Garry's Opinionated OpenClaw/Hermes Agent Brain
TypeScript
{
"decision_id": "p2-e1-hub-dampening-heldout",
"hypothesis": "P2 E1 search.hub_dampening (H = 32) on sealed hub-world seeds 2 and 3",
"computed_by": "second custodian (gbra49), P0 unresponsive; from P0's custody/p2-hub and vmjobs/hubC pull",
"builds": {
"candidate": "gbrain#6020 af1225d7b (arms off, 32, graph-off)",
"rival_remove": "31a792c94 (BACKLINK_BOOST_COEF = 0)",
"rival_cap": "4e2bd760e (backlink and graph-signal factors capped at 1.02)"
},
"bar": [
"concept nDCG@5 delta >= +1.0 point with 95% CI above 0",
"better than both rivals on concept",
"hub-as-answer nDCG@5 delta >= -0.5 point",
"one-hop recall@5 delta >= -0.5 point"
],
"bootstrap": "paired over probes, 10,000 draws, seed 42; points",
"workloads": {
"seed2-keyword": {
"abs_off": {
"concept": 38.0,
"hub-answer": 96.3
},
"H32": {
"concept_ndcg5": [
3.29,
[
2.46,
4.18
]
],
"hub_answer_ndcg5": [
-30.62,
[
-45.0,
-16.77
]
],
"onehop_recall5": [
-5.36,
[
-9.07,
-2.01
]
],
"onehop_hit1": [
-1.38,
[
-8.28,
5.52
]
]
},
"graph-off": {
"concept_ndcg5": [
0.02,
[
-0.13,
0.2
]
],
"hub_answer_ndcg5": [
1.85,
[
0.0,
5.54
]
],
"onehop_recall5": [
-3.59,
[
-6.0,
-1.52
]
],
"onehop_hit1": [
-4.14,
[
-8.28,
0.0
]
]
},
"remove": {
"concept_ndcg5": [
6.41,
[
5.17,
7.66
]
],
"hub_answer_ndcg5": [
-1.31,
[
-8.15,
4.88
]
],
"onehop_recall5": [
-9.8,
[
-14.01,
-5.85
]
],
"onehop_hit1": [
-13.1,
[
-20.69,
-5.52
]
]
},
"cap": {
"concept_ndcg5": [
5.85,
[
4.69,
7.04
]
],
"hub_answer_ndcg5": [
0.0,
[
-5.54,
5.54
]
],
"onehop_recall5": [
-9.63,
[
-14.03,
-5.46
]
],
"onehop_hit1": [
-10.34,
[
-17.93,
-2.76
]
]
},
"H32_minus_remove_concept": [
-3.12,
[
-3.98,
-2.33
]
],
"H32_minus_cap_concept": [
-2.56,
[
-3.32,
-1.85
]
],
"n_valid_H32": {
"concept": 428,
"hub-answer": 20,
"one-hop": 145
},
"gate": {
"1_primary": true,
"2_beats_rivals": false,
"3_hub_answer": false,
"4_onehop": false
}
},
"seed3-keyword": {
"abs_off": {
"concept": 38.0,
"hub-answer": 98.2
},
"H32": {
"concept_ndcg5": [
3.63,
[
2.77,
4.52
]
],
"hub_answer_ndcg5": [
-36.15,
[
-52.97,
-19.85
]
],
"onehop_recall5": [
-1.03,
[
-3.79,
1.38
]
],
"onehop_hit1": [
-2.07,
[
-8.28,
4.83
]
]
},
"graph-off": {
"concept_ndcg5": [
-0.04,
[
-0.18,
0.11
]
],
"hub_answer_ndcg5": [
1.85,
[
0.0,
5.54
]
],
"onehop_recall5": [
-3.24,
[
-5.66,
-1.17
]
],
"onehop_hit1": [
-5.52,
[
-9.66,
-1.38
]
]
},
"remove": {
"concept_ndcg5": [
6.47,
[
5.32,
7.67
]
],
"hub_answer_ndcg5": [
-11.73,
[
-19.76,
-3.69
]
],
"onehop_recall5": [
-7.21,
[
-10.83,
-3.97
]
],
"onehop_hit1": [
-14.48,
[
-22.07,
-7.59
]
]
},
"cap": {
"concept_ndcg5": [
5.9,
[
4.84,
7.01
]
],
"hub_answer_ndcg5": [
-2.5,
[
-9.35,
3.69
]
],
"onehop_recall5": [
-6.69,
[
-10.48,
-3.24
]
],
"onehop_hit1": [
-11.72,
[
-18.62,
-4.83
]
]
},
"H32_minus_remove_concept": [
-2.84,
[
-3.6,
-2.14
]
],
"H32_minus_cap_concept": [
-2.27,
[
-2.93,
-1.67
]
],
"n_valid_H32": {
"concept": 428,
"hub-answer": 20,
"one-hop": 145
},
"gate": {
"1_primary": true,
"2_beats_rivals": false,
"3_hub_answer": false,
"4_onehop": false
}
},
"seed2-hybrid": {
"valid_rows_only": {
"abs_off": {
"concept": 47.9,
"hub-answer": 94.5
},
"H32": {
"concept_ndcg5": [
0.96,
[
0.28,
1.89
]
],
"hub_answer_ndcg5": [
-24.46,
[
-42.62,
-8.15
]
],
"onehop_recall5": [
0.31,
[
-4.59,
5.2
]
],
"onehop_hit1": [
1.02,
[
-4.08,
