File size: 4,515 Bytes
4fa3363
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
{
  "training_scope": "all_data",
  "seed": 42,
  "input_mode": "package32k",
  "max_length": 32768,
  "hypothesis": "This AI agent skill is malicious.",
  "dataset": {
    "repo_id": "SkillScreen/SkillScreen-Data",
    "revision": "d59aabbd7e0e8d95b464a2d335f36e948577b832"
  },
  "native_score": "entailment softmax",
  "thresholds": {
    "mode": "package32k",
    "seed": 42,
    "records": 11831,
    "malicious": 6436,
    "benign": 5395,
    "auprc_average_precision": 0.9990312216598736,
    "auroc": 0.9984704751021387,
    "folds_sha256": "c11c6f68dd23fddb70d7090512d93f5be54bb216596175d44fcaa7e7df80e202",
    "oof_sha256": "f46a75a97b3716c787bd7116fb837ee0124be34055c1dbce75e6f6d1c27933f9",
    "folds": [
      {
        "fold": 1,
        "predictions_sha256": "e9e07bbeca1b9550d997de45dad000527ee6e212cf140247b7ab5885691aa6eb",
        "adapter_sha256": "1407f6319076816ea490e8e11b139c01ec3ff80947bd8fa962e16a57488eb85f"
      },
      {
        "fold": 2,
        "predictions_sha256": "94ba50e421d4385b7c0fd42cf6f1da076df7f81ec4b3e2a2872a21e015afb2b0",
        "adapter_sha256": "a01b9e8ab0623a54a7ae845137fb4fa0c877c9ddda6c57ac00d1600e0caafb1c"
      },
      {
        "fold": 3,
        "predictions_sha256": "f007afc29a1b5aa7407287501fc52619bb56fd1340197874aae5b5faccb50e14",
        "adapter_sha256": "e2261670e1f652da8bc316b3449c8ee03b0f5d993b7b958ddeb7b89a638bf116"
      }
    ],
    "created_at": "2026-11-10T07:34:36.103352+00:00",
    "default": {
      "raw_threshold": 0.0,
      "native_threshold": 0.5,
      "tp": 6367,
      "fp": 18,
      "fn": 69,
      "tn": 5377,
      "recall": 0.9892790553138595,
      "fpr": 0.003336422613531047,
      "precision": 0.9971808927173061,
      "f1": 0.9932142578582014,
      "flagged_fraction": 0.5396838813287127
    },
    "targets": {
      "0.95": {
        "raw_threshold": 7.327948239240634,
        "native_threshold": 0.9993435112125237,
        "tp": 6115,
        "fp": 1,
        "fn": 321,
        "tn": 5394,
        "recall": 0.9501243008079553,
        "fpr": 0.00018535681186283596,
        "precision": 0.999836494440811,
        "f1": 0.974346717654557,
        "flagged_fraction": 0.5169470036345195
      },
      "0.98": {
        "raw_threshold": 5.646373107267836,
        "native_threshold": 0.9964821215635228,
        "tp": 6308,
        "fp": 4,
        "fn": 128,
        "tn": 5391,
        "recall": 0.9801118707271598,
        "fpr": 0.0007414272474513438,
        "precision": 0.9993662864385298,
        "f1": 0.989645434577973,
        "flagged_fraction": 0.5335136505789874
      },
      "0.99": {
        "raw_threshold": -0.9062583981072989,
        "native_threshold": 0.2877660945090445,
        "tp": 6372,
        "fp": 19,
        "fn": 64,
        "tn": 5376,
        "recall": 0.9900559353635798,
        "fpr": 0.0035217794253938834,
        "precision": 0.997027069316226,
        "f1": 0.9935292741872612,
        "flagged_fraction": 0.5401910235821148
      }
    },
    "balanced": {
      "target": 0.99,
      "raw_threshold": -0.9062583981072989,
      "native_threshold": 0.2877660945090445,
      "tp": 6372,
      "fp": 19,
      "fn": 64,
      "tn": 5376,
      "recall": 0.9900559353635798,
      "fpr": 0.0035217794253938834,
      "precision": 0.997027069316226,
      "f1": 0.9935292741872612,
      "flagged_fraction": 0.5401910235821148
    },
    "high_recall": {
      "raw_threshold": -0.9062583981072989,
      "native_threshold": 0.2877660945090445,
      "tp": 6372,
      "fp": 19,
      "fn": 64,
      "tn": 5376,
      "recall": 0.9900559353635798,
      "fpr": 0.0035217794253938834,
      "precision": 0.997027069316226,
      "f1": 0.9935292741872612,
      "flagged_fraction": 0.5401910235821148
    },
    "selection": "Balanced maximizes recall minus FPR among predeclared 95/98/99% OOF recall gates; ties prefer higher recall. High-recall targets 99% OOF recall.",
    "scope": "Pooled three-fold out-of-fold scores. Every score is from a model that excluded that skill group. Thresholds transfer to a separate full-data refit; these are not held-out test results of that refit."
  },
  "metric_scope": "OOF metrics belong to the fold models, not an independent test of this final refit.",
  "repo_id": "SkillScreen/SkillScreen-2B-Package",
  "release": "v0.1.0",
  "default_policy": "high-recall",
  "base_model": {
    "repo_id": "AlexWortega/openjev",
    "subfolder": "qwen3.5-2b-nli-v5",
    "revision": "a20448012c213128955ca0c693e7c943865cab77"
  }
}