@tangle-network/agent-eval 0.23.1 → 0.25.0

package/dist/{chunk-IOXMGMHQ.js → chunk-2A5XJB43.js}

@@ -1,263 +1,12 @@
+import {
+  cohensD,
+  confidenceInterval,
+  wilcoxonSignedRank
+} from "./chunk-I4MBDTY5.js";
 import {
   canonicalize,
   hashJson
-} from "./chunk-6M774GY6.js";
-
-// src/statistics.ts
-var INVERTED_DIMENSIONS = /* @__PURE__ */ new Set([
-  "hallucination",
-  "false_confidence",
-  "worst_failure"
-]);
-function normalizeScores(scores) {
-  return scores.map((s) => {
-    if (INVERTED_DIMENSIONS.has(s.dimension)) {
-      return s;
-    }
-    return s;
-  });
-}
-function weightedMean(scores) {
-  if (scores.length === 0) return 0;
-  let totalWeight = 0;
-  let weightedSum = 0;
-  for (const { score, weight } of scores) {
-    const w = weight ?? 1;
-    weightedSum += score * w;
-    totalWeight += w;
-  }
-  return totalWeight > 0 ? weightedSum / totalWeight : 0;
-}
-function confidenceInterval(scores, confidence = 0.95) {
-  if (scores.length === 0) return { mean: 0, lower: 0, upper: 0 };
-  if (scores.length === 1) return { mean: scores[0], lower: scores[0], upper: scores[0] };
-  const n = scores.length;
-  const mean = scores.reduce((a, b) => a + b, 0) / n;
-  const B = 1e3;
-  const bootstrapMeans = [];
-  for (let i = 0; i < B; i++) {
-    let sum = 0;
-    for (let j = 0; j < n; j++) {
-      sum += scores[Math.floor(Math.random() * n)];
-    }
-    bootstrapMeans.push(sum / n);
-  }
-  bootstrapMeans.sort((a, b) => a - b);
-  const alpha = 1 - confidence;
-  const lowerIdx = Math.floor(alpha / 2 * B);
-  const upperIdx = Math.floor((1 - alpha / 2) * B) - 1;
-  return {
-    mean,
-    lower: bootstrapMeans[lowerIdx],
-    upper: bootstrapMeans[Math.min(upperIdx, B - 1)]
-  };
-}
-function interRaterReliability(judgeScores) {
-  if (judgeScores.length < 2) return 1;
-  const dimensionMap = /* @__PURE__ */ new Map();
-  for (const judgeSet of judgeScores) {
-    for (const s of judgeSet) {
-      if (!dimensionMap.has(s.dimension)) dimensionMap.set(s.dimension, []);
-      const arr = dimensionMap.get(s.dimension);
-      if (arr.length === 0 || arr[arr.length - 1].length >= judgeScores.length) {
-        arr.push([s.score]);
-      } else {
-        arr[arr.length - 1].push(s.score);
-      }
-    }
-  }
-  const allValues = [];
-  const pairDiffs = [];
-  for (const items of dimensionMap.values()) {
-    for (const ratings of items) {
-      if (ratings.length < 2) continue;
-      for (const v of ratings) allValues.push(v);
-      for (let i = 0; i < ratings.length; i++) {
-        for (let j = i + 1; j < ratings.length; j++) {
-          pairDiffs.push((ratings[i] - ratings[j]) ** 2);
-        }
-      }
-    }
-  }
-  if (pairDiffs.length === 0 || allValues.length < 2) return 1;
-  const observedDisagreement = pairDiffs.reduce((a, b) => a + b, 0) / pairDiffs.length;
-  let expectedDisagreement = 0;
-  let expectedCount = 0;
-  for (let i = 0; i < allValues.length; i++) {
-    for (let j = i + 1; j < allValues.length; j++) {
-      expectedDisagreement += (allValues[i] - allValues[j]) ** 2;
-      expectedCount++;
-    }
-  }
-  expectedDisagreement = expectedCount > 0 ? expectedDisagreement / expectedCount : 0;
-  if (expectedDisagreement === 0) return 1;
-  return 1 - observedDisagreement / expectedDisagreement;
-}
-function mannWhitneyU(a, b) {
-  if (a.length === 0 || b.length === 0) return { u: 0, p: 1 };
-  const n1 = a.length;
