120 lines
No EOL
2.5 KiB
JSON
120 lines
No EOL
2.5 KiB
JSON
{
|
|
"floor_proxy_a0_vs_a0": {
|
|
"mean_a0_spread": 0.1145,
|
|
"max_a0_spread": 0.406,
|
|
"n_traps": 20
|
|
},
|
|
"modes": {
|
|
"q3a-src": {
|
|
"primary_paired_mean": -0.0047,
|
|
"primary_paired_CI90": [
|
|
-0.1167,
|
|
0.1073
|
|
],
|
|
"primary_n": 20,
|
|
"primary_paired_p": 0.943,
|
|
"primary_MDE90": 0.1813,
|
|
"primary_marginal_armacc": -0.0048,
|
|
"tana_only_paired_mean": 0.042,
|
|
"tana_only_CI90": [
|
|
-0.0311,
|
|
0.1151
|
|
],
|
|
"tana_n": 18,
|
|
"loo_mean_diff": {
|
|
"without_grok-4.3": 0.0183,
|
|
"without_mistral-large-latest": -0.0125
|
|
},
|
|
"primary_catastrophes_arm_vs_a0": [
|
|
66,
|
|
73
|
|
]
|
|
},
|
|
"q3a-draft": {
|
|
"primary_paired_mean": -0.0085,
|
|
"primary_paired_CI90": [
|
|
-0.1439,
|
|
0.127
|
|
],
|
|
"primary_n": 20,
|
|
"primary_paired_p": 0.915,
|
|
"primary_MDE90": 0.2194,
|
|
"primary_marginal_armacc": -0.0084,
|
|
"tana_only_paired_mean": 0.0184,
|
|
"tana_only_CI90": [
|
|
-0.1002,
|
|
0.137
|
|
],
|
|
"tana_n": 18,
|
|
"loo_mean_diff": {
|
|
"without_grok-4.3": 0.0559,
|
|
"without_mistral-large-latest": -0.0159
|
|
},
|
|
"primary_catastrophes_arm_vs_a0": [
|
|
51,
|
|
58
|
|
]
|
|
},
|
|
"q3a-final": {
|
|
"primary_paired_mean": -0.0548,
|
|
"primary_paired_CI90": [
|
|
-0.1946,
|
|
0.0849
|
|
],
|
|
"primary_n": 20,
|
|
"primary_paired_p": 0.505,
|
|
"primary_MDE90": 0.2262,
|
|
"primary_marginal_armacc": -0.0549,
|
|
"tana_only_paired_mean": -0.0286,
|
|
"tana_only_CI90": [
|
|
-0.149,
|
|
0.0919
|
|
],
|
|
"tana_n": 18,
|
|
"loo_mean_diff": {
|
|
"without_grok-4.3": -0.0609,
|
|
"without_mistral-large-latest": -0.0375
|
|
},
|
|
"primary_catastrophes_arm_vs_a0": [
|
|
50,
|
|
49
|
|
]
|
|
}
|
|
},
|
|
"pooled_fix_rate": {
|
|
"A0_failed_traps": [
|
|
"T-ana-17",
|
|
"T-ana-2",
|
|
"T-ana-32",
|
|
"T-ana-36",
|
|
"T-ana-42",
|
|
"T-ell-40"
|
|
],
|
|
"n_failed": 6,
|
|
"by_mode": {
|
|
"q3a-src": "2/6 (33%)",
|
|
"q3a-draft": "2/6 (33%)",
|
|
"q3a-final": "2/6 (33%)"
|
|
}
|
|
},
|
|
"T-ell-43_disclosure": {
|
|
"q3a-src": {
|
|
"a0_acc": 1.0,
|
|
"arm_acc": 0.0,
|
|
"cat_arm": 11,
|
|
"cat_a0": 0
|
|
},
|
|
"q3a-draft": {
|
|
"a0_acc": 1.0,
|
|
"arm_acc": 0.125,
|
|
"cat_arm": 8,
|
|
"cat_a0": 0
|
|
},
|
|
"q3a-final": {
|
|
"a0_acc": 1.0,
|
|
"arm_acc": 0.042,
|
|
"cat_arm": 9,
|
|
"cat_a0": 0
|
|
}
|
|
}
|
|
} |