mcvay-mind 1.0.0 → 1.0.2

package/SKILL.md

@@ -66,9 +66,11 @@ node ~/.openclaw/skills/mcvay-mind/index.js query --type preference --days 7
 
 # Search (unified)
 node ~/.openclaw/skills/mcvay-mind/index.js search "query terms"
+node ~/.openclaw/skills/mcvay-mind/index.js search "latest timeout preference" --mode hybrid-v2 --graph-boost 0.25
 
 # Active recall (context surfacing)
 node ~/.openclaw/skills/mcvay-mind/index.js recall "topic"
+node ~/.openclaw/skills/mcvay-mind/index.js recall "codex coding" --graph-boost 0.3 --include-stale
 
 # Entity linking
 node ~/.openclaw/skills/mcvay-mind/index.js link
@@ -142,6 +144,12 @@ created: 2026-02-16T12:00:00.000Z
 updated: 2026-02-16T12:00:00.000Z
 tags: [tag1, tag2]
 links: [decision/choice-1, preference/user-pref]
+entities: [codex, timeout]
+valid_from: 2026-02-16T12:00:00.000Z
+valid_to: 2026-03-01T00:00:00.000Z
+supersedes: [preference/old-timeout]
+conflicts_with: [lesson/contrary-observation]
+salience: 0.75
 confidence: 90
 source: agent
 ---
@@ -188,6 +196,7 @@ Keywords that should trigger recall:
 ├── lib/
 │   ├── store.js       # Memory CRUD operations
 │   ├── search.js      # Full-text search
+│   ├── graph-temporal.js # Typed graph traversal + temporal ranking
 │   └── entity-linker.js # Wiki-link extraction & knowledge graph
 └── schema/            # YAML schemas
     ├── base.yaml

package/bench/README.md

@@ -0,0 +1,49 @@
+# Benchmark Harness
+
+Offline benchmark + auto-tuning for McVay Mind retrieval.
+
+## Commands
+
+From workspace root:
+
+- `npm run bench:all` - tune + benchmark + regression gate
+- `npm run bench:run` - benchmark with latest/default config
+- `npm run bench:tune` - run hyperparameter tuning only
+- `npm run bench:gate` - evaluate latest run vs baseline
+
+From `skills/mcvay-mind` (equivalent):
+
+- `npm run bench:all` - tune + benchmark + regression gate
+- `npm run bench:run` - benchmark with latest/default config
+- `npm run bench:tune` - run hyperparameter tuning only
+- `npm run bench:gate` - evaluate latest run vs baseline
+
+## Metrics
+
+- Recall@k
+- MRR
+- nDCG@k
+- Latency (mean, p50, p95)
+- Bootstrap confidence intervals
+- Paired bootstrap p-value for recall lift
+
+## Offline
+
+For strict offline runs, set:
+
+```bash
+MCVAY_EMBED_ENABLED=false npm run bench:all
+```
+
+## Artifacts
+
+Written to `bench/artifacts/`:
+
+- `runs/<timestamp>/summary.md`
+- `runs/<timestamp>/metrics.csv`
+- `runs/<timestamp>/results.json`
+- `best-config-<timestamp>.json`
+- `best-config-latest.json`
+- `baseline.json`
+- `latest.json`
+- `gate-latest.json`