-  const n2 = b.length;
-  const combined = [
-    ...a.map((v) => ({ v, group: "a" })),
-    ...b.map((v) => ({ v, group: "b" }))
-  ].sort((x, y) => x.v - y.v);
-  const ranks = new Array(combined.length);
-  let i = 0;
-  while (i < combined.length) {
-    let j = i;
-    while (j < combined.length && combined[j].v === combined[i].v) j++;
-    const avgRank = (i + 1 + j) / 2;
-    for (let k = i; k < j; k++) ranks[k] = avgRank;
-    i = j;
-  }
-  let r1 = 0;
-  for (let k = 0; k < combined.length; k++) {
-    if (combined[k].group === "a") r1 += ranks[k];
-  }
-  const u1 = r1 - n1 * (n1 + 1) / 2;
-  const u2 = n1 * n2 - u1;
-  const u = Math.min(u1, u2);
-  const mu = n1 * n2 / 2;
-  const sigma = Math.sqrt(n1 * n2 * (n1 + n2 + 1) / 12);
-  if (sigma === 0) return { u, p: 1 };
-  const z = Math.abs(u - mu) / sigma;
-  const p = 2 * (1 - normalCdf(z));
-  return { u, p };
-}
-function partialCredit(current, target) {
-  if (target <= 0) return 1;
-  return Math.min(1, Math.max(0, current / target));
-}
-function pairedTTest(before, after) {
-  if (before.length !== after.length) {
-    throw new Error(`pairedTTest: unequal sample sizes (${before.length} vs ${after.length})`);
-  }
-  const n = before.length;
-  if (n < 2) return { t: 0, df: 0, p: 1 };
-  const diffs = before.map((b, i) => after[i] - b);
-  const mean = diffs.reduce((a, b) => a + b, 0) / n;
-  const variance = diffs.reduce((acc, d) => acc + (d - mean) ** 2, 0) / (n - 1);
-  const se = Math.sqrt(variance / n);
-  if (se === 0) return { t: mean === 0 ? 0 : Infinity, df: n - 1, p: mean === 0 ? 1 : 0 };
-  const t = mean / se;
-  const df = n - 1;
-  const p = 2 * (1 - studentTCdf(Math.abs(t), df));
-  return { t, df, p };
-}
-function wilcoxonSignedRank(before, after) {
-  if (before.length !== after.length) {
-    throw new Error(`wilcoxonSignedRank: unequal sample sizes (${before.length} vs ${after.length})`);
-  }
-  const diffs = before.map((b, i2) => after[i2] - b).filter((d) => d !== 0);
-  const n = diffs.length;
-  if (n < 6) return { w: 0, p: 1 };
-  const absRanks = diffs.map((d, i2) => ({ abs: Math.abs(d), sign: Math.sign(d), i: i2 })).sort((a, b) => a.abs - b.abs);
-  const ranks = new Array(n);
-  let i = 0;
-  while (i < n) {
-    let j = i;
-    while (j < n && absRanks[j].abs === absRanks[i].abs) j++;
-    const avg2 = (i + 1 + j) / 2;
-    for (let k = i; k < j; k++) ranks[absRanks[k].i] = avg2;
-    i = j;
-  }
-  let wPlus = 0;
-  for (let k = 0; k < n; k++) if (diffs[k] > 0) wPlus += ranks[k];
-  const mean = n * (n + 1) / 4;
-  const variance = n * (n + 1) * (2 * n + 1) / 24;
-  const z = (wPlus - mean) / Math.sqrt(variance);
-  const p = 2 * (1 - normalCdf(Math.abs(z)));
-  return { w: wPlus, p };
-}
-function cohensD(a, b) {
-  if (a.length < 2 || b.length < 2) return 0;
-  const meanA = a.reduce((x, y) => x + y, 0) / a.length;
-  const meanB = b.reduce((x, y) => x + y, 0) / b.length;
-  const varA = a.reduce((acc, x) => acc + (x - meanA) ** 2, 0) / (a.length - 1);
-  const varB = b.reduce((acc, x) => acc + (x - meanB) ** 2, 0) / (b.length - 1);
-  const pooled = Math.sqrt(
-    ((a.length - 1) * varA + (b.length - 1) * varB) / (a.length + b.length - 2)
-  );
-  if (pooled === 0) return 0;
-  return (meanB - meanA) / pooled;
-}
-function studentTCdf(t, df) {
-  if (df <= 0) return 0.5;
-  if (df > 100) return normalCdf(t);
-  const x = df / (df + t * t);
-  const a = df / 2;
-  const b = 0.5;
