAAmB,QAAQ,IAAI,KAAK;AAC7E,QAAI,KAAK;AAAA,MACP,UAAU;AAAA,MACV,MAAM;AAAA,MACN,OAAO,+BAA+B,IAAI,mBAAmB,MAAM,qBAAgB,IAAI,mBAAmB,SAAS,QAAQ,CAAC,CAAC;AAAA,MAC7H,QAAQ,yFAAyF,IAAI,mBAAmB,MAAM,2CAA2C,IAAI,mBAAmB,MAAM;AAAA,MACtM,cAAc;AAAA,IAChB,CAAC;AAAA,EACH;AAEA,SAAO;AACT;;;AC9yBA,SAAS,kBACP,WACA,UAC8C;AAE9C,WAAS,KAAK,GAAmB;AAC/B,QAAI,IAAI,eAAe;AACvB,aAAS,IAAI,GAAG,IAAI,EAAE,QAAQ,KAAK;AACjC,WAAK,EAAE,WAAW,CAAC;AACnB,UAAI,KAAK,KAAK,GAAG,QAAQ,MAAM;AAAA,IACjC;AACA,WAAO;AAAA,EACT;AACA,QAAM,SAAS,CAAC,GAAG,SAAS,EAAE,KAAK,CAAC,GAAG,MAAM,KAAK,EAAE,EAAE,IAAI,KAAK,EAAE,EAAE,CAAC;AACpE,QAAM,WAAW,KAAK,IAAI,GAAG,KAAK,IAAI,OAAO,SAAS,GAAG,KAAK,MAAM,OAAO,SAAS,QAAQ,CAAC,CAAC;AAC9F,SAAO;AAAA,IACL,SAAS,OAAO,MAAM,GAAG,QAAQ;AAAA,IACjC,OAAO,OAAO,MAAM,QAAQ;AAAA,EAC9B;AACF;AAEA,SAAS,cAAc,YAGrB;AACA,QAAM,cAAsC,CAAC;AAC7C,QAAM,SAAmB,CAAC;AAC1B,aAAW,CAAC,IAAI,GAAG,KAAK,OAAO,QAAQ,UAAU,GAAG;AAClD,gBAAY,EAAE,IAAI,IAAI;AACtB,WAAO,KAAK,IAAI,aAAa;AAAA,EAC/B;AACA,SAAO;AAAA,IACL,eAAe,OAAO,WAAW,IAAI,IAAI,OAAO,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,OAAO;AAAA,IACpF;AAAA,EACF;AACF;AAEA,IAAM,8BAA8B;AAAA,EAClC;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF;AA2BA,eAAsB,YACpB,MACkD;AAClD,QAAM,YAAY,KAAK,IAAI;AAE3B,QAAM,SAAS,KAAK,UAAU,CAAC;AAC/B,QAAM,cAAc,OAAO,eAAe;AAC1C,QAAM,iBAAiB,OAAO,kBAAkB;AAChD,QAAM,iBAAiB,OAAO,kBAAkB;AAChD,QAAM,kBAAkB,OAAO,mBAAmB;AAClD,QAAM,cAAc,OAAO;AAE3B,QAAM,kBAAkB,OAAO;AAC/B,QAAM,EAAE,OAAO,QAAQ,IAAI,kBACvB;AAAA,IACE,OAAO,KAAK,UAAU,OAAO,CAAC,MAAM,CAAC,gBAAgB,KAAK,CAAC,MAAM,EAAE,OAAO,EAAE,EAAE,CAAC;AAAA,IAC/E,SAAS;AAAA,EACX,IACA,kBAAkB,KAAK,WAAW,eAAe;AAErD,MAAI,MAAM,WAAW,GAAG;AACtB,UAAM,IAAI;AAAA,MACR;AAAA,IACF;AAAA,EACF;AACA,MAAI,QAAQ,WAAW,GAAG;AACxB,UAAM,IAAI,MAAM,2DAA2D;AAAA,EAC7E;AAEA,QAAM,SACJ,KAAK,UACL,WAAW;AAAA,IACT,KAAK;AAAA,MACH,SAAS,KAAK,KAAK,WAAW;AAAA,MAC9B,QAAQ,KAAK,KAAK,UAAU,QAAQ,IAAI,kBAAkB;AAAA,IAC5D;AAAA,IACA,OAAO,KAAK,KAAK,SAAS;AAAA,IAC1B,QACE,KAAK,gBACL;AAAA,IACF,oBAAoB,KAAK,sBAAsB;AAAA,EACjD,CAAC;AAEH,QAAM,OACJ,KAAK,QACL,sBAA4C;AAAA,IAC1C,kBAAkB;AAAA,IAClB,gBAAgB;AAAA,EAClB,CAAC;AAEH,QAAM,UAAU,KAAK,WAAW,wBAAwB;AACxD,QAAM,SAAS,KAAK,UAAU,qBAAqB,SAAS;AAE5D,MAAI,KAAK,YAAY;AACnB,SAAK,WAAW,EAAE,MAAM,oBAAoB,WAAW,KAAK,UAAU,OAAO,CAAC;AAAA,EAChF;AAEA,QAAM,SAAS,MAAM,mBAAyC;AAAA,IAC5D,WAAW;AAAA,IACX,iBAAiB,KAAK;AAAA,IACtB,qBAAqB,KAAK;AAAA,IAC1B;AAAA,IACA,QAAQ,CAAC,KAAK,KAAK;AAAA,IACnB;AAAA,IACA,gBAAgB;AAAA,IAChB,kBAAkB;AAAA,IAClB;AAAA,IACA,eAAe,KAAK,iBAAiB;AAAA,IACrC,SAAS,KAAK;AAAA,IACd,QAAQ,KAAK;AAAA,IACb;AAAA,IACA;AAAA,IACA;AAAA,IACA,eAAe,KAAK;AAAA,IACpB;AAAA,EACF,CAAC;AAED,QAAM,WAAW,cAAc,OAAO,kBAAkB,WAAW,UAAU;AAC7E,QAAM,cAAc,cAAc,OAAO,gBAAgB,WAAW,UAAU;AAE9E,MAAI,KAAK,YAAY;AACnB,SAAK,WAAW;AAAA,MACd,MAAM;AAAA,MACN,eAAe,SAAS;AAAA,MACxB,YAAY,KAAK,IAAI,IAAI;AAAA,IAC3B,CAAC;AACD,SAAK,WAAW;AAAA,MACd,MAAM;AAAA,MACN,UAAU,OAAO,WAAW;AAAA,MAC5B,MAAM,YAAY,gBAAgB,SAAS;AAAA,IAC7C,CAAC;AAAA,EACH;AAEA,QAAM,YACJ,OAAO,iBAAiB,WAAW,eACnC,OAAO,YAAY;AAAA,IACjB,CAAC,KAAK,QACJ,MAAM,IAAI,SAAS,OAAO,CAAC,GAAG,OAAO,IAAI,GAAG,SAAS,WAAW,cAAc,CAAC;AAAA,IACjF;AAAA,EACF;AAMF,QAAM,UAAU,MAAM,YAAY;AAAA,IAChC,MAAM;AAAA,MACJ,GAAG,kBAAkB,OAAO,iBAAiB,OAAO,YAAY,MAAM;AAAA,MACtE,GAAG,kBAAkB,OAAO,gBAAgB,OAAO,UAAU,MAAM;AAAA,IACrE;AAAA,IACA,qBAAqB;AAAA,IACrB,sBAAsB;AAAA,EACxB,CAAC;AAED,QAAM,UAAmD;AAAA,IACvD;AAAA,IACA,QAAQ;AAAA,MACN,GAAG;AAAA,MACH,SAAS,OAAO;AAAA,IAClB;AAAA,IACA,MAAM,YAAY,gBAAgB,SAAS;AAAA,IAC3C,cAAc,OAAO,WAAW;AAAA,IAChC,qBAAqB,OAAO,YAAY;AAAA,IACxC,YAAY,KAAK,IAAI,IAAI;AAAA,IACzB,cAAc;AAAA,IACd;AAAA,IACA,KAAK;AAAA,EACP;AAKA,MAAI,KAAK,cAAc;AACrB,QAAI;AACF,YAAM,oBAAoB,KAAK,cAAc,MAAM,SAAS,QAAQ,MAAM;AAAA,IAC5E,SAAS,KAAK;AACZ,YAAM,MAAM,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AAE3D,cAAQ,KAAK,mDAAmD,GAAG,EAAE;AAAA,IACvE;AAAA,EACF;AAEA,SAAO;AACT;AAEA,eAAe,oBACb,QACA,MACA,SACA,KACA,QACe;AACf,QAAM,SAAS,mBAAmB,MAAM;AAExC,WAAS,qBACP,OACA,SACA,UACA,YAC2B;AAC3B,UAAM,QAA4B,SAAS,MAAM,IAAI,CAAC,SAAS;AAC7D,YAAM,cAAc,OAAO,OAAO,KAAK,WAAW;AAClD,YAAM,YACJ,YAAY,WAAW,IACnB,IACA,YAAY,OAAO,CAAC,GAAG,MAAM,IAAI,EAAE,WAAW,CAAC,IAAI,YAAY;AACrE,aAAO;AAAA,QACL,YAAY,KAAK;AAAA,QACjB,KAAK,KAAK;AAAA,QACV,eAAe;AAAA,QACf,YAAY,OAAO;AAAA,UACjB,OAAO,QAAQ,KAAK,WAAW,EAAE,IAAI,CAAC,CAAC,MAAM,KAAK,MAAM,CAAC,MAAM,MAAM,UAAU,CAAC;AAAA,QAClF;AAAA,QACA,cAAc,KAAK,SAAS;AAAA,MAC9B;AAAA,IACF,CAAC;AACD,UAAM,gBACJ,MAAM,WAAW,IAAI,IAAI,MAAM,OAAO,CAAC,GAAG,MAAM,IAAI,EAAE,eAAe,CAAC,IAAI,MAAM;AAClF,WAAO;AAAA,MACL;AAAA,MACA,aACE,OAAO,YAAY,WACf,WAAW,OAAO,IAClB,WAAW,KAAK,UAAU,WAAW,EAAE,CAAC;AAAA,MAC9C;AAAA,MACA;AAAA,MACA;AAAA,MACA,SAAS,SAAS,WAAW;AAAA,MAC7B;AAAA,IACF;AAAA,EACF;AAEA,QAAM,cAA2C,CAAC;AAElD,cAAY,KAAK,qBAAqB,GAAG,KAAK,iBAAiB,IAAI,kBAAkB,CAAC,CAAC;AAGvF,aAAW,OAAO,IAAI,aAAa;AACjC,UAAM,SAAS,IAAI,SAAS;AAAA,MAC1B,CAAC,MAAM,MACL,EAAE,SAAS,WAAW,gBAAgB,MACrC,SAAS,UAAa,iBAAiB,EAAE,QAAQ,IAAI,iBAAiB,KAAK,QAAQ,KAChF,IACA;AAAA,MACN,IAAI,SAAS,CAAC;AAAA,IAChB;AACA,QAAI,CAAC,OAAQ;AACb,gBAAY;AAAA,MACV,qBAAqB,IAAI,OAAO,kBAAkB,GAAG,OAAO,SAAS,OAAO,UAAU,CAAC;AAAA,IACzF;AAAA,EACF;AAEA,QAAM,QAAsB;AAAA,IAC1B,OAAO,GAAG,MAAM,IAAI,KAAK,IAAI,CAAC;AAAA,IAC9B;AAAA,IACA,YAAW,oBAAI,KAAK,GAAE,YAAY;AAAA,IAClC,QAAQ;AAAA,IACR,QAAQ,KAAK,gBAAgB,CAAC;AAAA,IAC9B,UAAU,YAAY,CAAC;AAAA,IACvB;AAAA,IACA,cAAc,QAAQ;AAAA,IACtB,aAAa,QAAQ;AAAA,IACrB,cAAc,QAAQ;AAAA,IACtB,iBAAiB,QAAQ;AAAA,IACzB,eAAe,QAAQ;AAAA,EACzB;AAEA,QAAM,OAAO,cAAc,KAAK;AAClC;AAEA,SAAS,iBACP,UACQ;AACR,QAAM,OAAO,OAAO,OAAO,SAAS,WAAW,UAAU;AACzD,SAAO,KAAK,WAAW,IAAI,IAAI,KAAK,OAAO,CAAC,GAAG,MAAM,IAAI,EAAE,eAAe,CAAC,IAAI,KAAK;AACtF;AAEA,SAAS,WAAW,GAAmB;AACrC,MAAI,IAAI,eAAe;AACvB,WAAS,IAAI,GAAG,IAAI,EAAE,QAAQ,KAAK;AACjC,SAAK,EAAE,WAAW,CAAC;AACnB,QAAI,KAAK,KAAK,GAAG,QAAQ,MAAM;AAAA,EACjC;AACA,SAAO,EAAE,SAAS,EAAE,EAAE,SAAS,GAAG,GAAG;AACvC;AAOA,SAAS,kBACP,OACA,aACA,OACa;AACb,SAAO,MAAM,IAAI,CAAC,SAAS;AACzB,UAAM,WAAmD,CAAC;AAC1D,UAAM,kBAA8D,CAAC;AACrE,QAAI,eAAe;AACnB,QAAI,iBAAiB;AACrB,eAAW,CAAC,SAAS,KAAK,KAAK,OAAO,QAAQ,KAAK,WAAW,GAAG;AAC/D,eAAS,OAAO,IAAI,EAAE,GAAG,MAAM,WAAW;AAC1C,iBAAW,CAAC,KAAK,KAAK,KAAK,OAAO,QAAQ,MAAM,UAAU,GAAG;AAC3D,YAAI,CAAC,OAAO,SAAS,KAAK,EAAG;AAC7B,cAAM,QAAQ,gBAAgB,GAAG,KAAK,EAAE,KAAK,GAAG,GAAG,EAAE;AACrD,cAAM,OAAO;AACb,cAAM,KAAK;AACX,wBAAgB,GAAG,IAAI;AAAA,MACzB;AACA,UAAI,OAAO,SAAS,MAAM,SAAS,GAAG;AACpC,wBAAgB,MAAM;AACtB,0BAAkB;AAAA,MACpB;AAAA,IACF;AACA,UAAM,aAAqC,CAAC;AAC5C,eAAW,CAAC,KAAK,EAAE,KAAK,EAAE,CAAC,KAAK,OAAO,QAAQ,eAAe,GAAG;AAC/D,iBAAW,GAAG,IAAI,MAAM,IAAI,IAAI,MAAM;AAAA,IACxC;AACA,UAAM,YAAY,mBAAmB,IAAI,IAAI,eAAe;AAC5D,UAAM,cAAiC;AAAA,MACrC;AAAA,MACA;AAAA,MACA;AAAA,IACF;AACA,WAAO;AAAA,MACL,OAAO,GAAG,KAAK,KAAK,WAAW,KAAK,KAAK,MAAM;AAAA,MAC/C,cAAc;AAAA,MACd;AAAA;AAAA;AAAA,MAGA,MACE,KAAK,MAAM,MACX,WAAW,KAAK,UAAU,EACvB,MAAM,GAAG,CAAC,EACV,MAAM,EAAE,EACR,OAAO,CAAC,GAAG,MAAO,IAAI,KAAK,EAAE,WAAW,CAAC,MAAO,GAAG,CAAC;AAAA,MACzD,OAAO;AAAA,MACP,YAAY;AAAA,MACZ,YAAY;AAAA,MACZ,WAAW;AAAA,MACX,QAAQ,KAAK;AAAA,MACb,SAAS,KAAK;AAAA,MACd,YAAY,EAAE,OAAO,GAAG,QAAQ,EAAE;AAAA,MAClC,SAAS;AAAA,QACP,cAAc;AAAA,QACd,KAAK,CAAC;AAAA,QACN;AAAA,MACF;AAAA,MACA,UAAU;AAAA,MACV,GAAI,KAAK,QAAQ,EAAE,aAAa,KAAK,MAAM,IAAI,CAAC;AAAA,IAClD;AAAA,EACF,CAAC;AACH;;;ACvgBO,SAAS,kBAAkB,MAAyD;AACzF,QAAM,EAAE,SAAS,OAAO,CAAC,GAAG,OAAO,kBAAkB,KAAK,IAAI;AAC9D,QAAM,YAAY,IAAI,IAAI,KAAK,IAAI,CAAC,MAAM,CAAC,EAAE,OAAO,CAAC,CAAC,CAAC;AAGvD,QAAM,YAAY,CAAC,WAAqC;AACtD,QAAI,OAAO,WAAW,UAAW,QAAO,SAAS,IAAI;AACrD,QAAI,CAAC,OAAO,SAAS,MAAM,EAAG,QAAO,OAAO;AAC5C,QAAI,CAAC,MAAO,QAAO;AACnB,UAAM,EAAE,KAAK,IAAI,IAAI;AACrB,QAAI,QAAQ,IAAK,QAAO;AACxB,YAAQ,SAAS,QAAQ,MAAM;AAAA,EACjC;AAGA,QAAM,QAAQ,oBAAI,IAAgC;AAClD,aAAW,OAAO,SAAS;AACzB,UAAM,OAAO,MAAM,IAAI,IAAI,KAAK,KAAK,CAAC;AACtC,SAAK,KAAK,GAAG;AACb,UAAM,IAAI,IAAI,OAAO,IAAI;AAAA,EAC3B;AAEA,QAAM,OAAoB,CAAC;AAC3B,QAAM,cAAsD,CAAC;AAE7D,aAAW,CAAC,OAAO,UAAU,KAAK,OAAO;AACvC,UAAM,aAAa,WAChB,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