{
  "tool": "modelaudit",
  "tool_version": "0.2.40",
  "scanned_at": "2026-04-26T23:37:34Z",
  "files": {
    "model.onnx": {
      "size_mb": 131.1,
      "sha256": "5289497ae11bb612806e56eefa168c6338e3e5d9f548e8012f1e10560e19fa8f"
    }
  },
  "audit": {
    "bytes_scanned": 134726908,
    "issues": [
      {
        "message": "Layer 'model.text_projector.linear_1.weight' has 1 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "model.text_projector.linear_1.weight",
          "outlier_neurons": [
            43
          ],
          "total_outliers": 1,
          "outlier_percentage": 0.26041666666666663,
          "z_scores": [
            3.000617742538452
          ],
          "weight_norms": [
            0.5273075103759766
          ],
          "mean_norm": 0.5908768773078918,
          "std_norm": 0.02118542604148388,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.8968546,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'model.text_projector.linear_2.weight' has 2 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "model.text_projector.linear_2.weight",
          "outlier_neurons": [
            214,
            265
          ],
          "total_outliers": 2,
          "outlier_percentage": 0.5208333333333333,
          "z_scores": [
            3.4721202850341797,
            3.839141368865967
          ],
          "weight_norms": [
            0.6805276870727539,
            0.6899271607398987
          ],
          "mean_norm": 0.591606080532074,
          "std_norm": 0.025610174983739853,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.8978963,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2032' has 4 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2032",
          "outlier_neurons": [
            730,
            760,
            762,
            764
          ],
          "total_outliers": 4,
          "outlier_percentage": 0.3472222222222222,
          "z_scores": [
            3.254678249359131,
            3.366765260696411,
            3.440369129180908,
            3.077902317047119
          ],
          "weight_norms": [
            3.0971879959106445,
            3.1553280353546143,
            3.1935067176818848,
            3.005493402481079
          ],
          "mean_norm": 1.4089704751968384,
          "std_norm": 0.518704891204834,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.8985758,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2053' has 2 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2053",
          "outlier_neurons": [
            358,
            363
          ],
          "total_outliers": 2,
          "outlier_percentage": 0.5208333333333333,
          "z_scores": [
            6.91820764541626,
            10.892086029052734
          ],
          "weight_norms": [
            1.986020803451538,
            2.709052085876465
          ],
          "mean_norm": 0.7272806167602539,
          "std_norm": 0.18194599449634552,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.899232,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2055' has 3 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2055",
          "outlier_neurons": [
            77,
            358,
            363
          ],
          "total_outliers": 3,
          "outlier_percentage": 0.78125,
          "z_scores": [
            3.1322412490844727,
            5.322834491729736,
            6.233516693115234
          ],
          "weight_norms": [
            2.7848806381225586,
            3.6990485191345215,
            4.079090118408203
          ],
          "mean_norm": 1.4777485132217407,
          "std_norm": 0.417315274477005,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.899889,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2075' has 1 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2075",
          "outlier_neurons": [
            363
          ],
          "total_outliers": 1,
          "outlier_percentage": 0.26041666666666663,
          "z_scores": [
            7.383401870727539
          ],
          "weight_norms": [
            1.4844307899475098
          ],
          "mean_norm": 0.7127715945243835,
          "std_norm": 0.10451269149780273,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.900539,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2076' has 6 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2076",
          "outlier_neurons": [
            27,
            194,
            456,
            688,
            770,
            1095
          ],
          "total_outliers": 6,
          "outlier_percentage": 0.5208333333333333,
          "z_scores": [
            3.035195827484131,
            3.0513501167297363,
            3.05649995803833,
            3.1463961601257324,
            7.347891330718994,
            3.1944825649261475
          ],
          "weight_norms": [
            0.3287472128868103,
            0.3244526982307434,
            1.948188066482544,
            1.9720864295959473,
            3.089028835296631,
            1.9848699569702148
          ],
          "mean_norm": 1.1356358528137207,
          "std_norm": 0.2658440172672272,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.9011943,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2078' has 2 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2078",
          "outlier_neurons": [
            170,
            486
          ],
          "total_outliers": 2,
          "outlier_percentage": 0.1736111111111111,
          "z_scores": [
            3.227228879928589,
            3.220970392227173
          ],
          "weight_norms": [
            1.981728196144104,
            1.9800732135772705