-  const ib = incompleteBeta(x, a, b);
-  return t >= 0 ? 1 - 0.5 * ib : 0.5 * ib;
-}
-function incompleteBeta(x, a, b) {
-  if (x <= 0) return 0;
-  if (x >= 1) return 1;
-  const lnBeta = lnGamma(a) + lnGamma(b) - lnGamma(a + b);
-  const front = Math.exp(Math.log(x) * a + Math.log(1 - x) * b - lnBeta) / a;
-  const maxIter = 200;
-  const eps = 3e-7;
-  let c = 1;
-  let d = 1 - (a + b) * x / (a + 1);
-  if (Math.abs(d) < 1e-30) d = 1e-30;
-  d = 1 / d;
-  let f = d;
-  for (let m = 1; m <= maxIter; m++) {
-    const m2 = 2 * m;
-    let num = m * (b - m) * x / ((a + m2 - 1) * (a + m2));
-    d = 1 + num * d;
-    if (Math.abs(d) < 1e-30) d = 1e-30;
-    c = 1 + num / c;
-    if (Math.abs(c) < 1e-30) c = 1e-30;
-    d = 1 / d;
-    f *= d * c;
-    num = -((a + m) * (a + b + m) * x) / ((a + m2) * (a + m2 + 1));
-    d = 1 + num * d;
-    if (Math.abs(d) < 1e-30) d = 1e-30;
-    c = 1 + num / c;
-    if (Math.abs(c) < 1e-30) c = 1e-30;
-    d = 1 / d;
-    const delta = d * c;
-    f *= delta;
-    if (Math.abs(delta - 1) < eps) break;
-  }
-  return front * f;
-}
-function lnGamma(z) {
-  const g = 7;
-  const coefs = [
-    0.9999999999998099,
-    676.5203681218851,
-    -1259.1392167224028,
-    771.3234287776531,
-    -176.6150291621406,
-    12.507343278686905,
-    -0.13857109526572012,
-    9984369578019572e-21,
-    15056327351493116e-23
-  ];
-  if (z < 0.5) {
-    return Math.log(Math.PI / Math.sin(Math.PI * z)) - lnGamma(1 - z);
-  }
-  z -= 1;
-  let x = coefs[0];
-  for (let i = 1; i < g + 2; i++) x += coefs[i] / (z + i);
-  const t = z + g + 0.5;
-  return 0.5 * Math.log(2 * Math.PI) + (z + 0.5) * Math.log(t) - t + Math.log(x);
-}
-function normalCdf(x) {
-  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 = x < 0 ? -1 : 1;
-  const absX = Math.abs(x);
-  const t = 1 / (1 + p * absX);
-  const y = 1 - ((((a5 * t + a4) * t + a3) * t + a2) * t + a1) * t * Math.exp(-absX * absX / 2);
-  return 0.5 * (1 + sign * y);
-}
+} from "./chunk-4F5DQN55.js";
 
 // src/power-analysis.ts
 function requiredSampleSize(opts) {
@@ -268,7 +17,7 @@ function requiredSampleSize(opts) {
   const twoSided = opts.twoSided ?? true;
   const zAlpha = zQuantile(twoSided ? 1 - alpha / 2 : 1 - alpha);
   const zBeta = zQuantile(power);
-  const n = 2 * Math.pow((zAlpha + zBeta) / effect, 2);
+  const n = 2 * ((zAlpha + zBeta) / effect) ** 2;
   return Math.ceil(n);
 }
 function pairedMde(opts) {
@@ -294,10 +43,11 @@ function benjaminiHochberg(pValues, fdr = 0.05) {
   let minRight = 1;
   for (let k = n - 1; k >= 0; k--) {
     const rank = k + 1;
-    const raw = indexed[k].p * n / rank;
+    const entry = indexed[k];
+    const raw = entry.p * n / rank;
     const bounded = Math.min(minRight, raw);
     minRight = bounded;
-    q[indexed[k].i] = Math.min(1, bounded);
+    q[entry.i] = Math.min(1, bounded);
   }
   const significant = q.map((v) => v < fdr);
   return { qValues: q, significant };
@@ -308,9 +58,29 @@ function zQuantile(p) {
     if (p === 1) return Infinity;
     return NaN;
   }
-  const a = [-39.69683028665376, 220.9460984245205, -275.9285104469687, 138.357751867269, -30.66479806614716, 2.506628277459239];
-  const b = [-54.47609879822406, 161.5858368580409, -155.6989798598866, 66.80131188771972, -13.28068155288572];
-  const c = [-0.007784894002430293, -0.3223964580411365, -2.400758277161838, -2.549732539343734, 4.374664141464968, 2.938163982698783];
+  const a = [
