Skip to content

Garry's Opinionated OpenClaw/Hermes Agent Brain

TypeScript
7.5 KB
Raw Blame
{
  "decision_id": "entity-anchoring-query",
  "stage": "dev",
  "decided_at": "2026-10-05T22:00:00Z",
  "measured_build": "garrytan/gbrain@2fa44f149",
  "gate_1": {
    "status": "pass",
    "metered_spend_usd": 1.5243,
    "seeds": [
      {
        "seed": 42,
        "on_accuracy": 1,
        "off_accuracy": 0.875,
        "on_lag": 0,
        "off_lag": 0.15,
        "on_stale": 0,
        "off_stale": 0.125,
        "wins": true,
        "accuracy_lower": false
      },
      {
        "seed": 7,
        "on_accuracy": 1,
        "off_accuracy": 0.9027777777777778,
        "on_lag": 0,
        "off_lag": 0.11666666666666667,
        "on_stale": 0,
        "off_stale": 0.09722222222222222,
        "wins": true,
        "accuracy_lower": false
      },
      {
        "seed": 1234,
        "on_accuracy": 1,
        "off_accuracy": 0.875,
        "on_lag": 0,
        "off_lag": 0.15,
        "on_stale": 0,
        "off_stale": 0.125,
        "wins": true,
        "accuracy_lower": false
      }
    ],
    "triggered": {
      "42": "72/72",
      "7": "72/72",
      "1234": "72/72"
    },
    "aggregate": [
      {
        "reads_per_write": 1,
        "arm": "anchoring_on",
        "seeds": 3,
        "accuracy": {
          "mean": 1,
          "min": 1,
          "max": 1
        },
        "stale_wrong_rate": {
          "mean": 0,
          "min": 0,
          "max": 0
        },
        "mean_lag_batches": {
          "mean": 0,
          "min": 0,
          "max": 0
        },
        "correct_at_change": {
          "mean": 1,
          "min": 1,
          "max": 1
        },
        "lifecycle_usd": {
          "mean": 0,
          "min": 0,
          "max": 0
        },
        "usd_per_correct": {
          "mean": 0.001270787037037037,
          "min": 0.0012603888888888888,
          "max": 0.0012843055555555555
        }
      },
      {
        "reads_per_write": 1,
        "arm": "anchoring_off",
        "seeds": 3,
        "accuracy": {
          "mean": 0.8842592592592592,
          "min": 0.875,
          "max": 0.9027777777777778
        },
        "stale_wrong_rate": {
          "mean": 0.11574074074074074,
          "min": 0.09722222222222222,
          "max": 0.125
        },
        "mean_lag_batches": {
          "mean": 0.13888888888888887,
          "min": 0.11666666666666667,
          "max": 0.15
        },
        "correct_at_change": {
          "mean": 0.8611111111111112,
          "min": 0.85,
          "max": 0.8833333333333333
        },
        "lifecycle_usd": {
          "mean": 0,
          "min": 0,
          "max": 0
        },
        "usd_per_correct": {
          "mean": 0.001333720309320309,
          "min": 0.001280430769230769,
          "max": 0.0013843174603174605
        }
      },
      {
        "reads_per_write": 1,
        "arm": "anchoring_on_full",
        "seeds": 3,
        "accuracy": {
          "mean": 1,
          "min": 1,
          "max": 1
        },
        "stale_wrong_rate": {
          "mean": 0,
          "min": 0,
          "max": 0
        },
        "mean_lag_batches": {
          "mean": 0,
          "min": 0,
          "max": 0
        },
        "correct_at_change": {
          "mean": 1,
          "min": 1,
          "max": 1
        },
        "lifecycle_usd": {
          "mean": 0,
          "min": 0,
          "max": 0
        },
        "usd_per_correct": {
          "mean": 0.0022993333333333334,
          "min": 0.002282222222222222,
          "max": 0.002310111111111111
        }
      },
      {
        "reads_per_write": 1,
        "arm": "anchoring_off_full",
        "seeds": 3,
        "accuracy": {
          "mean": 0.9814814814814815,
          "min": 0.9722222222222222,
          "max": 0.9861111111111112
        },
