@tangle-network/agent-eval 0.119.0 → 0.120.0

package/dist/{chunk-XJ7JVCHB.js → chunk-3A246TSA.js}

@@ -1,6 +1,6 @@
 import {
   summarizeBackendIntegrity
-} from "./chunk-PAHNGS65.js";
+} from "./chunk-XDIRG3TO.js";
 import {
   paretoChart
 } from "./chunk-DPZAEKA6.js";
@@ -995,4 +995,4 @@ export {
   summarizeExecution,
   analyzeRuns
 };
-//# sourceMappingURL=chunk-XJ7JVCHB.js.map
+//# sourceMappingURL=chunk-3A246TSA.js.map

package/dist/{chunk-LMJ2TGWJ.js → chunk-JM2SKQMS.js}

@@ -5,7 +5,7 @@ import {
   aggregateRunScore,
   clamp01,
   computeFindingId
-} from "./chunk-7A4LIMMY.js";
+} from "./chunk-S5TT5R3L.js";
 import {
   callLlmJson,
   costReceiptFromLlm,
@@ -747,4 +747,4 @@ export {
   runSemanticConceptJudge,
   createSemanticConceptJudge
 };
-//# sourceMappingURL=chunk-LMJ2TGWJ.js.map
+//# sourceMappingURL=chunk-JM2SKQMS.js.map

package/dist/{chunk-F6YUH3L4.js → chunk-JN2FCO5W.js}

@@ -3,6 +3,7 @@ import {
   assertCodeSurfaceIdentity,
   campaignBreakdown,
   campaignMeanComposite,
+  comparePairedArms,
   costReceiptFromTCloud,
   defaultProductionGate,
   gepaProposer,
@@ -15,7 +16,7 @@ import {
   runImprovementLoop,
   surfaceContentHash,
   surfaceHash
-} from "./chunk-RLCJQ6VZ.js";
+} from "./chunk-32BZXMSO.js";
 import {
   SearchLedgerConflictError,
   SearchLedgerError,
@@ -30,7 +31,7 @@ import {
   runCampaign,
   summarizeBackendIntegrity,
   withSearchLedgerFileLock
-} from "./chunk-PAHNGS65.js";
+} from "./chunk-XDIRG3TO.js";
 import {
   Mutex
 } from "./chunk-3YYRZDON.js";
@@ -42,13 +43,14 @@ import {
   admitPolicyEdit,
   applyPolicyEditToSurface,
   assertNoJudgeVerdict,
+  createAnalystAi,
   createTraceAnalystKind,
   isPolicyEdit,
   makePolicyEdit,
   makePolicyEditCandidateRecord,
   policyEditsFromFindings,
   validatePolicyEditCandidateRecord
-} from "./chunk-7A4LIMMY.js";
+} from "./chunk-S5TT5R3L.js";
 import {
   callLlm,
   callLlmJson,
@@ -58,11 +60,8 @@ import {
 } from "./chunk-NJC7U437.js";
 import {
   eProcess,
-  mcnemar,
   mulberry32,
-  pairedBootstrap,
-  pairedRiskDifference,
-  wilcoxonSignedRank
+  pairedBootstrap
 } from "./chunk-PJQFMIOX.js";
 import {
   CostAccountingIncompleteError,
@@ -70,10 +69,10 @@ import {
 } from "./chunk-JHOJHHU7.js";
 import {
   analyzeTraces
-} from "./chunk-FIUFRXP3.js";
+} from "./chunk-WW2A73HW.js";
 import {
   OtlpFileTraceStore
-} from "./chunk-6QIM2EAP.js";
+} from "./chunk-PXD6ZFNY.js";
 import {
   modelHasSnapshot,
   validateRunRecord
@@ -569,155 +568,6 @@ function failureModeRecallJudge(opts = {}) {
   };
 }
 