6.12
]
]
},
"graph-off": {
"concept_ndcg5": [
0.0,
[
0.0,
0.0
]
],
"hub_answer_ndcg5": [
0.0,
[
0.0,
0.0
]
],
"onehop_recall5": [
-1.03,
[
-2.76,
0.0
]
],
"onehop_hit1": [
-0.69,
[
-3.45,
1.38
]
]
},
"remove": {
"concept_ndcg5": [
-0.3,
[
-1.38,
0.64
]
],
"hub_answer_ndcg5": [
-11.88,
[
-23.81,
-2.62
]
],
"onehop_recall5": [
-0.14,
[
-2.9,
2.76
]
],
"onehop_hit1": [
-2.76,
[
-6.21,
0.0
]
]
},
"cap": {
"concept_ndcg5": [
0.79,
[
0.06,
1.57
]
],
"hub_answer_ndcg5": [
-10.72,
[
-19.88,
-3.0
]
],
"onehop_recall5": [
0.21,
[
-2.76,
3.1
]
],
"onehop_hit1": [
-2.76,
[
-6.9,
0.69
]
]
},
"H32_minus_remove_concept": [
0.39,
[
-1.21,
2.56
]
],
"H32_minus_cap_concept": [
-0.86,
[
-1.59,
-0.3
]
],
"n_valid_H32": {
"concept": 129,
"hub-answer": 20,
"one-hop": 98
},
"gate": {
"1_primary": false,
"2_beats_rivals": false,
"3_hub_answer": false,
"4_onehop": true
}
},
"telemetry_noop_rows_imputed_as_off": {
"abs_off": {
"concept": 47.9,
"hub-answer": 94.5
},
"H32": {
"concept_ndcg5": [
0.29,
[
0.08,
0.57
]
],
"hub_answer_ndcg5": [
-24.46,
[
-42.62,
-8.15
]
],
"onehop_recall5": [
0.21,
[
-3.03,
3.52
]
],
"onehop_hit1": [
0.69,
[
-2.76,
4.14
]
]
},
"graph-off": {
"concept_ndcg5": [
0.0,
[
0.0,
0.0
]
],
"hub_answer_ndcg5": [
0.0,
[
0.0,
0.0
]
],
"onehop_recall5": [
-1.03,
[
-2.76,
0.0
]
],
"onehop_hit1": [
-0.69,
[
-3.45,
1.38
]
]
},
"remove": {
"concept_ndcg5": [
-0.3,
[
-1.38,
0.64
]
],
"hub_answer_ndcg5": [
-11.88,
[
-23.81,
-2.62
]
],
"onehop_recall5": [
-0.14,
[
-2.9,
2.76
]
],
"onehop_hit1": [
-2.76,
[
-6.21,
0.0
]
]
},
"cap": {
"concept_ndcg5": [
0.79,
[
0.06,
1.57
]
],
"hub_answer_ndcg5": [
-10.72,
[
-19.88,
-3.0
]
],
"onehop_recall5": [
0.21,
[
-2.76,
3.1
]
],
"onehop_hit1": [
-2.76,
[
-6.9,
0.69
]
]
},
"H32_minus_remove_concept": [
0.59,
[
-0.29,
1.63
]
],
"H32_minus_cap_concept": [
-0.5,
[
-1.2,
0.15
]
],
"n_valid_H32": {
"concept": 428,
"hub-answer": 20,
"one-hop": 145
},
"gate": {
"1_primary": false,
"2_beats_rivals": false,
"3_hub_answer": false,
"4_onehop": true
}
}
},
"seed3-hybrid": {
"missing": [
"candidate cell (arms off, 32, graph-off)",
"rival_cap cell"
],
"present": [
"rival_remove cell"
],
"where": "P0 machine, Ubicloud VM ubirun-gbra49-1791261782-aeed909c (job hubD), still running at 14:19 UTC 2026-10-06"
}
},
"hybrid_telemetry_note": "In seed2-hybrid, 346 of 625 H = 32 probe rows (299 concept, 47 one-hop) are recorded as errors because search telemetry carried no hub_dampening stamp. On those queries the lexical metadata-boost gate skipped the backlink and graph-signal stages, so dampening had nothing to act on and the ranking equals arm off. Both treatments are reported; neither changes the verdict.",
"verdict": "fail",
"why": "Both keyword workloads fail conditions 2, 3 and 4: the rivals beat H = 32 on concept by 2.3 to 3.1 points with CIs below 0, and hub-as-answer drops 30.6 and 36.2 points, and one-hop recall@5 drops 5.4 and 1.0 points. Seed-2 hybrid fails conditions 1, 2 and 3 (concept +0.3 imputed / +1.0 valid-only; hub-as-answer -24.5). The missing seed-3 hybrid cells cannot change the outcome: hub-as-answer would need a gain of about +89 points there to bring even a pooled delta to -0.5.",
"consequence": "search.hub_dampening stays off and the search-side mechanism is removed (preregistered); the hubWeight helper stays for relational-chain.ts"
}