+    -39.69683028665376,
+    220.9460984245205,
+    -275.9285104469687,
+    138.357751867269,
+    -30.66479806614716,
+    2.506628277459239
+  ];
+  const b = [
+    -54.47609879822406,
+    161.5858368580409,
+    -155.6989798598866,
+    66.80131188771972,
+    -13.28068155288572
+  ];
+  const c = [
+    -0.007784894002430293,
+    -0.3223964580411365,
+    -2.400758277161838,
+    -2.549732539343734,
+    4.374664141464968,
+    2.938163982698783
+  ];
   const d = [0.007784695709041462, 0.3224671290700398, 2.445134137142996, 3.754408661907416];
   const pLow = 0.02425;
   const pHigh = 1 - pLow;
@@ -332,9 +102,7 @@ function zQuantile(p) {
 // src/paired-stats.ts
 function pairedBootstrap(before, after, opts = {}) {
   if (before.length !== after.length) {
-    throw new Error(
-      `pairedBootstrap: unequal sample sizes (${before.length} vs ${after.length})`
-    );
+    throw new Error(`pairedBootstrap: unequal sample sizes (${before.length} vs ${after.length})`);
   }
   const confidence = opts.confidence ?? 0.95;
   const resamples = opts.resamples ?? 2e3;
@@ -686,7 +454,9 @@ async function researchReport(runs, opts = {}) {
   const generatedAt = opts.generatedAt ?? (/* @__PURE__ */ new Date()).toISOString();
   const preregistrationHash = opts.preregistrationHash ?? null;
   if (rope && !(Number.isFinite(rope.low) && Number.isFinite(rope.high) && rope.low <= rope.high)) {
-    throw new Error(`researchReport: rope must satisfy low \u2264 high with finite bounds, got ${JSON.stringify(rope)}`);
+    throw new Error(
+      `researchReport: rope must satisfy low \u2264 high with finite bounds, got ${JSON.stringify(rope)}`
+    );
   }
   const summary = summaryTable(runs, {
     comparator: comparator ?? void 0,
@@ -696,24 +466,29 @@ async function researchReport(runs, opts = {}) {
   });
   const pareto = paretoChart(runs, { split, gateDecisions: opts.gateDecisions });
   const candidateIds = opts.candidateIds ?? summary.rows.map((r) => r.candidateId).filter((id) => id !== comparator);
-  const gains = comparator ? candidateIds.map((id) => gainHistogram(runs, id, comparator, {
-    split,
-    confidence,
-    seed: opts.seed
-  })) : [];
+  const gains = comparator ? candidateIds.map(
+    (id) => gainHistogram(runs, id, comparator, {
+      split,
+      confidence,
+      seed: opts.seed
+    })
+  ) : [];
   const gainByCandidate = new Map(gains.map((g) => [g.candidateId, g]));
   const paretoByCandidate = new Map(pareto.points.map((p) => [p.candidateId, p]));
   const posteriorByCandidate = /* @__PURE__ */ new Map();
   if (comparator) {
     for (const id of candidateIds) {
-      posteriorByCandidate.set(id, pairedPosterior(runs, id, comparator, {
-        split,
-        confidence,
-        seed: opts.seed,
-        rope,
-        mdePower,
-        mdeAlpha
-      }));
+      posteriorByCandidate.set(
+        id,
+        pairedPosterior(runs, id, comparator, {
+          split,
+          confidence,
+          seed: opts.seed,
+          rope,
+          mdePower,
+          mdeAlpha
+        })
+      );
     }
   }
   const candidates = summary.rows.map((row) => {
@@ -767,12 +542,23 @@ async function researchReport(runs, opts = {}) {
     failureClusters: opts.failureClusters,
     preregistrationHash
   });
-  const methodology = buildMethodology({ split, comparator, fdr, minPairs, rope, confidence, mdePower, mdeAlpha });