          ],
          "mean_norm": 1.1282992362976074,
          "std_norm": 0.2644463777542114,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.9018571,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2098' has 4 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2098",
          "outlier_neurons": [
            67,
            490,
            573,
            643
          ],
          "total_outliers": 4,
          "outlier_percentage": 0.3472222222222222,
          "z_scores": [
            10.147518157958984,
            3.2569260597229004,
            3.1233837604522705,
            12.771944999694824
          ],
          "weight_norms": [
            4.698851108551025,
            2.2743470668792725,
            2.2273592948913574,
            5.622274398803711
          ],
          "mean_norm": 1.128374457359314,
          "std_norm": 0.35185712575912476,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.9025378,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2099' has 2 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2099",
          "outlier_neurons": [
            358,
            363
          ],
          "total_outliers": 2,
          "outlier_percentage": 0.5208333333333333,
          "z_scores": [
            7.9214091300964355,
            10.449893951416016
          ],
          "weight_norms": [
            2.7937510013580322,
            3.2416014671325684
          ],
          "mean_norm": 1.3906946182250977,
          "std_norm": 0.1771220713853836,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.9031963,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2119' has 1 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2119",
          "outlier_neurons": [
            363
          ],
          "total_outliers": 1,
          "outlier_percentage": 0.26041666666666663,
          "z_scores": [
            4.427882671356201
          ],
          "weight_norms": [
            1.6316750049591064
          ],
          "mean_norm": 0.7824981808662415,
          "std_norm": 0.19177943468093872,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.9038618,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2121' has 2 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2121",
          "outlier_neurons": [
            358,
            363
          ],
          "total_outliers": 2,
          "outlier_percentage": 0.5208333333333333,
          "z_scores": [
            6.758997440338135,
            13.026949882507324
          ],
          "weight_norms": [
            2.784144401550293,
            4.112109184265137
          ],
          "mean_norm": 1.3521441221237183,
          "std_norm": 0.21186578273773193,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.904512,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2122' has 8 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2122",
          "outlier_neurons": [
            9,
            266,
            276,
            279,
            284,
            285,
            306,
            554
          ],
          "total_outliers": 8,
          "outlier_percentage": 0.6944444444444444,
          "z_scores": [
            3.212179660797119,
            3.245704412460327,
            3.1307146549224854,
            3.316779851913452,
            3.2148783206939697,
            3.2838239669799805,
            3.1412272453308105,
            3.164783477783203
          ],
          "weight_norms": [
            1.7544312477111816,
            1.7613235712051392,
            1.7376829385757446,
            1.7759358882904053,
            1.754986047744751,
            1.769160509109497,
            1.7398442029953003,
            1.7446870803833008
          ],
          "mean_norm": 1.0940423011779785,
          "std_norm": 0.20558904111385345,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.9051602,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2143' has 2 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2143",
          "outlier_neurons": [
            358,
            363
          ],
          "total_outliers": 2,
          "outlier_percentage": 0.5208333333333333,
          "z_scores": [
            6.859812259674072,
            10.17564582824707
          ],
          "weight_norms": [
            2.7563867568969727,
            3.4448232650756836
          ],
          "mean_norm": 1.3321459293365479,
          "std_norm": 0.20762096345424652,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.9058108,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2144' has 6 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2144",
          "outlier_neurons": [
            67,
            138,
            140,
            554,
            556,
            614
          ],
          "total_outliers": 6,
          "outlier_percentage": 0.5208333333333333,
          "z_scores": [
            3.020602226257324,
            3.1754822731018066,
            3.826287031173706,
            3.2961931228637695,
            3.650977611541748,
            3.5217597484588623
          ],
          "weight_norms": [
            0.3887397348880768,
            1.7699042558670044,
            1.9149746894836426,
            1.796811819076538,
            1.8758965730667114,
            1.8470927476882935
          ],
          "mean_norm": 1.0620598793029785,
          "std_norm": 0.22290925681591034,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.9065673,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2163' has 1 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2163",
          "outlier_neurons": [
            317
          ],
          "total_outliers": 1,