,OAAO,OAAO,UAAU,EAAE,MAAM,EAAE,EAAE,EAC3D,OAAO,CAAC,MAAM,OAAO,SAAS,EAAE,KAAK,CAAC;AACzC,QAAI,WAAW,WAAW,EAAG;AAE7B,UAAM,YAAY,WAAW,OAAO,CAAC,GAAG,MAAM,IAAI,EAAE,OAAO,CAAC,IAAI,WAAW;AAE3E,UAAM,UAAU,UAAU,IAAI,KAAK,KAAM,EAAE,MAAM;AAEjD,UAAM,cAAiC;AAAA,MACrC,UAAU,OAAO,YAAY,WAAW,IAAI,CAAC,MAAM,CAAC,EAAE,OAAO,EAAE,QAAQ,EAAE,MAAM,CAAC,CAAC,CAAC;AAAA,MAClF,YAAY,EAAE,QAAQ,UAAU;AAAA,MAChC,WAAW;AAAA,IACb;AAEA,UAAM,UAAsB;AAAA;AAAA;AAAA;AAAA,MAI1B,cAAc;AAAA,MACd,KAAK,OAAO,YAAY,WAAW,IAAI,CAAC,MAAM,CAAC,SAAS,EAAE,KAAK,IAAI,EAAE,KAAK,CAAC,CAAC;AAAA,MAC5E;AAAA,IACF;AAEA,SAAK,KAAK;AAAA,MACR;AAAA,MACA,cAAc,QAAQ,gBAAgB;AAAA,MACtC,aAAa,QAAQ,eAAe;AAAA,MACpC,MAAM;AAAA,MACN,OAAO,QAAQ,SAAS;AAAA,MACxB,YAAY,QAAQ,cAAc;AAAA,MAClC,YAAY,QAAQ,cAAc;AAAA,MAClC,WAAW,QAAQ,aAAa;AAAA,MAChC,QAAQ,QAAQ,UAAU;AAAA,MAC1B,SAAS,QAAQ,WAAW;AAAA,MAC5B,YAAY,EAAE,OAAO,GAAG,QAAQ,EAAE;AAAA,MAClC;AAAA,MACA,UAAU,QAAQ,YAAY;AAAA,IAChC,CAAc;AAEd,QAAI,iBAAiB;AACnB,iBAAW,KAAK,WAAY,aAAY,KAAK,EAAE,OAAO,OAAO,EAAE,OAAO,OAAO,EAAE,MAAM,CAAC;AAAA,IACxF;AAAA,EACF;AAEA,SAAO,EAAE,MAAM,YAAY;AAC7B;;;AC5HA,IAAM,aAAa,CAAC,gBAAgB,cAAc,OAAO;AACzD,IAAM,aAAa,CAAC,gBAAgB,wBAAwB,aAAa,OAAO;AAChF,IAAM,YAAY,CAAC,mBAAmB,yBAAyB,YAAY,MAAM;AACjF,IAAM,mBAAmB,CAAC,6BAA6B,oBAAoB,WAAW;AACtF,IAAM,oBAAoB,CAAC,8BAA8B,qBAAqB,YAAY;AAC1F,IAAM,mBAAmB,CAAC,sBAAsB,aAAa;AAC7D,IAAM,mBAAmB,CAAC,sBAAsB,aAAa;AAUtD,SAAS,cAAc,MAAyC;AACrE,QAAM,EAAE,OAAO,eAAe,WAAW,eAAe,cAAc,IAAI;AAC1E,QAAM,UAAU,WAAW,KAAK;AAEhC,QAAM,OAAoB,CAAC;AAC3B,aAAW,CAAC,UAAUC,WAAU,KAAK,SAAS;AAC5C,UAAM,OAAO,SAASA,WAAU;AAChC,QAAI,CAAC,KAAM;AAEX,UAAM,SAAS,KAAK,IAAI,IAAI,KAAK,kBAAkB,KAAK,qBAAqB,GAAS;AACtF,UAAM,QAAQ,eAAeA,aAAY,UAAU,KAAK;AACxD,UAAM,UAAU,eAAeA,aAAY,SAAS,KAAK;AACzD,UAAM,cAAc,eAAeA,aAAY,gBAAgB,KAAK;AACpE,UAAM,eAAe,eAAeA,aAAY,iBAAiB,KAAK;AACtE,UAAM,aAAa,eAAeA,aAAY,gBAAgB,KAAK;AACnE,UAAM,aAAa,eAAeA,aAAY,gBAAgB,KAAK;AACnE,UAAM,QAAQ,eAAeA,aAAY,UAAU;AAEnD,UAAM,aAAa,oBAAoBA,WAAU;AACjD,UAAM,aAA4B;AAAA,MAChC,OAAO;AAAA,MACP,QAAQ;AAAA,IACV;AAEA,UAAM,cACJ,UAAU,SACN;AAAA,MACE,UAAU,EAAE,gBAAgB,EAAE,MAAM,EAAE;AAAA,MACtC,YAAY,EAAE,MAAM;AAAA,MACpB,WAAW;AAAA,IACb,IACA;AAEN,UAAM,YAAYA,YAAW,KAAK,CAAC,MAAM,EAAE,QAAQ,SAAS,OAAO;AACnE,UAAM,UAAsB;AAAA,MAC1B,GAAI,KAAK,iBAAiB,WAAW,EAAE,aAAa,MAAM,IAAI,EAAE,cAAc,MAAM;AAAA,MACpF,KAAK;AAAA,MACL,GAAI,cAAc,EAAE,YAAY,IAAI,CAAC;AAAA,IACvC;AAEA,SAAK,KAAK;AAAA,MACR,OAAO;AAAA,MACP;AAAA,MACA,aAAc,KAAK,WAAW,oBAAoB,KAA4B;AAAA,MAC9E,MAAM;AAAA,MACN;AAAA,MACA;AAAA,MACA;AAAA,MACA,WAAY,KAAK,WAAW,mBAAmB,KAA4B;AAAA,MAC3E;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA,UAAU;AAAA,MACV,GAAI,YAAY,EAAE,aAAa,UAAU,KAAK,IAAI,CAAC;AAAA,IACrD,CAAc;AAAA,EAChB;AACA,SAAO;AACT;AAIA,SAAS,WAAW,OAAwD;AAC1E,QAAM,IAAI,oBAAI,IAA8B;AAC5C,aAAW,QAAQ,OAAO;AACxB,UAAM,MAAO,KAAK,cAAc,KAA4B,KAAK;AACjE,UAAM,OAAO,EAAE,IAAI,GAAG,KAAK,CAAC;AAC5B,SAAK,KAAK,IAAI;AACd,MAAE,IAAI,KAAK,IAAI;AAAA,EACjB;AACA,SAAO;AACT;AAEA,SAAS,SAAS,OAAqD;AACrE,SAAO,MAAM,KAAK,CAAC,MAAM,CAAC,EAAE,YAAY,KAAK,MAAM,CAAC;AACtD;AAEA,SAAS,eAAe,OAAyB,MAAoC;AACnF,aAAW,QAAQ,OAAO;AACxB,eAAW,OAAO,MAAM;AACtB,YAAM,IAAI,KAAK,WAAW,GAAG;AAC7B,UAAI,OAAO,MAAM,YAAY,EAAE,SAAS,EAAG,QAAO;AAAA,IACpD;AAAA,EACF;AACA,SAAO;AACT;AAEA,SAAS,eAAe,OAAyB,MAAoC;AACnF,aAAW,QAAQ,OAAO;AACxB,eAAW,OAAO,MAAM;AACtB,YAAM,IAAI,KAAK,WAAW,GAAG;AAC7B,UAAI,OAAO,MAAM,YAAY,OAAO,SAAS,CAAC,EAAG,QAAO;AACxD,UAAI,OAAO,MAAM,UAAU;AACzB,cAAM,SAAS,OAAO,CAAC;AACvB,YAAI,OAAO,SAAS,MAAM,EAAG,QAAO;AAAA,MACtC;AAAA,IACF;AAAA,EACF;AACA,SAAO;AACT;AAEA,SAAS,oBAAoB,OAAiD;AAC5E,QAAM,MAA8B,CAAC;AACrC,aAAW,QAAQ,OAAO;AACxB,eAAW,CAAC,GAAG,CAAC,KAAK,OAAO,QAAQ,KAAK,UAAU,GAAG;AACpD,UAAI,OAAO,MAAM,YAAY,OAAO,SAAS,CAAC,EAAG,KAAI,CAAC,IAAI;AAAA,IAC5D;AAAA,EACF;AACA,SAAO;AACT;","names":["cohensD","mean","i","groupSpans"]}

package/dist/hosted/index.d.ts

@@ -1,4 +1,4 @@
-export { E as EvalRunCellScore, d as EvalRunEvent, e as EvalRunGenerationSnapshot, f as EvalRunStatus, g as HOSTED_WIRE_VERSION, H as HostedClient, h as HostedIngestHeaders, a as HostedTenant, i as HostedWireVersion, j as IngestEvalRunsRequest, k as IngestResponse, l as IngestTracesRequest, T as TraceSpanEvent, m as createHostedClient } from '../index-DQHtWQ57.js';
+export { E as EvalRunCellScore, d as EvalRunEvent, e as EvalRunGenerationSnapshot, f as EvalRunStatus, g as HOSTED_WIRE_VERSION, H as HostedClient, h as HostedIngestHeaders, a as HostedTenant, i as HostedWireVersion, j as IngestEvalRunsRequest, k as IngestResponse, l as IngestTracesRequest, T as TraceSpanEvent, m as createHostedClient } from '../index-C7RhhEME.js';
 import '../types-Dbj5gu8n.js';
 import '../summary-report-B7gNRX-r.js';
 import '../run-record-BGY6bHRh.js';

package/dist/{index-DQHtWQ57.d.ts → index-C7RhhEME.d.ts}

@@ -81,6 +81,12 @@ interface InsightReport {
      *  ActionableSideInfo bag) calls `evaluateReleaseConfidence()` directly;
      *  this summary captures the analyzeRuns-derived axes. */
     release: ReleaseSummary;
+    /** Delta vs a prior period when `baselineRuns` is passed. Per-metric
+     *  current vs baseline with Welch CI + Cohen's d + significance flag.
+     *  Answers "did my last change help?" — the customer-conversion question.