-  const runFingerprint = await hashJson(canonicalize({
-    triples: runs.filter((r) => r.splitTag === split).map((r) => ({ runId: r.runId, candidateId: r.candidateId, splitTag: r.splitTag })).sort((a, b) => a.runId.localeCompare(b.runId)),
+  const methodology = buildMethodology({
+    split,
     comparator,
-    split
-  }));
+    fdr,
+    minPairs,
+    rope,
+    confidence,
+    mdePower,
+    mdeAlpha
+  });
+  const runFingerprint = await hashJson(
+    canonicalize({
+      triples: runs.filter((r) => r.splitTag === split).map((r) => ({ runId: r.runId, candidateId: r.candidateId, splitTag: r.splitTag })).sort((a, b) => a.runId.localeCompare(b.runId)),
+      comparator,
+      split
+    })
+  );
   const markdown = renderResearchMarkdown({
     title,
     generatedAt,
@@ -818,7 +604,9 @@ function buildMethodology(ctx) {
     `Decisions are pre-specified at fdr=${ctx.fdr}, minPairs=${ctx.minPairs}, confidence=${ctx.confidence}; deviating from these post-hoc invalidates the false-discovery control.`
   ];
   if (ctx.rope) {
-    assumptions.push(`The Region of Practical Equivalence ${formatRope(ctx.rope)} is supplied by the domain owner; equivalent verdicts are only meaningful if that range is treated as the standing definition of "no material difference."`);
+    assumptions.push(
+      `The Region of Practical Equivalence ${formatRope(ctx.rope)} is supplied by the domain owner; equivalent verdicts are only meaningful if that range is treated as the standing definition of "no material difference."`
+    );
   }
   if (ctx.comparator === null) {
     assumptions.push("No comparator was configured; this run is descriptive, not causal.");
@@ -884,7 +672,10 @@ function classifyCandidate(row, ctx) {
   const gainPositive = ci.low > 0;
   const gainNegative = ci.high < 0;
   if (gainNegative) {
-    return { decision: "reject", reason: `Paired-delta CI [${fmt(ci.low)}, ${fmt(ci.high)}] lies entirely below zero.` };
+    return {
+      decision: "reject",
+      reason: `Paired-delta CI [${fmt(ci.low)}, ${fmt(ci.high)}] lies entirely below zero.`
+    };
   }
   if (ctx.posterior.n < ctx.minPairs) {
     return {
@@ -916,7 +707,9 @@ function buildRecommendation(candidates, ctx) {
     rationale.push(`${chosen.candidateId}: ${chosen.decisionReason}`);
     if (chosen.gainCi) {
       const probSummary = chosen.prGreaterThanZero !== null ? `, Pr(\u0394>0)=${fmt(chosen.prGreaterThanZero)}` : "";
-      rationale.push(`Median paired gain CI: [${fmt(chosen.gainCi.low)}, ${fmt(chosen.gainCi.high)}]${probSummary}.`);
+      rationale.push(
+        `Median paired gain CI: [${fmt(chosen.gainCi.low)}, ${fmt(chosen.gainCi.high)}]${probSummary}.`
+      );
     }
     if (chosen.mde !== null && Number.isFinite(chosen.mde)) {
       rationale.push(`MDE at current paired N=${chosen.pairedN}: ${fmt(chosen.mde)} score units.`);
@@ -927,22 +720,36 @@ function buildRecommendation(candidates, ctx) {
     nextActions.push("Re-run with a stable comparator candidate for paired inference.");
   }
   if (!ctx.preregistrationHash) {
-    risks.push("No preregistration hash supplied; readers cannot verify the analysis was specified before data inspection.");
-    nextActions.push("Sign a HypothesisManifest before the next sweep and pass `preregistrationHash` so the report cites it.");
+    risks.push(
+      "No preregistration hash supplied; readers cannot verify the analysis was specified before data inspection."