          "outlier_percentage": 0.26041666666666663,
          "z_scores": [
            3.0252268314361572
          ],
          "weight_norms": [
            0.36917099356651306
          ],
          "mean_norm": 1.0253673791885376,
          "std_norm": 0.21690815687179565,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.9072561,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2165' has 2 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2165",
          "outlier_neurons": [
            358,
            363
          ],
          "total_outliers": 2,
          "outlier_percentage": 0.5208333333333333,
          "z_scores": [
            3.6371476650238037,
            7.814241886138916
          ],
          "weight_norms": [
            2.133096933364868,
            3.1184256076812744
          ],
          "mean_norm": 1.2751353979110718,
          "std_norm": 0.2358885556459427,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.907795,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2166' has 4 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2166",
          "outlier_neurons": [
            217,
            218,
            219,
            602
          ],
          "total_outliers": 4,
          "outlier_percentage": 0.3472222222222222,
          "z_scores": [
            3.0621910095214844,
            3.151977300643921,
            3.188354253768921,
            3.0198326110839844
          ],
          "weight_norms": [
            1.8419299125671387,
            1.868815302848816,
            1.879707932472229,
            1.829246163368225
          ],
          "mean_norm": 0.924994170665741,
          "std_norm": 0.2994377911090851,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.908296,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2185' has 2 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2185",
          "outlier_neurons": [
            141,
            317
          ],
          "total_outliers": 2,
          "outlier_percentage": 0.5208333333333333,
          "z_scores": [
            3.005260944366455,
            3.345065116882324
          ],
          "weight_norms": [
            0.428337424993515,
            0.36667728424072266
          ],
          "mean_norm": 0.9736661911010742,
          "std_norm": 0.1814580261707306,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.908806,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2186' has 5 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2186",
          "outlier_neurons": [
            252,
            338,
            828,
            914,
            946
          ],
          "total_outliers": 5,
          "outlier_percentage": 0.4340277777777778,
          "z_scores": [
            11.505152702331543,
            4.314464569091797,
            12.327495574951172,
            9.297687530517578,
            3.366377115249634
          ],
          "weight_norms": [
            4.306219577789307,
            2.2771012783050537,
            4.538273811340332,
            3.6833016872406006,
            2.009563446044922
          ],
          "mean_norm": 1.0596158504486084,
          "std_norm": 0.28218692541122437,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.909304,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2187' has 3 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2187",
          "outlier_neurons": [
            147,
            358,
            363
          ],
          "total_outliers": 3,
          "outlier_percentage": 0.78125,
          "z_scores": [
            3.2084083557128906,
            3.8869893550872803,
            9.003568649291992
          ],
          "weight_norms": [
            2.062852144241333,
            2.2326014041900635,
            3.5125303268432617
          ],
          "mean_norm": 1.2602583169937134,
          "std_norm": 0.2501532733440399,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.9098086,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2188' has 3 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2188",
          "outlier_neurons": [
            326,
            651,
            742
          ],
          "total_outliers": 3,
          "outlier_percentage": 0.26041666666666663,
          "z_scores": [
            3.1055855751037598,
            3.0717461109161377,
            3.0304436683654785
          ],
          "weight_norms": [
            1.6304196119308472,
            1.6237001419067383,
            1.6154987812042236
          ],
          "mean_norm": 1.0137479305267334,
          "std_norm": 0.19856856763362885,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.9103193,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2207' has 1 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2207",
          "outlier_neurons": [
            141
          ],
          "total_outliers": 1,
          "outlier_percentage": 0.26041666666666663,
          "z_scores": [
            3.137122869491577
          ],
          "weight_norms": [
            0.4656246602535248
          ],
          "mean_norm": 1.142965316772461,
          "std_norm": 0.215911403298378,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.9108434,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2208' has 5 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2208",
          "outlier_neurons": [
            147,
            489,
            723,
            731,
            1111
          ],
          "total_outliers": 5,
          "outlier_percentage": 0.4340277777777778,
          "z_scores": [
            3.3600692749023438,
            3.3963823318481445,
            8.018890380859375,
