feat: optional Jev evaluation and release 1.18.0

.claude-plugin/marketplace.json

@@ -13,7 +13,7 @@
       "name": "sensegrep",
       "source": "./plugin/sensegrep-plugin",
       "description": "Semantic code search for AI agents. Search code by meaning, not text patterns. Adds sensegrep MCP tools + smart usage instructions to Claude Code.",
-      "version": "1.17.3",
+      "version": "1.18.0",
       "author": {
         "name": "sensegrep"
       },

.cursor-plugin/marketplace.json

@@ -13,7 +13,7 @@
       "name": "sensegrep",
       "source": "plugin/sensegrep-cursor",
       "description": "Semantic code search for AI agents. Search code by meaning, not text patterns.",
-      "version": "1.17.3"
+      "version": "1.18.0"
     }
   ]
 }

README.md

@@ -34,6 +34,8 @@ Traditional search tools (grep, ripgrep, ast-grep) match **text patterns**. sens
 
 ## Quickstart
 
+Optional remote evaluation: [Jev setup, evidence-aware reranking, group labels and duplicate judgments](docs/jev.md). It is off by default and works alongside local Ollama embeddings.
+
 ### Claude Code Plugin (recommended)
 
 The fastest way to get sensegrep into Claude Code — zero configuration:

docs/evaluations/jev-v2-2026-09-24/comparison-49.json

@@ -0,0 +1,39 @@
+{
+  "cache": "existing",
+  "totals": {
+    "baseline": {
+      "searches": 32,
+      "fileTop5": 30,
+      "symbolCases": 15,
+      "symbolTop5": 14,
+      "fileTop10": 32,
+      "reciprocalRank": 0.7669270833333333,
+      "contextCases": 11,
+      "contextHits": 11,
+      "jevRequests": 980,
+      "jevCacheHits": 0,
+      "jevReportedCost": 0.05617193399999999,
+      "medianMs": 5313,
+      "p95Ms": 11190,
+      "falseWeakWarnings": 0,
+      "negativesFlagged": 5
+    },
+    "candidate": {
+      "searches": 32,
+      "fileTop5": 30,
+      "symbolCases": 15,
+      "symbolTop5": 14,
+      "fileTop10": 32,
+      "reciprocalRank": 0.7721354166666666,
+      "contextCases": 11,
+      "contextHits": 11,
+      "jevRequests": 268,
+      "jevCacheHits": 0,
+      "jevReportedCost": 0.076875792,
+      "medianMs": 4742,
+      "p95Ms": 6044,
+      "falseWeakWarnings": 0,
+      "negativesFlagged": 4
+    }
+  }
+}

docs/evaluations/jev-v2-2026-09-24/duplicate-smoke.json

@@ -0,0 +1,39 @@
+{
+  "jev": {
+    "status": "complete",
+    "mode": "evidence",
+    "model": "typesafe/jev-1.13",
+    "models": [
+      "typesafe/jev-1.13-20260917"
+    ],
+    "evaluated": 1,
+    "candidates": 1,
+    "cacheHits": 0,
+    "requests": 1,
+    "inputTokens": 1356,
+    "outputTokens": 164,
+    "cost": 0.000056952,
+    "elapsedMs": 696,
+    "batchSize": 5,
+    "ranking": "score",
+    "splitRequests": 0
+  },
+  "judgements": [
+    {
+      "literalOperatorDifference": true,
+      "dimensions": {
+        "preconditions": 0.15,
+        "authorization": 0.55,
+        "effects": 0.19,
+        "errors": 0.23,
+        "mappings": 0.12
+      },
+      "relationship": "different-behavior",
+      "sameRule": 0.04,
+      "differentBehavior": 0.94,
+      "similarStructure": 0.92,
+      "advisory": true,
+      "sourceTruncated": false
+    }
+  ]
+}

docs/evaluations/jev-v2-2026-09-24/fallback-smoke.json

@@ -0,0 +1,47 @@
+{
+  "fallbackSameIds": true,
+  "scopePreserved": true,
+  "timeout": {
+    "status": "fallback",
+    "mode": "both",
+    "model": "typesafe/jev-1.13",
+    "models": [],
+    "evaluated": 0,
+    "candidates": 4,
+    "cacheHits": 0,
+    "requests": 1,
+    "inputTokens": 0,
+    "outputTokens": 0,
+    "cost": 0,
+    "elapsedMs": 115,
+    "batchSize": 5,
+    "ranking": "score",
+    "splitRequests": 0,
+    "reason": "deadline",
+    "packetStatus": "not-assessed"
+  },
+  "missing": {
+    "status": "fallback",
+    "mode": "both",
+    "model": "typesafe/jev-1.13",
+    "models": [],
+    "evaluated": 0,
+    "candidates": 4,
+    "cacheHits": 0,
+    "requests": 0,
+    "inputTokens": 0,
+    "outputTokens": 0,
+    "cost": 0,
+    "elapsedMs": 0,
+    "batchSize": 5,
+    "ranking": "score",
+    "splitRequests": 0,
+    "reason": "missing-api-key",
+    "packetStatus": "not-assessed"
+  },
+  "exact": {
+    "status": "skipped",
+    "reason": "exact-query",
+    "mode": "both"
+  }
+}

docs/evaluations/jev-v2-2026-09-24/packet-regressions-17.json

@@ -0,0 +1,35 @@
+{
+  "cache": "existing",
+  "totals": {
+    "baseline": {
+      "searches": 0,
+      "fileTop5": 0,
+      "symbolCases": 0,
+      "symbolTop5": 0,
+      "fileTop10": 0,
+      "reciprocalRank": null,
+      "contextCases": 11,
+      "contextHits": 11,
+      "jevRequests": 340,
+      "jevCacheHits": 0,
+      "jevReportedCost": 0.019264392000000005,
+      "falseWeakWarnings": 0,
+      "negativesFlagged": 5
+    },
+    "candidate": {
+      "searches": 0,
+      "fileTop5": 0,
+      "symbolCases": 0,
+      "symbolTop5": 0,
+      "fileTop10": 0,
+      "reciprocalRank": null,
+      "contextCases": 11,
+      "contextHits": 11,
+      "jevRequests": 108,
+      "jevCacheHits": 0,
+      "jevReportedCost": 0.027716808000000002,
+      "falseWeakWarnings": 0,
+      "negativesFlagged": 6
+    }
+  }
+}