@functional-systems/srcx 0.1.0-dev.14 → 0.1.0-dev.16

package/CHANGELOG.md

@@ -2,6 +2,12 @@
 
 ## Unreleased
 
+- Evaluate fixed-candidate CosQA semantic reranking with local Qwen, shared
+  duplicate scores, stable ties and separate official-ID/content metrics.
+
+- Cross-check the retained CosQA lexical baseline with an independent offline
+  BM25 ranker and the same StandardAnalyzer tokenization.
+
 - Add pinned CoIR CoSQA and MTEB CodeSearchNet Python/Go retrieval baselines,
   sharing srcx query construction and using trec_eval metrics on official labels.
 

package/VALIDATION.md

@@ -1,6 +1,21 @@
 # Validation record
 
-Date: 2026-09-25 (Asia/Seoul).
+Date: 2026-09-26 (Asia/Seoul).
+
+## Fixed-candidate CosQA reranking
+
+The [completed Qwen comparison](eval/COSQA-RERANK-RESULTS.md), run on September 26
+(Asia/Seoul), improved exact-content hit@10 from 57.4% to 64.8% with roughly one
+second of local model computation per query. All 500 original candidate pools
+and duplicate-ID ordering were preserved. The report separates official-ID
+metrics, content-equivalence diagnostics, regressions and local latency.
+
+## Independent CosQA BM25 cross-check
+
+The [independent cross-check](eval/COSQA-BM25-RESULTS.md) reproduces lexical
+quality after aligning score ties and documents duplicate-ID sensitivity in the
+official task. Its supplemental content-equivalence metrics are distinct from
+the unchanged official-label baseline.
 
 ## Public code retrieval baseline
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@functional-systems/srcx",
-  "version": "0.1.0-dev.14",
+  "version": "0.1.0-dev.16",
   "private": false,
   "description": "Version-aware code search on LambdaDB",
   "type": "module",
@@ -62,5 +62,5 @@
     "url": "https://github.com/lambdadb/srcx/issues"
   },
   "license": "Apache-2.0",
-  "gitHead": "5a214d6124a84f831f4595c159576d319b94f7fc"
+  "gitHead": "1da86320e6ef5446b41f3d7f9d7d92d536a33132"
 }