            3.6322925090789795,
            3.479583501815796
          ],
          "weight_norms": [
            0.2726581394672394,
            1.847255825996399,
            2.9245359897613525,
            1.9022349119186401,
            1.8666459321975708
          ],
          "mean_norm": 1.0557255744934082,
          "std_norm": 0.23305098712444305,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.9113624,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2209' has 2 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2209",
          "outlier_neurons": [
            358,
            363
          ],
          "total_outliers": 2,
          "outlier_percentage": 0.5208333333333333,
          "z_scores": [
            4.950748920440674,
            6.248614311218262
          ],
          "weight_norms": [
            2.431596517562866,
            2.7228353023529053
          ],
          "mean_norm": 1.3206568956375122,
          "std_norm": 0.22439830005168915,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.911881,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2230' has 7 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2230",
          "outlier_neurons": [
            252,
            299,
            715,
            793,
            798,
            843,
            853
          ],
          "total_outliers": 7,
          "outlier_percentage": 0.607638888888889,
          "z_scores": [
            3.6941657066345215,
            3.1229217052459717,
            3.9711530208587646,
            3.111565589904785,
            3.424854278564453,
            3.259798765182495,
            3.0462286472320557
          ],
          "weight_norms": [
            1.8310163021087646,
            1.7133229970932007,
            1.8880839347839355,
            1.7109832763671875,
            1.7755300998687744,
            1.7415237426757812,
            1.6975219249725342
          ],
          "mean_norm": 1.0699079036712646,
          "std_norm": 0.20602984726428986,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.9123945,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2231' has 1 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2231",
          "outlier_neurons": [
            363
          ],
          "total_outliers": 1,
          "outlier_percentage": 0.26041666666666663,
          "z_scores": [
            3.510470151901245
          ],
          "weight_norms": [
            2.2871758937835693
          ],
          "mean_norm": 1.3612122535705566,
          "std_norm": 0.26377198100090027,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.91292,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2252' has 7 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2252",
          "outlier_neurons": [
            316,
            661,
            736,
            815,
            839,
            942,
            1027
          ],
          "total_outliers": 7,
          "outlier_percentage": 0.607638888888889,
          "z_scores": [
            3.2526912689208984,
            3.417073965072632,
            3.306257486343384,
            3.0526297092437744,
            7.109450340270996,
            7.056085586547852,
            3.029418468475342
          ],
          "weight_norms": [
            1.8308838605880737,
            1.868377447128296,
            1.8431016206741333,
            1.785252332687378,
            2.710561990737915,
            2.698390245437622,
            1.7799581289291382
          ],
          "mean_norm": 1.0889859199523926,
          "std_norm": 0.2280874103307724,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.9134305,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2255' has 1 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2255",
          "outlier_neurons": [
            244
          ],
          "total_outliers": 1,
          "outlier_percentage": 0.26041666666666663,
          "z_scores": [
            3.056356430053711
          ],
          "weight_norms": [
            0.6583805680274963
          ],
          "mean_norm": 0.5920038223266602,
          "std_norm": 0.021717606112360954,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.913943,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2256' has 2 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2256",
          "outlier_neurons": [
            62,
            268
          ],
          "total_outliers": 2,
          "outlier_percentage": 0.5208333333333333,
          "z_scores": [
            3.048487663269043,
            3.0232150554656982
          ],
          "weight_norms": [
            0.7006304264068604,
            0.6999441981315613
          ],
          "mean_norm": 0.617854654788971,
          "std_norm": 0.02715306170284748,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.914449,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2257' has 2 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2257",
          "outlier_neurons": [
            206,
            250
          ],
          "total_outliers": 2,
          "outlier_percentage": 0.78125,
          "z_scores": [
            3.219963312149048,
            3.752326488494873
          ],
          "weight_norms": [
            0.9235143065452576,
            0.9379115700721741
          ],
          "mean_norm": 0.8364334106445312,
          "std_norm": 0.027044065296649933,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.9149594,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2258' has 1 output neurons with abnormal weight magnitudes",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2258",
          "outlier_neurons": [
