@tangle-network/agent-eval 0.114.0 → 0.115.0

package/CHANGELOG.md

@@ -4,6 +4,18 @@ All notable changes to `@tangle-network/agent-eval` and its sibling `agent-eval-
 
 ---
 
+## [0.115.0] — 2026-07-11 — auditable cross-surface improvement search
+
+### Added
+
+- `openSearchLedger()` records the predeclared candidate, task, and operation denominators for an improvement search in a durable hash-chained event stream.
+  Failed proposal slots, partial batches, task attempts, measured cost, surface firing and effect, and terminal selection decisions remain replayable after a crash.
+- `analyzeCrossSurfaceInteractions()` compares fixed, best-single, blind-union, and interaction-aware agent changes on the same task rows.
+  It preserves missing and invalid attempts, cost, firing, effect, synergy, interference, and every evaluated composition path, including combinations whose constituents are neutral alone.
+
+This release is additive.
+Existing consumers do not need to update unless they want durable improvement-search accounting or cross-surface composition.
+
 ## [0.114.0] — 2026-07-11 — exact directional paired inference
 
 ### Added

package/README.md

@@ -127,7 +127,7 @@ Each example: `README.md` + a single `index.ts` runnable via `pnpm tsx`. Prints
 | `…/adapters/otel` | `createOtelBridge` — forwards OpenTelemetry-shape spans into the hosted-tier ingest, no `@opentelemetry/*` dependency |
 | `…/adapters/langchain` | Wrap any LangChain `Runnable` as a `Dispatch` (or `JudgeConfig`), no `@langchain/core` peer dep |
 | `…/adapters/http` | `httpDispatch` + `runDispatchServer` — run a campaign's worker on another machine (multi-region, remote worker execution) |
-| `…/campaign` | **The measurement + improvement engine**: `runProfileMatrix`, `compareProposers`, every surface proposer (`gepaProposer`, `fapoProposer`, `parameterSweepProposer`, `haloProposer`, `skillOptProposer`, `aceProposer`, `memoryCurationProposer`, …), the gates, storage backends, and loop provenance. `/contract` re-exports the app-facing subset. |
+| `…/campaign` | **The measurement + improvement engine**: `runProfileMatrix`, `compareProposers`, `analyzeCrossSurfaceInteractions`, every surface proposer (`gepaProposer`, `fapoProposer`, `parameterSweepProposer`, `haloProposer`, `skillOptProposer`, `aceProposer`, `memoryCurationProposer`, …), the gates, storage backends, and loop provenance. `/contract` re-exports the app-facing subset. |
 | `…/rl` | Bridge from eval artifacts to training signal: verifiable rewards, preferences, OPE, tournaments, contamination, compute curves, trainer-format exporters, process rewards, plus the durable corpus + `buildRlDataset` / datasheet bundle |
 | `…/reporting` | Release-decision statistics: `pairedBootstrap`, `benjaminiHochberg`, anytime-valid sequential e-values, `evaluateReleaseConfidence`, and the report renderers |
 | `…/analyst` | The trace-analyst surface: `AnalystRegistry` + `buildDefaultAnalystRegistry` (run the failure-clustering panel), `FindingsStore`, and the LLM chat transports |

package/dist/analyst/index.js

@@ -11,7 +11,7 @@ import {
   diffFindings,
   emitSkillUsageFindings,
   runSemanticConceptJudge
-} from "../chunk-4FRTXH2M.js";
+} from "../chunk-I2HNIE6N.js";
 import {
   behavioralAnalyst,
   buildDefaultAnalystRegistry,
@@ -33,7 +33,7 @@ import {
   policyEditsFromFindings,
   scorePolicyEditReadiness,
   validatePolicyEdit
-} from "../chunk-HJWNHCD5.js";
+} from "../chunk-AN5UYSVD.js";
 import {
   ANALYST_SEVERITIES,
   AnalystRegistry,

package/dist/benchmarks/index.js

@@ -17,11 +17,11 @@ import {
   runBenchmarkAdapter,
   summarizeBenchmarkCampaign
 } from "../chunk-3LXTCTWL.js";
-import "../chunk-VHPD6AXX.js";
+import "../chunk-MNR6ZW4P.js";
 import "../chunk-N6MTC3GK.js";
 import "../chunk-VI2UW6B6.js";
 import "../chunk-FAOEFFRT.js";
-import "../chunk-HJWNHCD5.js";
+import "../chunk-AN5UYSVD.js";
 import "../chunk-DWLIGZBX.js";
 import "../chunk-ARU2PZFM.js";
 import "../chunk-PJQFMIOX.js";