@tangle-network/agent-eval 0.126.0 → 0.126.1

package/CHANGELOG.md

@@ -6,6 +6,20 @@ All notable changes to `@tangle-network/agent-eval` and its sibling `agent-eval-
 
 ## [Unreleased]
 
+## [0.126.1] - 2026-07-24 - optimizer lifecycle integrity
+
+### Changed
+
+- `compareOptimizationMethods()` uses one caller-visible spend limit across optimizer models, train and selection evaluation, and final scoring.
+- Candidate surfaces are detached before every optimizer and scoring callback.
+- SkillOpt receives complete non-secret model settings automatically while provider credentials remain in the local Node proxy.
+
+### Fixed
+
+- Aborting a comparison now stops official optimizer subprocess groups, callback work, and model proxy requests.
+- Resumed final scoring retains the original cost identity and reports cumulative spend instead of resetting cached calls to zero.
+- Model-backed GEPA fails before search when the full upstream dependency set is missing instead of silently returning the baseline after swallowed reflection errors.
+
 ## [0.126.0] - 2026-07-24 - official optimizer engines
 
 ### Added

package/dist/analyst/index.js

@@ -9,7 +9,7 @@ import {
   diffFindings,
   emitSkillUsageFindings,
   runSemanticConceptJudge
-} from "../chunk-4B7ZZHPX.js";
+} from "../chunk-KE2VWPZX.js";
 import {
   ANALYST_SEVERITIES,
   AnalystRegistry,
@@ -50,9 +50,9 @@ import {
   stripCodeFences,
   structureFindings,
   validateUsageSettlementTimeout
-} from "../chunk-CM4OILD2.js";
+} from "../chunk-LUNF2SEL.js";
 import "../chunk-HHWE3POT.js";
-import "../chunk-UI4YMIN2.js";
+import "../chunk-WGXIEX7P.js";
 import "../chunk-J5SQWP6Y.js";
 import {
   CostLedger

package/dist/benchmarks/index.js

@@ -17,10 +17,10 @@ import {
   runBenchmarkAdapter,
   summarizeBenchmarkCampaign
 } from "../chunk-W4L6C2XT.js";
-import "../chunk-KO2PZOGP.js";
-import "../chunk-NTOV7RU5.js";
+import "../chunk-NGUYT5CI.js";
+import "../chunk-VMUENW6F.js";
 import "../chunk-UCLVDLCH.js";
-import "../chunk-UI4YMIN2.js";
+import "../chunk-WGXIEX7P.js";
 import "../chunk-ARU2PZFM.js";
 import "../chunk-J5SQWP6Y.js";
 import "../chunk-PJQFMIOX.js";

package/dist/campaign/index.d.ts

@@ -2479,8 +2479,8 @@ declare function planCampaignRun<TScenario extends Scenario, TArtifact>(opts: Pl
  * data and compared with paired confidence intervals.
  */
 
-/** Per-method campaign settings. Each method receives its own spend account. */
-type OptimizationMethodRunOptions<TScenario extends Scenario, TArtifact> = Omit<RunCampaignOptions<TScenario, TArtifact>, 'costLedger' | 'dispatch' | 'judges' | 'runDir' | 'scenarios' | 'seed'>;
+/** Shared campaign settings applied to every optimization method. */
+type OptimizationMethodRunOptions<TScenario extends Scenario, TArtifact> = Omit<RunCampaignOptions<TScenario, TArtifact>, 'costCeiling' | 'costLedger' | 'dispatch' | 'judges' | 'runDir' | 'scenarios' | 'seed'>;
 /** Cost reported by a method or by final test scoring. */
 interface ComparisonCost {
     totalCostUsd: number;
@@ -2553,7 +2553,7 @@ interface OptimizationMethodInput<TScenario extends Scenario, TArtifact> {
     readonly seed: number;
     /** Shared defaults for every method. A method may override them explicitly. */
     readonly runOptions: Readonly<OptimizationMethodRunOptions<TScenario, TArtifact>>;
-    /** Durable spend account shared by the method's model and evaluation calls. */
+    /** Durable spend account shared by every method and final scoring. */
     readonly costLedger: CostLedgerHandle;
 }
 interface OptimizationMethodResult {
@@ -2662,8 +2662,7 @@ interface CompareOptimizationMethodsOptions<TScenario extends Scenario, TArtifac
     /** Simultaneous confidence across method-vs-baseline and method-vs-method contrasts.
      *  Each bootstrap interval is Bonferroni-adjusted. Default 0.95. */
     confidence?: number;
-    /** Shared spend limit across baseline and winner scoring on the final test partition.
-     *  Each method owns its optimization budget through `optimizationRunOptions.costCeiling`. */
+    /** Shared spend limit across every method's optimizer and evaluation calls plus final scoring. */
     costCeiling?: number;
 }
 /**

