@tangle-network/agent-eval 0.196.0 → 0.198.0

package/dist/{skillopt-optimization-method-IbEhlA1V.js → skillopt-optimization-method-BMP2qV_1.js}

@@ -1,13 +1,14 @@
 import { a as hashCanonical } from "./canonical-BsZVgBhW.js";
-import { a as heldoutSignificance, o as pairHoldout } from "./statistical-heldout-DANVmyug.js";
+import { a as heldoutSignificance, o as pairHoldout } from "./statistical-heldout-BAFDCo07.js";
 import { n as contentHash, t as canonicalJson } from "./verdict-cache-BQEsVy9n.js";
-import { S as developmentClaim, T as surfaceNode, d as combineComparisonCosts, f as costFromLedgerSummary, l as optimizationTokenUsageFromSummary, m as readGepaCandidatePopulationArtifact, p as assertGepaCandidatePopulationSummary, t as llmJudge, ut as runCampaign, w as surfaceDiff, x as SearchRecorder } from "./llm-judge-CxfoYqdk.js";
-import { F as surfaceContentHash, b as campaignBreakdown } from "./campaign-evidence-DJc60F9C.js";
+import { D as surfaceDiff, O as surfaceNode, T as developmentClaim, d as combineComparisonCosts, f as costFromLedgerSummary, l as optimizationTokenUsageFromSummary, m as readGepaCandidatePopulationArtifact, p as assertGepaCandidatePopulationSummary, rt as runCampaign, t as llmJudge, w as SearchRecorder } from "./llm-judge-BYWXsxhZ.js";
+import { F as surfaceContentHash, b as campaignBreakdown } from "./campaign-evidence-jL_U7Ikv.js";
 import { t as deepFreezeCanonicalJson } from "./deep-freeze-CurcOQg3.js";
-import { C as fsCampaignStorage, S as createRunCostLedger, _ as removeCredentialEnvironment, b as safePathComponent, d as assertExternalOptimizerModelBudget, f as assertJsonValue, g as isRecord, h as isExternalTextCandidate, i as startExternalOptimizerModelProxy, m as isCandidateText, n as closeExternalOptimizerResources, p as assertNoCredentialValues, r as runWithCleanup, t as runExternalOptimizerProcess, v as resolveExternalOptimizerCallbackLimits, x as acquireSingleRunLock, y as resolveExternalOptimizerProcessLimits } from "./external-optimizer-subprocess-BrghTvEp.js";
-import { n as openSearchLedger } from "./search-ledger-9W0g5FP6.js";
+import { S as fsCampaignStorage, _ as removeCredentialEnvironment, b as safePathComponent, d as assertExternalOptimizerModelBudget, f as assertJsonValue, g as isRecord, h as isExternalTextCandidate, i as startExternalOptimizerModelProxy, m as isCandidateText, n as closeExternalOptimizerResources, p as assertNoCredentialValues, r as runWithCleanup, t as runExternalOptimizerProcess, v as resolveExternalOptimizerCallbackLimits, x as createRunCostLedger, y as resolveExternalOptimizerProcessLimits } from "./external-optimizer-subprocess-laM1-UDV.js";
+import { n as openSearchLedger } from "./search-ledger-BnUytR8X.js";
+import { t as acquireSingleRunLock } from "./single-run-lock-fVu3MZWH.js";
 import { g as assertServedModelPolicy } from "./llm-client-c_bKHTXz.js";
-import { t as startExternalOptimizerCallback } from "./external-optimizer-process-_YIECYJ7.js";
+import { t as startExternalOptimizerCallback } from "./external-optimizer-process-CEXcCKqd.js";
 import { createHash, randomBytes } from "node:crypto";
 import { z } from "zod";
 import { mkdir } from "node:fs/promises";
@@ -1609,12 +1610,22 @@ function externalSurface(candidate) {
 * GEPA's choice as the selected node, and the close. GEPA's own selection is a
 * budget decision, so the ledger carries no claim; the final comparison runs
 * outside this search.
+*
+* A composed recipe reports no population: pass `population: null` and GEPA's
+* returned best as `selected`, and every evaluated candidate enters with an
+* `unknown` edge, as SkillOpt's do.
 */
 async function recordGepaSearch(input) {
-	const best = input.population.candidates.find((candidate) => candidate.index === input.population.bestIndex);
-	if (!best) throw new Error(`GEPA candidate population has no bestIndex entry ${input.population.bestIndex}`);
