renkin 1.0.6 → 1.0.8

package/README.md

@@ -17,6 +17,14 @@ RENKIN has two uses:
 Auditing is local and reproducible. Reports include structural checks, stock
 coverage, forward replay, provenance, and a verifiable audit manifest.
 
+Current release: **v1.0.8**. Public inputs are validated before execution;
+WASM applies bounded search limits, and MCP numeric and element-filter
+arguments fail closed when invalid. Standard MCP search also accepts an
+optional cooperative `timeout_secs` budget.
+
+The audit and private-stock policy layers are split into deterministic,
+testable steps so their report schema and policy behavior remain stable.
+
 ## Install
 
 ```bash
@@ -121,14 +129,16 @@ explanation, constraints, diagnostics, and audit receipts. See the [MCP guide](h
 
 ## Benchmark status
 
-The latest checked-in formal comparison uses a declared shared-stock endpoint.
-RENKIN and AiZynthFinder reached the same route count in the published
-equal-condition comparison; universal CASP superiority is not proven.
-Success rate, speed, stock definition, validation, and route quality must be
-reported separately.
+The registered Phase 55 shared-stock, shared-budget TEST comparison completed
+690 targets: RENKIN found 481 strict routes (69.71%) and AiZynthFinder 4.4.1
+found 32 (4.64%). The paired coverage difference was +65.07 percentage points
+(95% CI +61.45 to +68.55). This is evidence only for the pinned cohort,
+assets, stock, and budget; it is neither a universal CASP claim nor evidence
+of laboratory success. Peak-RSS and time-to-first-route receipts remain open,
+so it is not a whole-cohort performance claim.
 
 See the [benchmark documentation](https://kent-tokyo.github.io/renkin/benchmark/)
-for the full measurements, methodology, confidence intervals, and limitations.
+for protocol, artifacts, and claim boundaries.
 
 ## Development
 
@@ -146,6 +156,8 @@ Important boundaries:
 - stock identity is exact standardized canonical-SMILES membership;
 - route success is not experimental success;
 - external model output is evidence or candidates, not automatic validity;
+- comparison manifests bind the tool, configuration, input files, and checked
+  worktree state so resumed runs cannot silently mix configurations;
 - benchmark claims must name the dataset, stock, versions, and endpoint.
 
 ## License

package/package.json

@@ -5,7 +5,7 @@
     "kent-tokyo <kent-tokyo@users.noreply.github.com>"
   ],
   "description": "Ultra-fast retrosynthesis engine for computer-aided synthesis planning (CASP) — pure Rust, WASM-ready, Python bindings via PyO3",
-  "version": "1.0.6",
+  "version": "1.0.8",
   "license": "MIT",
   "repository": {
     "type": "git",