package/bench/artifacts/baseline.json

@@ -0,0 +1,106 @@
+{
+  "created": "2026-02-20T01:21:09.340Z",
+  "runId": "20260220T011934Z",
+  "hybridSummary": {
+    "sampleSize": 4,
+    "metrics": {
+      "recallAtK": {
+        "mean": 0.25,
+        "lower": 0,
+        "upper": 0.75
+      },
+      "mrr": {
+        "mean": 0.125,
+        "lower": 0,
+        "upper": 0.375
+      },
+      "ndcgAtK": {
+        "mean": 0.15773243839286438,
+        "lower": 0,
+        "upper": 0.47319731517859315
+      }
+    },
+    "latency": {
+      "meanMs": 3445.17079825,
+      "p50Ms": 3344.280292,
+      "p95Ms": 3533.913265
+    },
+    "perTask": [
+      {
+        "rr": 0,
+        "recall": 0,
+        "ndcg": 0,
+        "latencyMs": 3344.280292
+      },
+      {
+        "rr": 0,
+        "recall": 0,
+        "ndcg": 0,
+        "latencyMs": 4020.542328
+      },
+      {
+        "rr": 0.5,
+        "recall": 1,
+        "ndcg": 0.6309297535714575,
+        "latencyMs": 3533.913265
+      },
+      {
+        "rr": 0,
+        "recall": 0,
+        "ndcg": 0,
+        "latencyMs": 2881.947308
+      }
+    ]
+  },
+  "candidateSummary": {
+    "sampleSize": 4,
+    "metrics": {
+      "recallAtK": {
+        "mean": 0,
+        "lower": 0,
+        "upper": 0
+      },
+      "mrr": {
+        "mean": 0,
+        "lower": 0,
+        "upper": 0
+      },
+      "ndcgAtK": {
+        "mean": 0,
+        "lower": 0,
+        "upper": 0
+      }
+    },
+    "latency": {
+      "meanMs": 0.76737475,
+      "p50Ms": 0.746852,
+      "p95Ms": 0.758073
+    },
+    "perTask": [
+      {
+        "rr": 0,
+        "recall": 0,
+        "ndcg": 0,
+        "latencyMs": 0.746852
+      },
+      {
+        "rr": 0,
+        "recall": 0,
+        "ndcg": 0,
+        "latencyMs": 0.858205
+      },
+      {
+        "rr": 0,
+        "recall": 0,
+        "ndcg": 0,
+        "latencyMs": 0.758073
+      },
+      {
+        "rr": 0,
+        "recall": 0,
+        "ndcg": 0,
+        "latencyMs": 0.706369
+      }
+    ]
+  }
+}

package/bench/artifacts/best-config-20260220T011934Z.json

@@ -0,0 +1,13 @@
+{
+  "config": {
+    "minSimilarity": 0.25,
+    "rrfK": 60,
+    "bm25TopN": 60,
+    "annTopN": 80,
+    "graphBoost": 0
+  },
+  "recallAtK": 0.3333333333333333,
+  "mrr": 0.25,
+  "ndcgAtK": 0.271821625595243,
+  "latencyP95Ms": 106.768316
+}

package/bench/artifacts/best-config-20260220T014624Z.json

@@ -0,0 +1,13 @@
+{
+  "config": {
+    "minSimilarity": 0.25,
+    "rrfK": 40,
+    "bm25TopN": 40,
+    "annTopN": 60,
+    "graphBoost": 0
+  },
+  "recallAtK": 0.5714285714285714,
+  "mrr": 0.369047619047619,
+  "ndcgAtK": 0.42091351862245524,
+  "latencyP95Ms": 3378.864337
+}

package/bench/artifacts/best-config-latest.json

@@ -0,0 +1,13 @@
+{
+  "config": {
+    "minSimilarity": 0.25,
+    "rrfK": 40,
+    "bm25TopN": 40,
+    "annTopN": 60,
+    "graphBoost": 0
+  },
+  "recallAtK": 0.5714285714285714,
+  "mrr": 0.369047619047619,
+  "ndcgAtK": 0.42091351862245524,
+  "latencyP95Ms": 3378.864337
+}

package/bench/artifacts/gate-latest.json

@@ -0,0 +1,22 @@
+{
+  "timestamp": "2026-02-20T02:56:46.872Z",
+  "baselineRunId": "20260220T011934Z",
+  "latestRunId": "20260220T014624Z",
+  "gate": {
+    "pass": false,
+    "recallDrop": -833333333.3333334,
+    "latencyRegression": 3761.516071671198,
+    "failRecall": false,
+    "failLatency": true
+  },
+  "acceptance": {
+    "pass": false,
+    "latencyRegression": -0.07326701690082414,
+    "pValueThreshold": 0.05,
+    "maxLatencyRegression": 0.2,
+    "failLift": true,
+    "failPValue": true,
+    "failLatency": false,
+    "failReproducibility": true
+  }
+}