-// src/paired-arms.ts
-function pairArms(rows, opts) {
-  const { baselineArm, treatmentArm } = opts;
-  if (baselineArm === treatmentArm) {
-    throw new ValidationError(
-      `pairArms: baselineArm and treatmentArm are both '${baselineArm}' \u2014 an arm cannot be compared to itself`
-    );
-  }
-  const byArm = /* @__PURE__ */ new Map();
-  const armsSeen = /* @__PURE__ */ new Set();
-  for (const row of rows) {
-    armsSeen.add(row.arm);
-    if (row.arm !== baselineArm && row.arm !== treatmentArm) continue;
-    const byKey = byArm.get(row.arm) ?? /* @__PURE__ */ new Map();
-    const group = byKey.get(row.pairKey) ?? [];
-    group.push(row);
-    byKey.set(row.pairKey, group);
-    byArm.set(row.arm, byKey);
-  }
-  for (const arm of [baselineArm, treatmentArm]) {
-    if (!byArm.has(arm)) {
-      const seen = [...armsSeen].sort().join(", ") || "<none>";
-      throw new ValidationError(`pairArms: no rows for arm '${arm}' (arms present: ${seen})`);
-    }
-  }
-  const baselineByKey = byArm.get(baselineArm);
-  const treatmentByKey = byArm.get(treatmentArm);
-  const allKeys = [.../* @__PURE__ */ new Set([...baselineByKey.keys(), ...treatmentByKey.keys()])].sort();
-  const pairs = [];
-  const unpairedBaseline = [];
-  const unpairedTreatment = [];
-  for (const pairKey of allKeys) {
-    const b = baselineByKey.get(pairKey) ?? [];
-    const t = treatmentByKey.get(pairKey) ?? [];
-    if (b.length <= 1 && t.length <= 1) {
-      if (b.length === 1 && t.length === 1) {
-        pairs.push({ pairKey, repIndex: 0, baseline: b[0], treatment: t[0] });
-      } else {
-        unpairedBaseline.push(...b);
-        unpairedTreatment.push(...t);
-      }
-      continue;
-    }
-    const bByRep = indexByRepKey(b, pairKey, baselineArm);
-    const tByRep = indexByRepKey(t, pairKey, treatmentArm);
-    const repKeys = [.../* @__PURE__ */ new Set([...bByRep.keys(), ...tByRep.keys()])].sort();
-    let repIndex = 0;
-    for (const repKey of repKeys) {
-      const baseline = bByRep.get(repKey);
-      const treatment = tByRep.get(repKey);
-      if (baseline !== void 0 && treatment !== void 0) {
-        pairs.push({ pairKey, repIndex: repIndex++, baseline, treatment });
-      } else if (baseline !== void 0) {
-        unpairedBaseline.push(baseline);
-      } else if (treatment !== void 0) {
-        unpairedTreatment.push(treatment);
-      }
-    }
-  }
-  return { pairs, unpairedBaseline, unpairedTreatment };
-}
-function indexByRepKey(group, pairKey, arm) {
-  const byRep = /* @__PURE__ */ new Map();
-  for (const row of group) {
-    if (row.repKey === void 0) {
-      throw new ValidationError(
-        `pairArms: pairKey '${pairKey}' has multiple reps in an arm, but a row in arm '${arm}' is missing repKey \u2014 multi-rep items require an explicit repKey on every row so reps pair by identity (pairing reps by outcome or by index would bias the paired statistics)`
-      );
-    }
-    if (byRep.has(row.repKey)) {
-      throw new ValidationError(
-        `pairArms: duplicate repKey '${row.repKey}' for pairKey '${pairKey}' in arm '${arm}' \u2014 (pairKey, repKey) must uniquely identify a rep within an arm`
-      );
-    }
-    byRep.set(row.repKey, row);
-  }
-  return byRep;
-}
-function comparePairedArms(rows, opts) {
-  const { pairs, unpairedBaseline, unpairedTreatment } = pairArms(rows, opts);
-  let correctness = null;
-  const baselinePass = [];
-  const treatmentPass = [];
-  for (const pair of pairs) {
-    if (pair.baseline.pass === void 0 || pair.treatment.pass === void 0) continue;
-    baselinePass.push(pair.baseline.pass ? 1 : 0);
-    treatmentPass.push(pair.treatment.pass ? 1 : 0);
-  }
-  if (baselinePass.length > 0) {
-    const mc = mcnemar(baselinePass, treatmentPass);
-    correctness = {
-      b10: mc.b,
-      b01: mc.c,
-      mcnemar: mc,
-      riskDifference: pairedRiskDifference(baselinePass, treatmentPass)
-    };
-  }
-  const metricNames = opts.metricNames ?? [
-    ...new Set(
-      pairs.flatMap((p) => [
-        ...Object.keys(p.baseline.metrics ?? {}),
-        ...Object.keys(p.treatment.metrics ?? {})
-      ])
-    )
-  ].sort();
-  const metricDeltas = metricNames.map((name) => {
-    const before = [];
-    const after = [];
-    let nMissing = 0;
-    for (const pair of pairs) {
-      const b = metricValue(pair.baseline, name);
-      const t = metricValue(pair.treatment, name);
-      if (b === void 0 || t === void 0) {
-        nMissing++;
-        continue;
-      }
-      before.push(b);
-      after.push(t);
-    }
-    const bootstrapCi = before.length === 0 ? null : pairedBootstrap(before, after, opts.bootstrap);
-    return {
-      name,
-      n: before.length,
-      nMissing,
-      medianDelta: bootstrapCi === null ? Number.NaN : bootstrapCi.median,
-      meanDelta: bootstrapCi === null ? Number.NaN : bootstrapCi.mean,
-      bootstrapCi,
-      wilcoxon: before.length === 0 ? null : wilcoxonSignedRank(before, after)
-    };
-  });
-  return {
-    nPairs: pairs.length,
-    nUnpairedBaseline: unpairedBaseline.length,
-    nUnpairedTreatment: unpairedTreatment.length,
-    correctness,
-    metricDeltas
-  };
-}
-function metricValue(row, name) {
-  const v = row.metrics?.[name];
-  if (v === void 0) return void 0;
-  if (!Number.isFinite(v)) {
-    throw new ValidationError(
-      `comparePairedArms: non-finite value for metric '${name}' on pairKey '${row.pairKey}' (arm '${row.arm}'): ${v}`
-    );
-  }
-  return v;
-}
-
 // src/campaign/cross-surface-context.ts
 function validateCrossSurfaceInput(input) {
   assertNonEmptyUnique(input.taskOrder, "taskOrder");
@@ -2204,8 +2054,7 @@ function sequentialPairedGate(options) {
         direction: cfg.direction,
         minEffect: cfg.minEffect,
         pairedN: deltas.length,
-        preRegisteredId: manifest?.id,
-        metric: manifest?.metric
+        ...manifest ? { preRegisteredId: manifest.id, metric: manifest.metric } : {}
       };
       const meanDelta = deltas.length === 0 ? void 0 : deltas.reduce((s, d) => s + d, 0) / deltas.length;
       if (last === void 0) {
@@ -6294,7 +6143,6 @@ ${block}
 }
 
