@tangle-network/agent-eval 0.122.8 → 0.123.0

package/dist/index.d.ts

@@ -14252,6 +14252,10 @@ interface RunImprovementLoopResult<TArtifact, TScenario extends Scenario> extend
     neutralizedOnHoldout?: CampaignResult<TArtifact, TScenario>;
     neutralizedSurface?: MutableSurface;
     gateResult: Awaited<ReturnType<Gate<TArtifact, TScenario>['decide']>>;
+    /** Present iff the loop ran with `holdout: 'deferred'`. When set,
+     *  `baselineOnHoldout`/`winnerOnHoldout` are the shared EMPTY campaign (zero
+     *  cells dispatched) and the gate verdict is the forced `'hold'`. */
+    holdout?: 'deferred';
     /** Unified baseline→winner surface diff. Computed UNCONDITIONALLY (not only
      *  when `autoOnPromote === 'pr'`) so the diff that the gate decided on is
      *  always present on the result + in the emitted provenance record. Empty

package/dist/index.js

@@ -53,7 +53,7 @@ import {
   pairArms,
   parseCorrectnessResponse,
   verifyCompletion
-} from "./chunk-ZDAKRZG5.js";
+} from "./chunk-GS3FJGUF.js";
 import {
   DEFAULT_MUTATION_PRIMITIVES,
   DEFAULT_RED_TEAM_CORPUS,
@@ -91,7 +91,7 @@ import {
   scoreRedTeamOutput,
   surfaceContentHash,
   toolNamesForRun
-} from "./chunk-S6IL3R5I.js";
+} from "./chunk-HZJF4IUO.js";
 import {
   BackendIntegrityError,
   assertRealAgentReceipts,
@@ -116,12 +116,12 @@ import {
   defaultIsMaterial,
   diffFindings,
   runSemanticConceptJudge
-} from "./chunk-MI2H23RV.js";
+} from "./chunk-G2GPNLSX.js";
 import {
   buildDefaultAnalystRegistry,
   computeTraceMetrics,
   createChatClient
-} from "./chunk-BVMWQDC5.js";
+} from "./chunk-VJ7T5WIO.js";
 import "./chunk-HHWE3POT.js";
 import {
   AnalystRegistry,
@@ -156,7 +156,7 @@ import {
   scorePolicyEditReadiness,
   validatePolicyEdit,
   validatePolicyEditCandidateRecord
-} from "./chunk-MUS5DOH7.js";
+} from "./chunk-FC5NDO3E.js";
 import {
   allCriticalPassed,
   controlFailureClassFromVerification,
@@ -305,7 +305,7 @@ import {
   scoreTraceInsightReadiness,
   tokenizeDomainWords,
   traceAnalystOnRunComplete
-} from "./chunk-ULOKLHIQ.js";
+} from "./chunk-A5S77LSE.js";
 import "./chunk-H5UD2323.js";
 import {
   extractUsage,
@@ -327,7 +327,7 @@ import {
   TRACE_ANALYST_ACTOR_DESCRIPTION,
   TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION,
   analyzeTraces
-} from "./chunk-WW2A73HW.js";
+} from "./chunk-NUKSVU3W.js";
 import {
   DEFAULT_REDACTION_RULES,
   REDACTION_VERSION,
@@ -352,7 +352,7 @@ import {
   stringField,
   traceAnalystFunctionGroup,
   traceSpanKindToOpenInferenceKind
-} from "./chunk-PXD6ZFNY.js";
+} from "./chunk-IR3KBHOY.js";
 import {
   RunIntegrityError,
   assertRunCaptured,

package/dist/openapi.json

@@ -2,7 +2,7 @@
   "openapi": "3.1.0",
   "info": {
     "title": "@tangle-network/agent-eval — wire protocol",
-    "version": "0.122.8",
+    "version": "0.123.0",
     "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/dist/traces.js

@@ -27,7 +27,7 @@ import {
   scoreTraceInsightReadiness,
   tokenizeDomainWords,
   traceAnalystOnRunComplete
-} from "./chunk-ULOKLHIQ.js";
+} from "./chunk-A5S77LSE.js";
 import "./chunk-H5UD2323.js";
 import {
   extractUsage,
@@ -49,7 +49,7 @@ import {
   TRACE_ANALYST_ACTOR_DESCRIPTION,
   TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION,
   analyzeTraces
-} from "./chunk-WW2A73HW.js";
+} from "./chunk-NUKSVU3W.js";
 import {
   DEFAULT_REDACTION_RULES,
   REDACTION_VERSION,
@@ -74,7 +74,7 @@ import {
   stringField,
   traceAnalystFunctionGroup,
   traceSpanKindToOpenInferenceKind
-} from "./chunk-PXD6ZFNY.js";
+} from "./chunk-IR3KBHOY.js";
 import {
   RunIntegrityError,
   assertRunCaptured,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tangle-network/agent-eval",
-  "version": "0.122.8",
+  "version": "0.123.0",
   "description": "Evaluate and improve AI agents from runs, traces, judges, and feedback. Compare candidates, cluster failures, measure lift, and gate releases.",
   "homepage": "https://github.com/tangle-network/agent-eval#readme",
   "repository": {