            60
          ],
          "total_outliers": 1,
          "outlier_percentage": 0.78125,
          "z_scores": [
            3.0868079662323
          ],
          "weight_norms": [
            0.6551735401153564
          ],
          "mean_norm": 0.5127619504928589,
          "std_norm": 0.0461355522274971,
          "analysis_method": "structural_analysis",
          "architecture_confidence": 0.8999999999999999
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.9154654,
        "type": "onnx_check"
      },
      {
        "message": "Layer 'onnx::MatMul_2259' has neurons with extremely large weight values",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "layer": "onnx::MatMul_2259",
          "affected_neurons": [
            0
          ],
          "total_affected": 1,
          "num_extreme_weights": 1,
          "threshold": 0.08830882608890533,
          "max_weight": 0.09035700559616089,
          "total_outputs": 1,
          "analysis_method": "structural_analysis"
        },
        "why": "Weight values that are orders of magnitude larger than typical can cause numerical instability, overflow attacks, or may encode hidden data. Detection uses statistical analysis rather than name-based classification to avoid security bypasses.",
        "timestamp": 1777246651.9158247,
        "type": "onnx_check",
        "rule_code": "S802"
      }
    ],
    "checks": [
      {
        "name": "Path Exists",
        "status": "passed",
        "message": "Path exists",
        "location": "/opt/sas/model-gli-text/models/class-edge/README.md",
        "details": {
          "path": "/opt/sas/model-gli-text/models/class-edge/README.md"
        },
        "timestamp": 1777246575.039381
      },
      {
        "name": "Path Readable",
        "status": "passed",
        "message": "Path is readable",
        "location": "/opt/sas/model-gli-text/models/class-edge/README.md",
        "details": {
          "path": "/opt/sas/model-gli-text/models/class-edge/README.md"
        },
        "timestamp": 1777246575.0394099
      },
      {
        "name": "File Type Validation",
        "status": "passed",
        "message": "File type validation passed",
        "location": "/opt/sas/model-gli-text/models/class-edge/README.md",
        "details": {},
        "timestamp": 1777246575.0394428
      },
      {
        "name": "Path Exists",
        "status": "passed",
        "message": "Path exists",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "path": "/opt/sas/model-gli-text/models/class-edge/model.onnx"
        },
        "timestamp": 1777246575.0802407
      },
      {
        "name": "Path Readable",
        "status": "passed",
        "message": "Path is readable",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "path": "/opt/sas/model-gli-text/models/class-edge/model.onnx"
        },
        "timestamp": 1777246575.0802708
      },
      {
        "name": "File Type Validation",
        "status": "passed",
        "message": "File type validation passed",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {},
        "timestamp": 1777246575.0802965
      },
      {
        "name": "File Integrity Hash",
        "status": "passed",
        "message": "File integrity hashes calculated",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "md5": "20298350a5e7fe666dd69ced6481d792",
          "sha256": "5289497ae11bb612806e56eefa168c6338e3e5d9f548e8012f1e10560e19fa8f",
          "sha512": "709fd6baee90e7e7ab1b1cc502ffbbd2133d857211bdaacc74fc7fae76ba02e207af9e458c990873c59ac38fd9d8df090a4994a7e410d5a8d7228f994660d13c",
          "file_size": 131130181
        },
        "timestamp": 1777246575.7523339
      },
      {
        "name": "JIT/Script Code Execution Detection",
        "status": "passed",
        "message": "No JIT/Script code execution risks detected",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {},
        "timestamp": 1777246649.262821
      },
      {
        "name": "Network Communication Detection",
        "status": "passed",
        "message": "No network communication patterns detected",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {},
        "timestamp": 1777246649.2629182
      },
      {
        "name": "Custom Operator Domain Check",
        "status": "passed",
        "message": "All operators use standard ONNX domains",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "safe_nodes": 1694
        },
        "timestamp": 1777246649.2702723
      },
      {
        "name": "Python Operator Detection",
        "status": "passed",
        "message": "No Python operators detected",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "nodes_checked": 1694
        },
        "timestamp": 1777246649.2702963
      },
      {
        "name": "Tensor Size Validation",
        "status": "passed",
        "message": "Tensor Size Validation completed successfully",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx (tensor: model.text_projector.linear_1.weight)",
        "details": {
          "component_count": 76
        },
        "timestamp": 1777246649.4810627
      },
      {
        "name": "Weight Distribution Anomaly Detection",
        "status": "failed",
        "message": "Weight Distribution Anomaly Detection found 33 issues",
        "severity": "info",
        "location": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "details": {
          "component_count": 33,
          "findings": [
            {
              "layer": "model.text_projector.linear_1.weight",
              "outlier_neurons": [
                43
              ],
              "total_outliers": 1,
              "outlier_percentage": 0.26041666666666663,