+	if (input.population === null) return recordExternalSearch({
+		...input,
+		optimizer: "GEPA"
+	});
+	const { population } = input;
+	const best = population.candidates.find((candidate) => candidate.index === population.bestIndex);
+	if (!best) throw new Error(`GEPA candidate population has no bestIndex entry ${population.bestIndex}`);
 	return recordExternalSearch({
 		...input,
+		population,
 		optimizer: "GEPA",
 		selected: externalSurface(best.candidate)
 	});
@@ -2050,37 +2061,39 @@ function gepaOptimizationMethod(config) {
 				resumed: result.resumed
 			});
 			if (callback.evaluations() !== result.totalEvaluations) throw new Error(`${name}: GEPA reported ${result.totalEvaluations} evaluations but the callback received ${callback.evaluations()}`);
-			let searchHistory;
-			if (result.candidatePopulation) {
-				const population = readGepaCandidatePopulationArtifact({
-					summary: result.candidatePopulation,
-					storage
-				});
+			const population = result.candidatePopulation ? readGepaCandidatePopulationArtifact({
+				summary: result.candidatePopulation,
+				storage
+			}) : null;
+			if (population) {
 				const selected = population.candidates[population.bestIndex];
 				const selectedHash = contentHash({
 					kind: "external-text-candidate",
 					candidate: result.bestCandidate
 				});
 				if (selected?.candidateHash !== selectedHash) throw new Error(`${name}: GEPA candidate population identifies a different winner`);
-				if (config.searchLedger) searchHistory = await recordGepaSearch({
-					name,
-					path: config.searchLedger.path ?? `${runDir}/search-ledger.jsonl`,
-					searchId: runId,
-					identity: config.searchLedger.identity,
-					storage,
-					seed: input.seed,
-					baselineSurface: input.baselineSurface,
-					trainScenarios: input.trainScenarios,
-					selectionScenarios: input.selectionScenarios,
-					evaluationLimit,
-					population,
-					observations: readExternalOptimizerObservationArtifact({
-						summary: observationLog.summary(),
-						storage
-					}),
-					generationAccounting: optimizerAccounting(result.tokenUsage, result.proposerCostUsd)
-				});
 			}
+			const searchHistory = config.searchLedger ? await recordGepaSearch({
+				name,
+				path: config.searchLedger.path ?? `${runDir}/search-ledger.jsonl`,
+				searchId: runId,
+				identity: config.searchLedger.identity,
+				storage,
+				seed: input.seed,
+				baselineSurface: input.baselineSurface,
+				trainScenarios: input.trainScenarios,
+				selectionScenarios: input.selectionScenarios,
+				evaluationLimit,
+				...population ? { population } : {
+					population: null,
+					selected: decodeExternalTextCandidate(result.bestCandidate)
+				},
+				observations: readExternalOptimizerObservationArtifact({
+					summary: observationLog.summary(),
+					storage
+				}),
+				generationAccounting: optimizerAccounting(result.tokenUsage, result.proposerCostUsd)
+			}) : void 0;
 			const evaluationCost = costFromLedgerSummary(costLedger.summary({
 				phase: "gepa.external-evaluation",
 				tags: runBudget.runTags
@@ -2570,4 +2583,4 @@ function skillOptOptimizationMethod(config) {
 //#endregion
 export { importGepaPopulation as a, heldOutGate as c, decodeExternalTextCandidate as d, readExternalOptimizerObservationArtifact as f, runReferenceEquivalenceJudge as g, createReferenceEquivalenceJudge as h, importExternalEvaluations as i, composeGate as l, REFERENCE_EQUIVALENCE_JUDGE_VERSION as m, gepaOptimizationMethod as n, recordGepaSearch as o, REFERENCE_EQUIVALENCE_INPUT_LIMITS as p, externalSurface as r, recordSkillOptSearch as s, skillOptOptimizationMethod as t, externalTextOptimizationMethod as u };
 
-//# sourceMappingURL=skillopt-optimization-method-IbEhlA1V.js.map
+//# sourceMappingURL=skillopt-optimization-method-BMP2qV_1.js.map