+    );
+    nextActions.push(
+      "Sign a HypothesisManifest before the next sweep and pass `preregistrationHash` so the report cites it."
+    );
   }
   if (ctx.rope === null && nonComparator.length > 0) {
-    risks.push('No ROPE configured; the report cannot distinguish "equivalent" from "inconclusive".');
-    nextActions.push("Define a domain-specific Region of Practical Equivalence and pass it to lock in the equivalence threshold.");
+    risks.push(
+      'No ROPE configured; the report cannot distinguish "equivalent" from "inconclusive".'
+    );
+    nextActions.push(
+      "Define a domain-specific Region of Practical Equivalence and pass it to lock in the equivalence threshold."
+    );
   }
   const inconclusive = nonComparator.filter((c) => c.decision === "needs_more_data");
   if (inconclusive.length > 0) {
     const worst = inconclusive.reduce((a, b) => b.pairedN < a.pairedN ? b : a);
-    risks.push(`${inconclusive.length} candidate(s) below soft floor (${ctx.minPairs} pairs); thinnest is ${worst.candidateId} with ${worst.pairedN}.`);
-    nextActions.push(`Collect at least ${ctx.minPairs - worst.pairedN} more matched holdout runs for ${worst.candidateId}.`);
+    risks.push(
+      `${inconclusive.length} candidate(s) below soft floor (${ctx.minPairs} pairs); thinnest is ${worst.candidateId} with ${worst.pairedN}.`
+    );
+    nextActions.push(
+      `Collect at least ${ctx.minPairs - worst.pairedN} more matched holdout runs for ${worst.candidateId}.`
+    );
   }
   const rejected = nonComparator.filter((c) => c.decision === "reject");
   if (rejected.length > 0) {
-    risks.push(`${rejected.length} candidate(s) failed the paired test or held-out gate; do not ship those variants.`);
+    risks.push(
+      `${rejected.length} candidate(s) failed the paired test or held-out gate; do not ship those variants.`
+    );
   }
   if (ctx.failureClusters && ctx.failureClusters.clusters.length > 0) {
     const top = ctx.failureClusters.clusters[0];
@@ -954,9 +761,13 @@ function buildRecommendation(candidates, ctx) {
   } else if (decision === "hold") {
     nextActions.push("Keep current production candidate while expanding holdout evidence.");
   } else if (decision === "equivalent") {
-    nextActions.push("Either keep the comparator (no quality regression) or promote on cost/latency grounds \u2014 equivalence does not justify either; the choice is a product decision, not a stats one.");
+    nextActions.push(
+      "Either keep the comparator (no quality regression) or promote on cost/latency grounds \u2014 equivalence does not justify either; the choice is a product decision, not a stats one."
+    );
   } else if (decision === "reject") {
-    nextActions.push("Do not promote this sweep; inspect failures and generate a revised candidate.");
+    nextActions.push(
+      "Do not promote this sweep; inspect failures and generate a revised candidate."