+     *  Surfaced metrics: composite, cost, duration, tokenUsage, plus any
+     *  per-dimension judge metric present in both windows. */
+    priorPeriodComparison?: PriorPeriodComparison;
     /** Top-N actionable recommendations, ranked by priority. The packet's
      *  human-readable layer; the numeric sections are the evidence. */
     recommendations: Recommendation[];
@@ -217,6 +223,46 @@ interface ReleaseSummary {
      *  consumers can post-process to populate. */
     issues: string[];
 }
+interface MetricDelta {
+    /** Current-period mean. */
+    current: number;
+    /** Baseline-period mean. */
+    baseline: number;
+    /** current - baseline. Positive means improved (or, for cost/duration,
+     *  the consumer-side interpretation: "higher current" — semantic
+     *  direction depends on the metric). */
+    delta: number;
+    /** Welch 95% confidence interval on the delta. Two-sample, unpaired —
+     *  the baseline and current run sets may have different scenarios. */
+    ci95: [number, number];
+    /** Welch t-test p-value (two-sided). */
+    pValue: number;
+    /** Cohen's d (pooled stddev). Effect size, signed. */
+    cohensD: number;
+    /** Sample sizes. */
+    baselineN: number;
+    currentN: number;
+    /** True when p < 0.05 AND |d| >= 0.2 (small-effect threshold). The
+     *  conjunction prevents large-effect-but-noisy and significant-but-
+     *  tiny from triggering recommendations. */
+    significant: boolean;
+}
+interface PriorPeriodComparison {
+    /** Sample counts. */
+    baselineN: number;
+    currentN: number;
+    /** Optional human-readable label — "vs prior 7 days", "vs v3 release". */
+    windowLabel?: string;
+    /** Every metric we could compare. Keys: 'composite', 'cost', 'duration',
+     *  'tokenUsage' for always-present ones; per-dimension keys when both
+     *  windows have judge scores on the same dimension. */
+    metrics: Record<string, MetricDelta>;
+    /** Metric names where current is significantly WORSE than baseline.
+     *  Direction-aware: for cost/duration, higher current = worse. */
+    regressedMetrics: string[];
+    /** Metric names where current is significantly BETTER than baseline. */
+    improvedMetrics: string[];
+}
 interface Recommendation {
     priority: 'critical' | 'high' | 'medium' | 'low';
     kind: 'ship' | 'hold' | 'investigate' | 'fix' | 'recalibrate' | 'expand-corpus';

package/dist/openapi.json

@@ -2,7 +2,7 @@
   "openapi": "3.1.0",
   "info": {
     "title": "@tangle-network/agent-eval — wire protocol",
-    "version": "0.52.0",
+    "version": "0.53.0",
     "description": "HTTP and stdio RPC interface to agent-eval. The TypeScript runtime is the source of truth; this spec is the contract that cross-language clients (Python, Rust, Go) generate from.\n\nWire-protocol version: 1.0.0. Bumps on breaking changes to request/response schemas.",
     "contact": {
       "name": "Tangle Network",

package/docs/design/self-improvement-protocol.md

@@ -0,0 +1,223 @@
+# Self-improvement protocol — the world-class architecture
+
+**Status:** Strategic design. The artifact that every roadmap entry maps to.
+**Date:** 2026-05-27.
+
+## Thesis
+
+**Self-improvement is a protocol, not a product.** We define the wire formats, surface abstractions, driver interface, gate interface, and insight format. We ship reference implementations. Customers plug in whatever framework, model, or runtime they already use — our infrastructure handles the rigorous middle (analysis, gating, version-safe deployment).
+
+No competitor ships this combination. LangSmith / Braintrust / Phoenix / LangFuse ship tracing. Hermes ships an agent. SkillOpt ships an academic optimizer. Anthropic's Claude Code ships skill-creation. **Nobody ships the protocol.**
+
+## The pipeline as a single abstract flow
+
+```
+┌──────────────────────────────────────────────────────────────────────┐
+│  WHATEVER YOU ALREADY USE                                            │
+│  LangChain · LlamaIndex · Anthropic SDK · OpenAI Assistants ·        │
+│  Hermes · Claude Code · Codex · agent-runtime · your own stack       │
+└─────────────────────────────────┬────────────────────────────────────┘
+                                  │ traces (any format)
+                                  ▼
+┌──────────────────────────────────────────────────────────────────────┐
+│  INGEST — universal trace adapters                                   │
+│  fromOtelSpans · fromFeedbackTable · fromLangChain · fromLlamaIndex ·│
+│  fromAnthropicSDK · fromOpenAISDK · fromHermesProfileLog · BYO       │
+│  → canonical RunRecord[]                                             │
+└─────────────────────────────────┬────────────────────────────────────┘
+                                  ▼
+┌──────────────────────────────────────────────────────────────────────┐
+│  ANALYZE — analyzeRuns({ runs, baselineRuns?, userFeedback? })       │
+│  paired-bootstrap CI · Pareto · failure clusters · prior-period      │
+│  delta · user-corrective-signal extraction · recommendations         │
+│  ← THE STATISTICAL EDGE NOBODY ELSE SHIPS                            │
+└─────────────────────────────────┬────────────────────────────────────┘
+                                  ▼
+┌──────────────────────────────────────────────────────────────────────┐
+│  IMPROVE — selfImprove() closed loop                                 │
+│  gepaDriver · evolutionaryDriver · BYO ImprovementDriver             │
+│  → ProfileDiff (versioned, hashed, content-addressable)              │
+└─────────────────────────────────┬────────────────────────────────────┘
+                                  ▼
+┌──────────────────────────────────────────────────────────────────────┐
+│  GATE — defaultProductionGate (paired-CI) · BYO gate                 │
+│  ship-substrate / ship-harness / merge / inconclusive                │
+│  ← STATISTICALLY STRICTER THAN ANY COMPETITOR                        │
+└─────────────────────────────────┬────────────────────────────────────┘
+                                  ▼
+┌──────────────────────────────────────────────────────────────────────┐
+│  DEPLOY — back into WHATEVER YOU ALREADY USE                         │
+│  agent-runtime · Hermes profile log · LangChain config · custom hook │
+└──────────────────────────────────────────────────────────────────────┘
+```
+
+## The integration promise
+
+Customers pick one of three integration shapes. All three work today (some are aspirational on adapter coverage). Every shape uses the same canonical types underneath.