 // src/campaign/proposers/trace-analyst.ts
-import { ai } from "@ax-llm/ax";
 function renderFindings(findings) {
   return findings.map((f, i) => {
     const action = f.recommended_action ? `
@@ -6308,11 +6156,11 @@ function traceAnalystProposer(opts) {
   if (!opts.model) throw new Error("traceAnalystProposer: model is required");
   const kinds = opts.kinds ?? DEFAULT_TRACE_ANALYST_KINDS;
   const produceFindings = opts.analyze ?? (async (path, c) => {
-    const aiService = ai({
-      name: opts.provider ?? "openai",
+    const aiService = createAnalystAi({
+      provider: opts.provider ?? "openai",
       apiKey: opts.apiKey,
-      apiURL: opts.baseUrl,
-      config: { model: opts.model }
+      baseUrl: opts.baseUrl,
+      model: opts.model
     });
     const registry = new AnalystRegistry();
     for (const spec of kinds) {
@@ -8026,8 +7874,6 @@ function resolveWorktreePath(surface, worktreeDir) {
 }
 
 export {
-  pairArms,
-  comparePairedArms,
   completionVerdict,
   verifyCompletion,
   parseCorrectnessResponse,
@@ -8109,4 +7955,4 @@ export {
   verifyCodeSurface,
   resolveWorktreePath
 };
-//# sourceMappingURL=chunk-F6YUH3L4.js.map
+//# sourceMappingURL=chunk-JN2FCO5W.js.map