+    );
   }
   return {
     decision,
@@ -969,20 +780,30 @@ function buildRecommendation(candidates, ctx) {
 function buildExecutiveSummary(candidates, recommendation, ctx) {
   const lines = [];
   const nonComparator = candidates.filter((c) => c.candidateId !== ctx.comparator);
-  lines.push(`Evaluated ${nonComparator.length} candidate(s) on the ${ctx.split} split${ctx.comparator ? ` against ${ctx.comparator}` : ""}.`);
-  lines.push(`Recommendation: ${recommendation.decision}${recommendation.candidateId ? ` ${recommendation.candidateId}` : ""}.`);
+  lines.push(
+    `Evaluated ${nonComparator.length} candidate(s) on the ${ctx.split} split${ctx.comparator ? ` against ${ctx.comparator}` : ""}.`
+  );
+  lines.push(
+    `Recommendation: ${recommendation.decision}${recommendation.candidateId ? ` ${recommendation.candidateId}` : ""}.`
+  );
   const promoted = nonComparator.filter((c) => c.decision === "promote").length;
   const held = nonComparator.filter((c) => c.decision === "hold").length;
   const equivalent = nonComparator.filter((c) => c.decision === "equivalent").length;
   const rejected = nonComparator.filter((c) => c.decision === "reject").length;
   const more = nonComparator.filter((c) => c.decision === "needs_more_data").length;
-  lines.push(`Decision mix: ${promoted} promote, ${equivalent} equivalent, ${held} hold, ${rejected} reject, ${more} need more data.`);
+  lines.push(
+    `Decision mix: ${promoted} promote, ${equivalent} equivalent, ${held} hold, ${rejected} reject, ${more} need more data.`
+  );
   const frontier = nonComparator.filter((c) => c.onParetoFrontier).map((c) => c.candidateId);
   if (frontier.length > 0) lines.push(`Pareto-frontier candidates: ${frontier.join(", ")}.`);
   if (ctx.failureClusters) {
-    lines.push(`Failure clustering found ${ctx.failureClusters.totalFailures}/${ctx.failureClusters.totalRuns} failed runs across ${ctx.failureClusters.clusters.length} reportable cluster(s).`);
+    lines.push(
+      `Failure clustering found ${ctx.failureClusters.totalFailures}/${ctx.failureClusters.totalRuns} failed runs across ${ctx.failureClusters.clusters.length} reportable cluster(s).`
+    );
   }
-  lines.push(ctx.preregistrationHash ? `Preregistered analysis: ${ctx.preregistrationHash.slice(0, 12)}\u2026` : "Analysis is post-hoc \u2014 no preregistration hash supplied.");
+  lines.push(
+    ctx.preregistrationHash ? `Preregistered analysis: ${ctx.preregistrationHash.slice(0, 12)}\u2026` : "Analysis is post-hoc \u2014 no preregistration hash supplied."
+  );
   return lines;
 }
 function renderResearchMarkdown(report) {
@@ -994,7 +815,9 @@ function renderResearchMarkdown(report) {
   lines.push(`**Comparator:** ${report.comparator ?? "not configured"}`);
   lines.push(`**ROPE:** ${report.rope ? formatRope(report.rope) : "not configured"}`);
   lines.push(`**Run fingerprint:** \`${report.runFingerprint}\``);
-  lines.push(`**Preregistration:** ${report.preregistrationHash ? `\`${report.preregistrationHash}\`` : "none"}`);
+  lines.push(
+    `**Preregistration:** ${report.preregistrationHash ? `\`${report.preregistrationHash}\`` : "none"}`
+  );
   lines.push("");
   lines.push("## Executive Summary");
   lines.push("");
@@ -1021,7 +844,9 @@ function renderResearchMarkdown(report) {
   lines.push("");
   lines.push("## Candidate Decision Table");
   lines.push("");