+
+### Shape A — offline analysis only
+
+You have traces, you want a decision packet. Zero LLM cost. Zero closed loop.
+
+```typescript
+import { fromOtelSpans, analyzeRuns } from '@tangle-network/agent-eval'
+
+const runs = fromOtelSpans({ spans: mySpans })
+const report = await analyzeRuns({ runs })
+// → InsightReport with composite, recommendations, Pareto, ...
+```
+
+Use case: dashboards, weekly post-mortems, "did anything regress" checks. The intelligence-kernel ships this.
+
+### Shape B — closed loop, your runtime
+
+You have an agent, you want to improve it. We provide drivers + gate + insight. You decide when to deploy.
+
+```typescript
+import { selfImprove, gepaDriver } from '@tangle-network/agent-eval'
+
+const result = await selfImprove({
+  scenarios,
+  agent: yourAgent,           // any function (surface, scenario) → artifact
+  judge: yourJudge,           // any function (artifact) → JudgeScore
+  baselineSurface,
+  driver: gepaDriver({ llm, model, target }),
+  budget: { generations: 3, populationSize: 4, holdoutFraction: 0.3 },
+})
+// → SelfImproveResult { baselineHash, diff, winningHash, lift, gateDecision, insight }
+```
+
+Use case: every product agent we ship. Hermes-on-our-sandbox. Claude Code with skills. Anyone wanting "ship if statistically better, else hold."
+
+### Shape C — hosted, cross-language
+
+You stream traces from anywhere, get InsightReports + selfImprove orchestration. Bills usage-based.
+
+```sh
+# Stream traces
+curl https://api.tangle.tools/v1/ingest/otel \
+  -H "Authorization: Bearer ${TANGLE_KEY}" \
+  --data-binary @traces.jsonl
+
+# Get the decision packet
+curl https://api.tangle.tools/v1/insight/${runId}
+
+# Or run a closed-loop campaign
+curl https://api.tangle.tools/v1/improve \
+  -d '{"scenarios": ..., "baselineHash": "...", "budget": {...}}'
+```
+
+Use case: Python customers, Go customers, customers behind firewalls, customers who don't want to operate the substrate.
+
+## The five non-negotiables
+
+The protocol claim only holds if all five of these survive integration. Customers shouldn't have to compromise on any.
+
+1. **Universal ingest.** Any trace format → canonical RunRecord. Coverage: OTel ✓, multi-rater feedback ✓, LangChain ⏳, LlamaIndex ⏳, Anthropic SDK ⏳, OpenAI Assistants ⏳, Hermes profile log ⏳.
+2. **Statistical rigor.** Every claim falsifiable. Paired bootstrap CI on lift, Cohen's d on effect size, MDE-aware sample-size recommendations, p-values. **SkillOpt's gate is literal `cand > current`. Hermes has no gate. Ours has all of the above.** This is the moat.
+3. **Plug-in everything.** Driver, judge, gate, intake adapter, storage all swappable. Customer brings their LLM, their judge, their scenarios. We bring the rigor.
+4. **Version-safe deployment.** AgentProfile is content-addressable. Two writers (harness + substrate) can both mutate without lost-update. Gate verdicts are scoped to baseline hash, not absolute. Tracked as #98.
+5. **Cross-language wire format.** Python client at parity with TypeScript. Hosted ingest spec versioned. Customers in any language consume the same shape.
+
+## Where we are honest about gaps
+
+| Component | Status | Customer impact when missing |
+|---|---|---|
+| `fromOtelSpans` ingest adapter | ✓ shipped 0.50.0 | — |
+| `fromFeedbackTable` multi-rater intake | ✓ shipped 0.50.0 | — |
+| `analyzeRuns` decision packet | ✓ shipped 0.50.0 / 0.50.2 actionability | — |
+| `selfImprove` closed loop | ✓ shipped 0.50.0 | — |
+| Paired-bootstrap gate | ✓ shipped early; still our edge | — |
+| `gepaDriver` reflection (not full Pareto — task #101) | ⚠ partial | OK; customers don't need Pareto until plateau hit |
+| **Prior-period comparison** in `analyzeRuns` | ✗ MISSING | "Did my last change help?" — the #1 customer question — has no rigorous answer today |
+| **User-corrective-feedback signal extraction** | ✗ MISSING | Hermes' first-class skill signal. We have the trace data. We don't mine it. |
+| **`init` CLI** scaffolding canonical eval/ layout | ✗ MISSING | Every new consumer wires it by hand; the skill describes 80 lines they have to copy |
+| **Framework-specific intake adapters** (LangChain, LlamaIndex, Anthropic SDK, OpenAI Assistants) | ✗ MISSING | Customers using these frameworks can't ingest without writing custom adapter code |
+| **Profile versioning** (task #98) | ✗ MISSING | Offline/online drift; gate verdicts can be stale by the time they're applied |
+| **Composite driver** (optimize all surfaces against one gate) | ✗ MISSING | Customers can optimize prompts OR skills, not both jointly |
+| **Empirical proof drivers work** | ✗ MISSING | We've never published "we ran gepaDriver on real customer data, here's the lift CI" |
+| Hosted-tier production launch | ⚠ in scaffolding (intelligence-kernel) | Customers must self-host today |
+
+## The roadmap — what closes each gap
+
+Mapping every roadmap entry back to a concrete protocol gap.
