Spaces:
Running
Running
File size: 4,741 Bytes
7a69d2b | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 | {
"schema_version": "mat-nexus-public-benchmark-catalog-v1",
"generated_on": "2026-08-02",
"scope": "Aggregate results only. Questions, answers, prompts, router tables and private implementation are excluded.",
"primary_comparisons": [
{
"id": "short-18-granite",
"campaign_id": "nexus-safe-short-levels-v1",
"model": "Granite 3.3 2B Instruct",
"domain": "mixed_verifiable",
"questions": 18,
"raw_correct": 3,
"nexus_correct": 18,
"wins": 15,
"losses": 0,
"classification": "GAIN",
"two_sided_exact_p_value": 0.00006103515625,
"evidence_status": "AUDITED",
"report_sha256": "554fc022f5f445927ce5f1ea5a24546a505865a359c82293940a9581c97e2aa4",
"audit_sha256": "758656d58407c1c08c9bc0005f453eaf2a9c6076e76d4a5276a80b768d2bdc17"
},
{
"id": "short-18-gemma",
"campaign_id": "nexus-safe-short-levels-v1",
"model": "Gemma 3 12B IT QAT",
"domain": "mixed_verifiable",
"questions": 18,
"raw_correct": 3,
"nexus_correct": 15,
"wins": 12,
"losses": 0,
"classification": "GAIN",
"two_sided_exact_p_value": 0.00048828125,
"evidence_status": "AUDITED",
"report_sha256": "554fc022f5f445927ce5f1ea5a24546a505865a359c82293940a9581c97e2aa4",
"audit_sha256": "758656d58407c1c08c9bc0005f453eaf2a9c6076e76d4a5276a80b768d2bdc17"
},
{
"id": "micro-6-llama",
"campaign_id": "nexus-safe-three-model-micro-v1",
"model": "Llama 3.1 8B",
"domain": "mixed_verifiable",
"questions": 6,
"raw_correct": 1,
"nexus_correct": 6,
"wins": 5,
"losses": 0,
"classification": "INCONCLUSIVE",
"two_sided_exact_p_value": 0.0625,
"evidence_status": "AUDITED",
"report_sha256": "dc0c55cb8fe517128a79a039e4a9f34be85ff00cc9f7326225bd6bb4774a10f6",
"audit_sha256": "c50311824a2c7ae215af8f02a8b1e5f99c176483f7f877a69aaba2edccf5b801"
},
{
"id": "code-5-granite",
"campaign_id": "mat9f-code-e2e-raw-vs-nexus-v1",
"model": "Granite 3.3 2B Instruct",
"domain": "code",
"questions": 5,
"raw_correct": 1,
"nexus_correct": 1,
"wins": 0,
"losses": 0,
"classification": "INCONCLUSIVE",
"two_sided_exact_p_value": 1.0,
"evidence_status": "AUDITED_SMALL_SAMPLE",
"report_sha256": "69db61ec98a2e5c445833f00be8127b1327effc13ce8194ce9e05de6cd7a095e",
"audit_sha256": null
}
],
"diagnostic_campaigns": [
{
"id": "nexus-four-arm-1000q-v1",
"questions": 1000,
"kind": "component_ablation",
"status": "HISTORICAL_DIAGNOSTIC",
"metrics": {
"raw_accuracy": 0.0,
"prepared_accuracy": 0.019,
"experts_accuracy": 0.812,
"experts_memory_accuracy": 0.905
},
"report_sha256": "e3e23ada5f197b14ccdbfe542e791033fe6d564a22ed06b08f907bc906ff5c64",
"caveat": "Component ablation; not pooled with the current paired safe-layer benchmark."
},
{
"id": "nexus-adapter-ab-1000q-v1-safe",
"questions": 1000,
"kind": "adapter_nexus_ablation",
"status": "HISTORICAL_DIAGNOSTIC",
"metrics": {
"base_raw_accuracy": 0.0,
"adapted_raw_accuracy": 0.0,
"base_nexus_accuracy": 0.125,
"adapted_nexus_accuracy": 0.186
},
"report_sha256": "14bb4052eb05de8580231c491ca7db7444a365ec813c7de235b3f73632f8a22f",
"caveat": "Useful for architecture diagnosis; raw-arm behavior makes it unsuitable as the headline gain claim."
},
{
"id": "full-circuit-all-functional-ab-v2-safe",
"questions": 10,
"kind": "controlled_preview",
"status": "EXPLORATORY",
"metrics": {
"llm_alone_accuracy": 0.0,
"experts_raw_plus_llm_accuracy": 0.0,
"experts_prepared_plus_llm_accuracy": 0.1
},
"report_sha256": "260f583eec371f6b1242d4fbb48b9c9056daf90bc8a2e431c7b9aa6b543cd822",
"caveat": "Controlled preview, explicitly not an official benchmark."
}
],
"excluded_campaigns": [
{
"id": "nexus-neutral-10x20-v1",
"status": "INVALIDATED",
"reason": "Uniform 96-token ceiling truncated model outputs; 36 Granite cases were consumed and excluded from scoring."
},
{
"id": "nexus-neutral-10x20-v2",
"status": "RUNNING",
"reason": "Larger neutral paired campaign remains unpublished until completion and audit."
}
],
"controls": {
"previous_test_answers_reused": false,
"sealed_test_used_for_training": false,
"automatic_learning_during_evaluation": false,
"raw_questions_or_targets_published": false,
"primary_and_diagnostic_results_pooled": false
}
}
|