-  lines.push("| Candidate | Decision | Mean | \u0394\u0304 | Pr(\u0394>0) | q | d | Paired N | Median Gain CI | MDE | Pareto | Gate |");
+  lines.push(
+    "| Candidate | Decision | Mean | \u0394\u0304 | Pr(\u0394>0) | q | d | Paired N | Median Gain CI | MDE | Pareto | Gate |"
+  );
   lines.push("|---|---|---:|---:|---:|---:|---:|---:|---|---:|---|---|");
   for (const c of report.candidates) {
     const delta = c.meanDeltaVsComparator === null ? "-" : signed(c.meanDeltaVsComparator);
@@ -1030,7 +855,9 @@ function renderResearchMarkdown(report) {
     const d = Number.isFinite(c.cohensD) ? c.cohensD.toFixed(3) : "-";
     const gain = c.gainCi ? `[${fmt(c.gainCi.low)}, ${fmt(c.gainCi.high)}]` : "-";
     const mde = c.mde === null || !Number.isFinite(c.mde) ? "-" : fmt(c.mde);
-    lines.push(`| ${c.candidateId} | ${c.decision} | ${fmt(c.mean)} | ${delta} | ${prGt} | ${q} | ${d} | ${c.pairedN} | ${gain} | ${mde} | ${c.onParetoFrontier ? "yes" : "no"} | ${c.gate ?? "-"} |`);
+    lines.push(
+      `| ${c.candidateId} | ${c.decision} | ${fmt(c.mean)} | ${delta} | ${prGt} | ${q} | ${d} | ${c.pairedN} | ${gain} | ${mde} | ${c.onParetoFrontier ? "yes" : "no"} | ${c.gate ?? "-"} |`
+    );
   }
   lines.push("");
   lines.push("## Statistical Summary");
@@ -1061,7 +888,9 @@ function renderResearchMarkdown(report) {
   lines.push("");
   lines.push("## Chart Specs");
   lines.push("");
-  lines.push("The report carries JSON chart specs for Pareto cost/quality and paired gain histograms.");
+  lines.push(
+    "The report carries JSON chart specs for Pareto cost/quality and paired gain histograms."
+  );
   lines.push("");
   lines.push("```json");
   lines.push(JSON.stringify({ pareto: report.pareto, gains: report.gains }, null, 2));
@@ -1073,7 +902,9 @@ function renderResearchMarkdown(report) {
     lines.push("| Failure Class | Runs | Scenarios | Tool | Example |");
     lines.push("|---|---:|---:|---|---|");
     for (const c of report.failureClusters.clusters.slice(0, 10)) {
-      lines.push(`| ${c.failureClass} | ${c.runCount} | ${c.scenarioIds.length} | ${c.toolName ?? "-"} | ${escapePipes(c.exampleError ?? c.exampleRunId)} |`);
+      lines.push(
+        `| ${c.failureClass} | ${c.runCount} | ${c.scenarioIds.length} | ${c.toolName ?? "-"} | ${escapePipes(c.exampleError ?? c.exampleRunId)} |`
+      );
     }
   }
   return lines.join("\n");
@@ -1161,7 +992,9 @@ function markdownToHtml(markdown) {
   return html.join("\n");
 }
 function renderMarkdownTable(lines) {
-  const rows = lines.filter((line) => !/^\|[-:\s|]+\|$/.test(line)).map((line) => line.slice(1, -1).split("|").map((cell) => inlineMarkdown(cell.trim())));
+  const rows = lines.filter((line) => !/^\|[-:\s|]+\|$/.test(line)).map(
+    (line) => line.slice(1, -1).split("|").map((cell) => inlineMarkdown(cell.trim()))
+  );
   if (rows.length === 0) return "";
   const [head, ...body] = rows;
   const th = head.map((cell) => `<th>${cell}</th>`).join("");
@@ -1202,15 +1035,6 @@ function fmt(x) {
 }
 
 export {
-  normalizeScores,
-  weightedMean,
-  confidenceInterval,
-  interRaterReliability,
-  mannWhitneyU,
-  partialCredit,
-  pairedTTest,
-  wilcoxonSignedRank,
-  cohensD,
   requiredSampleSize,
   bonferroni,
   benjaminiHochberg,
@@ -1223,4 +1047,4 @@ export {
   RESEARCH_REPORT_HARD_PAIR_FLOOR,
   researchReport
 };
-//# sourceMappingURL=chunk-IOXMGMHQ.js.map
+//# sourceMappingURL=chunk-2A5XJB43.js.map