+
+### 0.53.0 (this session-or-next) — answer "did my last change help?"
+
+- **`analyzeRuns({ runs, baselineRuns? })`** — when `baselineRuns` is provided, the report includes a `priorPeriodComparison?` block: per-metric delta with paired-bootstrap CI, MDE-aware significance judgment, "regressed metrics" surfaced in `recommendations`.
+- Built on top of existing `diffRuns()` primitive (already shipped 0.48.0).
+- 1 PR. Pure additive surface.
+- **Customer impact**: this is the conversion question for every prospect.
+
+### 0.54.0 — extract Hermes' missing signal
+
+- **`extractUserCorrections(runs)`** — new substrate primitive. Mines user messages in traces for corrective markers (regex pass + LLM classifier for nuance). Returns `UserCorrectionEvent[]` keyed by runId.
+- `analyzeRuns({ runs, userFeedback? })` includes a "common corrections" cluster in `recommendations`.
+- Bridge to Hermes-style signal without adopting Hermes' runtime.
+- **Customer impact**: distinctive — no competitor mines this signal.
+
+### 0.55.0 — framework-specific intake adapters
+
+- **`fromLangChain(traces)`**, **`fromLlamaIndex(traces)`**, **`fromAnthropicSDK(traces)`**, **`fromOpenAIAssistants(traces)`**.
+- Each maps the framework's native trace shape to RunRecord.
+- Top 4 frameworks = 80% of agent-builder market coverage.
+- **Customer impact**: removes "we don't support your framework" friction.
+
+### 0.56.0 — `init` CLI + worked examples
+
+- `pnpm dlx @tangle-network/agent-eval init` scaffolds the canonical `eval/scenarios.json` + 3 pnpm scripts + judges template + `.runs/` directory.
+- Adds 5+ end-to-end runnable examples covering Shapes A/B/C across the 4 framework adapters.
+- **Customer impact**: time-to-first-eval drops from 4 hours to 5 minutes.
+
+### 1.0.0 — profile versioning (#98) + composite driver
+
+- Content-addressable `AgentProfileVersion` + `ProfileDiff` + 3-way merge + 4-way `DriftGateDecision`.
+- `compositeDriver` — optimize all surfaces of one AgentProfile against one gate.
+- Hermes-on-sandbox forcing function validates the work before commit.
+- **Customer impact**: production-safe; the moat is locked.
+
+### 1.1.0 — empirical-proof publication
+
+- Pick one named customer or one synthetic-realistic corpus (legal-agent canonical).
+- Run gepaDriver end-to-end with real LLM cost.
+- Publish: "n=, lift=, CI=, p=, $cost=, vs no-driver baseline."
+- One blog post, one demo video, one runnable repro.
+- **Customer impact**: every other claim becomes credible because this one is verified.
+
+## Why this design is 100x
+
+Not a 10% improvement over LangSmith. A category change.
+
+| Capability | LangSmith / Braintrust / Phoenix | Hermes / Claude Code | Tangle (target) |
+|---|---|---|---|
+| Trace ingest | ✓ proprietary | ✓ own runtime | ✓ universal |
+| Decision packet | ⚠ scorecards (no CI) | ✗ | ✓ paired-bootstrap |
+| Closed loop | ✗ | ✓ heuristic | ✓ statistically rigorous |
+| Plug-in drivers | ✗ | ✗ | ✓ |
+| Profile versioning | ✗ | ✗ | ✓ (1.0.0) |
+| Composite multi-surface | ✗ | ✗ | ✓ (1.0.0) |
+| Cross-language | ✗ | ✗ | ✓ (Python at parity) |
+| Empirical-proof publication | ✗ | ✗ | ✓ (1.1.0) |
+
+Eight rows. Nobody else has eight. We can be the only one. The work is named, scoped, and queued.
+
+## What's NOT on the roadmap (and why)
+
+- **Building our own agent runtime.** Hermes / agent-runtime / Claude Code cover that. We are infrastructure, not a runtime.
+- **Single-vendor LLM.** Substrate stays model-agnostic.
+- **UI-first product.** API-first. UIs are downstream.
+- **LangChain replacement.** Wrong layer.
+- **"Self-improvement" without a held-out gate.** Hermes and SkillOpt both ship this; we explicitly refuse — every selfImprove() requires a holdout.
+
+## Decision log — what we committed to in 0.52.0 → 1.0.0
+
+1. **`skillOptDriver` removed; behavior in `gepaDriver({ constraints })`** — 0.52.0 ✓ shipped
+2. **Honest spec docs** — 0.52.0 ✓ shipped
+3. **Profile-versioning spec with symmetric-fork framing** — 0.52.0 ✓ shipped
+4. **No V2 names anywhere** — enforced
+5. **Forcing-function gate on profile-versioning work** — Hermes-on-sandbox experiment required before phases 1-5 commit
+6. **Single-PR-per-repo discipline** — enforced 0.52.0 onwards
+7. **Prior-period comparison as 0.53.0** — committed; the customer-conversion primitive
+8. **User-feedback extraction as 0.54.0** — committed; the Hermes-signal bridge
+9. **Framework intake adapters as 0.55.0** — committed; 80% market coverage
+10. **Empirical-proof publication as 1.1.0** — committed; the credibility lock

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tangle-network/agent-eval",
-  "version": "0.52.0",
+  "version": "0.53.0",
   "description": "Substrate for self-improving agents: traces, verifiable rewards, preferences, GEPA / reflective mutation, auto-research, replay, sequential anytime-valid stats, and release gates.",
   "homepage": "https://github.com/tangle-network/agent-eval#readme",
   "repository": {