package/dist/campaign/index.js

@@ -32,7 +32,7 @@ import {
   userStoryScoreboard,
   validateSearchLedgerEvent,
   verifyCodeSurface
-} from "../chunk-KO2PZOGP.js";
+} from "../chunk-NGUYT5CI.js";
 import {
   acquireSingleRunLock,
   assertCodeSurfaceIdentity,
@@ -78,7 +78,7 @@ import {
   surfaceContentHash,
   surfaceHash,
   verifyLoopProvenanceRecord
-} from "../chunk-NTOV7RU5.js";
+} from "../chunk-VMUENW6F.js";
 import {
   SearchLedgerConflictError,
   SearchLedgerError,
@@ -96,7 +96,7 @@ import {
   runCampaign,
   tangleTracesRoot
 } from "../chunk-UCLVDLCH.js";
-import "../chunk-UI4YMIN2.js";
+import "../chunk-WGXIEX7P.js";
 import "../chunk-ARU2PZFM.js";
 import "../chunk-J5SQWP6Y.js";
 import "../chunk-PJQFMIOX.js";

package/dist/{chunk-4B7ZZHPX.js → chunk-KE2VWPZX.js}

@@ -1,11 +1,11 @@
 import {
   computeFindingId
-} from "./chunk-CM4OILD2.js";
+} from "./chunk-LUNF2SEL.js";
 import {
   Mutex,
   aggregateRunScore,
   clamp01
-} from "./chunk-UI4YMIN2.js";
+} from "./chunk-WGXIEX7P.js";
 import {
   callLlmJson,
   costReceiptFromLlm,
@@ -749,4 +749,4 @@ export {
   runSemanticConceptJudge,
   createSemanticConceptJudge
 };
-//# sourceMappingURL=chunk-4B7ZZHPX.js.map
+//# sourceMappingURL=chunk-KE2VWPZX.js.map

package/dist/{chunk-CM4OILD2.js → chunk-LUNF2SEL.js}

@@ -1,6 +1,9 @@
 import {
   executionTrackByLane
 } from "./chunk-HHWE3POT.js";
+import {
+  combineAbortSignals
+} from "./chunk-WGXIEX7P.js";
 import {
   LlmClient,
   callLlm,
@@ -2108,11 +2111,6 @@ function validateTimeout(timeoutMs) {
   }
   return timeoutMs;
 }
-function combineAbortSignals(caller, timeout) {
-  if (!caller) return timeout;
-  if (!timeout) return caller;
-  return AbortSignal.any([caller, timeout]);
-}
 async function waitForOperation(operation, signal, abortGraceMs) {
   if (!signal) return operation;
   if (signal.aborted) {
@@ -2427,4 +2425,4 @@ export {
   AnalystRegistry,
   buildDefaultAnalystRegistry
 };
-//# sourceMappingURL=chunk-CM4OILD2.js.map
+//# sourceMappingURL=chunk-LUNF2SEL.js.map