chore(release): agent-eval 0.143.0

CHANGELOG.md

@@ -4,6 +4,24 @@ All notable changes to `@tangle-network/agent-eval` and its sibling `agent-eval-
 
 ---
 
+## [0.143.0] - 2026-08-03 - exact campaign cost provenance
+
+### Changed
+
+- **Breaking:** `CampaignCellResult.costEstimated` is replaced by the required `costProvenance` union, and `ProfileSummary.totalCostUsd` becomes `number | null` with an adjacent `costProvenance` value.
+- Campaign measurement digests now include the complete cost-provenance value, so digest equality is intentionally not preserved across the `0.142.x` to `0.143.0` data boundary.
+- Run-record raw cost fields now describe the total as observed, estimated, or uncaptured, while `cost_known_subtotal_usd` retains any measured subtotal when the total remains unknown.
+- `planCampaignRun` reports `cellsBlocked`, and each planned cell may be `blocked` when saved data cannot prove its cost or receipt identity.
+- Caches created before `0.143.0` have no complete cost provenance and therefore stop a resumable campaign by default.
+  Inspect the plan and set `rerunInvalidCachedCells: true` for a one-time selective rerun, or set `resumable: false` for an intentional full rerun.
+
+### Fixed
+
+- Failed provider calls retain input, cached-input, cache-write, output, and reasoning tokens without presenting an unknown USD total as zero.
+- Campaign execution inspects every saved cell and its exact billing receipts before any concurrent dispatch begins, so a late stale cell cannot waste spend in earlier cells.
+- Malformed, unreadable, mismatched, or receipt-incomplete saved cells require explicit rerun approval instead of being silently reused or re-executed.
+- `planCampaignRun` no longer creates the run directory while performing read-only inspection.
+
 ## [0.142.2] - 2026-08-02 - current shared type cohort
 
 ### Changed

clients/python/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "agent-eval-rpc"
-version = "0.142.2"
+version = "0.143.0"
 description = "Python RPC client, official optimizer bridge, and DSPy metric adapter for @tangle-network/agent-eval."
 readme = "README.md"
 requires-python = ">=3.10"

clients/python/src/agent_eval_rpc/__init__.py

@@ -53,7 +53,7 @@
 try:
     __version__ = version("agent-eval-rpc")
 except PackageNotFoundError:
-    __version__ = "0.142.2"
+    __version__ = "0.143.0"
 
 __all__ = [
     "Client",

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tangle-network/agent-eval",
-  "version": "0.142.2",
+  "version": "0.143.0",
   "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": {

src/analyst/benchmark-implementation.ts

@@ -10,7 +10,7 @@ export const ANALYST_BENCHMARK_DEPENDENCY_LOCK_FILES = Object.freeze([
 ])
 
 export const ANALYST_BENCHMARK_DEPENDENCY_LOCK_SHA256 =
-  '55744d067f70ffcb0223eae6121739021a9e2567208cfd14098582bb73cdbf97'
+  '2867725e505a600f4f1f9d934fb797f5fb94a71c906577d05009aecc254f85f3'
 
 /** The published benchmark evidence was produced at this package version, by
  * the retired one-shot direct runner, before trace analysts moved to the