        "stale_wrong_rate": {
          "mean": 0.009259259259259259,
          "min": 0,
          "max": 0.027777777777777776
        },
        "mean_lag_batches": {
          "mean": 0.022222222222222223,
          "min": 0.016666666666666666,
          "max": 0.03333333333333333
        },
        "correct_at_change": {
          "mean": 0.9777777777777777,
          "min": 0.9666666666666667,
          "max": 0.9833333333333333
        },
        "lifecycle_usd": {
          "mean": 0,
          "min": 0,
          "max": 0
        },
        "usd_per_correct": {
          "mean": 0.0023514169014084503,
          "min": 0.0023353802816901397,
          "max": 0.0023617999999999994
        }
      }
    ]
  },
  "gate_2": {
    "status": "pass",
    "reports": [
      {
        "corpus": "namedthing",
        "questions": 12,
        "triggered": 0,
        "recall10_lower": [],
        "recall10_higher": 0,
        "mean_recall10": {
          "on": 0.9166666666666667,
          "off": 0.9166666666666667
        },
        "mean_rr": {
          "on": 1,
          "off": 1
        },
        "negative_clean_lost": []
      },
      {
        "corpus": "namedthing+now",
        "questions": 12,
        "triggered": 0,
        "recall10_lower": [],
        "recall10_higher": 0,
        "mean_recall10": {
          "on": 0.8333333333333334,
          "off": 0.8333333333333334
        },
        "mean_rr": {
          "on": 0.9166666666666667,
          "off": 0.9166666666666667
        },
        "negative_clean_lost": []
      },
      {
        "corpus": "relational",
        "questions": 38,
        "triggered": 0,
        "recall10_lower": [],
        "recall10_higher": 0,
        "mean_recall10": {
          "on": 0.9999999999999992,
          "off": 0.9999999999999992
        },
        "mean_rr": {
          "on": 0.28890977443609034,
          "off": 0.28890977443609034
        },
        "negative_clean_lost": []
      },
      {
        "corpus": "relational+now",
        "questions": 38,
        "triggered": 24,
        "recall10_lower": [],
        "recall10_higher": 8,
        "mean_recall10": {
          "on": 0.13596491228070173,
          "off": 0
        },
        "mean_rr": {
          "on": 0.09649122807017543,
          "off": 0
        },
        "negative_clean_lost": []
      },
      {
        "corpus": "longmemeval-nightly",
        "questions": 10,
        "triggered": 0,
        "recall10_lower": [],
        "recall10_higher": 0,
        "mean_recall10": {
          "on": 0.9999999999999999,
          "off": 0.9999999999999999
        },
        "mean_rr": {
          "on": 0.9999999999999999,
          "off": 0.9999999999999999
        },
        "negative_clean_lost": []
      },
      {
        "corpus": "longmemeval-nightly+now",
        "questions": 10,
        "triggered": 0,
        "recall10_lower": [],
        "recall10_higher": 0,
        "mean_recall10": {
          "on": 0.9999999999999999,
          "off": 0.9999999999999999
        },
        "mean_rr": {
          "on": 0.9999999999999999,
          "off": 0.9999999999999999
        },
        "negative_clean_lost": []
      }
    ],
    "off_path": "key-off responses equal the pre-change build modulo clock-dependent recency scores and random revision UUIDs (120 responses)"
  },
  "gate_3": {
    "status": "pending",
    "owner": "harness lane",
    "requires": "retrieve through the query op (or gbrain query) with search.entity_anchoring=true"
  },
  "overall": "off_pending_gate_3",
  "caveats": [
    "The seeded workload was built around the failure mode: every question is entity-scoped current-state (216/216 triggered).",
    "Gate 2 never triggered on as-written questions; the cue-added relational diagnostic triggered 24/38 with 8 recall gains and 0 losses.",
    "Voyage embedding and rerank calls are not metered."
  ]
}