Release graph-tool-call 0.40.0

CHANGELOG.md

@@ -7,6 +7,22 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.40.0] - 2026-08-15
+
+### Added
+- Structured semantic retrieval signals for action, resource, result shape,
+  module, contract, and concise summary evidence in metadata-rich catalogs.
+- Deterministic score tie-breaking and graph traversal across insertion order
+  and Python hash seeds.
+
+### Changed
+- Dense dependency-graph construction now bounds ubiquitous structural fields
+  so large OpenAPI catalogs avoid low-value edge explosions.
+
+### Fixed
+- Optional Kiwi tokenization now runs an isolated runtime health check and
+  falls back safely when the native extension is unstable or nondeterministic.
+
 ## [0.39.0] - 2026-08-09
 
 ### Added
@@ -651,7 +667,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - **Tests**: 32 tests passing across all modules
 - **Example**: `quickstart.py` demonstrating full workflow
 
-[Unreleased]: https://github.com/SonAIengine/graph-tool-call/compare/v0.39.0...HEAD
+[Unreleased]: https://github.com/SonAIengine/graph-tool-call/compare/v0.40.0...HEAD
 [0.39.0]: https://github.com/SonAIengine/graph-tool-call/compare/v0.38.0...v0.39.0
 [0.38.0]: https://github.com/SonAIengine/graph-tool-call/compare/v0.37.0...v0.38.0
 [0.37.0]: https://github.com/SonAIengine/graph-tool-call/compare/v0.36.0...v0.37.0
@@ -689,3 +705,4 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 [0.1.0]: https://github.com/SonAIengine/graph-tool-call/releases/tag/v0.1.0
 [0.8.0]: https://github.com/SonAIengine/graph-tool-call/compare/v0.5.0...v0.8.0
 [0.28.0]: https://github.com/SonAIengine/graph-tool-call/compare/v0.27.0...v0.28.0
+[0.40.0]: https://github.com/SonAIengine/graph-tool-call/compare/v0.39.0...v0.40.0

README.md

@@ -230,7 +230,7 @@ selected target produced:
 | Candidate binding support | 14.3% | **100%** |
 | Target Recall@5 | - | **100%** |
 
-The [case-level v0.39.0 artifact](benchmarks/results/releases/v0.39.0/dependency-chain-evidence.json)
+The [case-level v0.40.0 artifact](benchmarks/results/releases/v0.40.0/dependency-chain-evidence.json)
 records fixture hashes, every expected target and producer, and replay commands:
 
 ```bash
@@ -239,7 +239,7 @@ make launch-evidence-check
 ```
 
 The separate
-[observability artifact](benchmarks/results/releases/v0.39.0/observability-evidence.json)
+[observability artifact](benchmarks/results/releases/v0.40.0/observability-evidence.json)
 checks that tracing leaves engine inputs unchanged, replays deterministically,
 scrubs secrets, explains every decision, and stays below the documented
 `5ms/span` p95 capture-cost gate:

benchmarks/results/releases/v0.40.0/README.md

@@ -0,0 +1,26 @@
+# v0.40.0 release evidence
+
+This directory contains the model-free release artifacts for graph-tool-call
+0.40.0.
+
+- `dependency-chain-evidence.json` records the deterministic target and
+  prerequisite-producer regression used by the README.
+- `observability-evidence.json` records result invariance, deterministic replay,
+  secret scrubbing, reason coverage, serialized trace size, and measured trace
+  capture overhead.
+- Dense-catalog dependency bounds, structured semantic retrieval, stable score
+  ordering, and optional Kiwi fallback are covered by the public retrieval,
+  scale, determinism, and tokenizer tests.
+
+Regenerate and validate the artifacts with:
+
+```bash
+make launch-evidence
+make launch-evidence-check
+make observability-evidence
+make observability-evidence-check
+```
+
+The observability latency value is a local Python microbenchmark. The check
+runs a fresh measurement and enforces the documented `5ms/span` p95 ceiling;
+it does not measure exporter backend or service network latency.