package/bench/artifacts/latest.json

@@ -0,0 +1,150 @@
+{
+  "created": "2026-02-20T02:56:46.869Z",
+  "runId": "20260220T014624Z",
+  "baseline": {
+    "sampleSize": 6,
+    "metrics": {
+      "recallAtK": {
+        "mean": 0.8333333333333334,
+        "lower": 0.5,
+        "upper": 1
+      },
+      "mrr": {
+        "mean": 0.6111111111111112,
+        "lower": 0.27777777777777773,
+        "upper": 0.888888888888889
+      },
+      "ndcgAtK": {
+        "mean": 0.6666666666666666,
+        "lower": 0.3333333333333333,
+        "upper": 0.9166666666666666
+      }
+    },
+    "latency": {
+      "meanMs": 2599.9500551666665,
+      "p50Ms": 2385.250471,
+      "p95Ms": 3077.760151
+    },
+    "perTask": [
+      {
+        "rr": 1,
+        "recall": 1,
+        "ndcg": 1,
+        "latencyMs": 3070.747261
+      },
+      {
+        "rr": 0,
+        "recall": 0,
+        "ndcg": 0,
+        "latencyMs": 3077.760151
+      },
+      {
+        "rr": 1,
+        "recall": 1,
+        "ndcg": 1,
+        "latencyMs": 3776.156462
+      },
+      {
+        "rr": 1,
+        "recall": 1,
+        "ndcg": 1,
+        "latencyMs": 2385.250471
+      },
+      {
+        "rr": 0.3333333333333333,
+        "recall": 1,
+        "ndcg": 0.5,
+        "latencyMs": 1707.033999
+      },
+      {
+        "rr": 0.3333333333333333,
+        "recall": 1,
+        "ndcg": 0.5,
+        "latencyMs": 1582.751987
+      }
+    ]
+  },
+  "candidate": {
+    "sampleSize": 6,
+    "metrics": {
+      "recallAtK": {
+        "mean": 0.8333333333333334,
+        "lower": 0.5,
+        "upper": 1
+      },
+      "mrr": {
+        "mean": 0.6111111111111112,
+        "lower": 0.27777777777777773,
+        "upper": 0.888888888888889
+      },
+      "ndcgAtK": {
+        "mean": 0.6666666666666666,
+        "lower": 0.3333333333333333,
+        "upper": 0.9166666666666666
+      }
+    },
+    "latency": {
+      "meanMs": 2237.8700451666664,
+      "p50Ms": 2144.505668,
+      "p95Ms": 2852.261846
+    },
+    "perTask": [
+      {
+        "rr": 1,
+        "recall": 1,
+        "ndcg": 1,
+        "latencyMs": 2216.417736
+      },
+      {
+        "rr": 0,
+        "recall": 0,
+        "ndcg": 0,
+        "latencyMs": 2852.261846
+      },
+      {
+        "rr": 1,
+        "recall": 1,
+        "ndcg": 1,
+        "latencyMs": 3482.541869
+      },
+      {
+        "rr": 1,
+        "recall": 1,
+        "ndcg": 1,
+        "latencyMs": 2144.505668
+      },
+      {
+        "rr": 0.3333333333333333,
+        "recall": 1,
+        "ndcg": 0.5,
+        "latencyMs": 1443.53379
+      },
+      {
+        "rr": 0.3333333333333333,
+        "recall": 1,
+        "ndcg": 0.5,
+        "latencyMs": 1287.959362
+      }
+    ]
+  },
+  "significance": {
+    "baseline": "mcvay-hybrid",
+    "candidate": "mcvay-hybrid-v2",
+    "lift": 0,
+    "pValue": 1
+  },
+  "reproducibility": {
+    "variancePct": 25.00000000000001,
+    "stable": false
+  },
+  "acceptance": {
+    "pass": false,
+    "latencyRegression": -0.07326701690082414,
+    "pValueThreshold": 0.05,
+    "maxLatencyRegression": 0.2,
+    "failLift": true,
+    "failPValue": true,
+    "failLatency": false,
+    "failReproducibility": true
+  }
+}

package/bench/artifacts/runs/20260220T011934Z/metrics.csv

@@ -0,0 +1,7 @@
+variant,recall_at_k,mrr,ndcg_at_k,latency_mean_ms,latency_p95_ms,sample_size
+mcvay-keyword,0.250000,0.125000,0.157732,3234.290,3370.587,4
+mcvay-semantic,0.000000,0.000000,0.000000,3083.167,3131.873,4
+mcvay-hybrid,0.250000,0.125000,0.157732,3445.171,3533.913,4
+mcvay-hybrid-v2-default,0.000000,0.000000,0.000000,0.874,0.920,4
+mcvay-hybrid-v2,0.000000,0.000000,0.000000,0.767,0.758,4
+mem0-local-fallback-lexical,1.000000,0.875000,0.907732,1.672,1.847,4