              "z_scores": [
                3.000617742538452
              ],
              "weight_norms": [
                0.5273075103759766
              ],
              "mean_norm": 0.5908768773078918,
              "std_norm": 0.02118542604148388,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "model.text_projector.linear_2.weight",
              "outlier_neurons": [
                214,
                265
              ],
              "total_outliers": 2,
              "outlier_percentage": 0.5208333333333333,
              "z_scores": [
                3.4721202850341797,
                3.839141368865967
              ],
              "weight_norms": [
                0.6805276870727539,
                0.6899271607398987
              ],
              "mean_norm": 0.591606080532074,
              "std_norm": 0.025610174983739853,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2032",
              "outlier_neurons": [
                730,
                760,
                762,
                764
              ],
              "total_outliers": 4,
              "outlier_percentage": 0.3472222222222222,
              "z_scores": [
                3.254678249359131,
                3.366765260696411,
                3.440369129180908,
                3.077902317047119
              ],
              "weight_norms": [
                3.0971879959106445,
                3.1553280353546143,
                3.1935067176818848,
                3.005493402481079
              ],
              "mean_norm": 1.4089704751968384,
              "std_norm": 0.518704891204834,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2053",
              "outlier_neurons": [
                358,
                363
              ],
              "total_outliers": 2,
              "outlier_percentage": 0.5208333333333333,
              "z_scores": [
                6.91820764541626,
                10.892086029052734
              ],
              "weight_norms": [
                1.986020803451538,
                2.709052085876465
              ],
              "mean_norm": 0.7272806167602539,
              "std_norm": 0.18194599449634552,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2055",
              "outlier_neurons": [
                77,
                358,
                363
              ],
              "total_outliers": 3,
              "outlier_percentage": 0.78125,
              "z_scores": [
                3.1322412490844727,
                5.322834491729736,
                6.233516693115234
              ],
              "weight_norms": [
                2.7848806381225586,
                3.6990485191345215,
                4.079090118408203
              ],
              "mean_norm": 1.4777485132217407,
              "std_norm": 0.417315274477005,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2075",
              "outlier_neurons": [
                363
              ],
              "total_outliers": 1,
              "outlier_percentage": 0.26041666666666663,
              "z_scores": [
                7.383401870727539
              ],
              "weight_norms": [
                1.4844307899475098
              ],
              "mean_norm": 0.7127715945243835,
              "std_norm": 0.10451269149780273,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2076",
              "outlier_neurons": [
                27,
                194,
                456,
                688,
                770,
                1095
              ],
              "total_outliers": 6,
              "outlier_percentage": 0.5208333333333333,
              "z_scores": [
                3.035195827484131,
                3.0513501167297363,
                3.05649995803833,
                3.1463961601257324,
                7.347891330718994,
                3.1944825649261475
              ],
              "weight_norms": [
                0.3287472128868103,
                0.3244526982307434,
                1.948188066482544,
                1.9720864295959473,
                3.089028835296631,
                1.9848699569702148
              ],
              "mean_norm": 1.1356358528137207,
              "std_norm": 0.2658440172672272,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2078",
              "outlier_neurons": [
                170,
                486
              ],
              "total_outliers": 2,
              "outlier_percentage": 0.1736111111111111,
              "z_scores": [
                3.227228879928589,
                3.220970392227173
              ],
              "weight_norms": [
                1.981728196144104,
                1.9800732135772705
              ],
              "mean_norm": 1.1282992362976074,
              "std_norm": 0.2644463777542114,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2098",
              "outlier_neurons": [
                67,
                490,
                573,
                643
              ],
              "total_outliers": 4,
              "outlier_percentage": 0.3472222222222222,
              "z_scores": [
                10.147518157958984,
                3.2569260597229004,
                3.1233837604522705,
                12.771944999694824
              ],
              "weight_norms": [
                4.698851108551025,
                2.2743470668792725,
                2.2273592948913574,
                5.622274398803711
              ],
              "mean_norm": 1.128374457359314,
              "std_norm": 0.35185712575912476,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2099",
              "outlier_neurons": [
                358,
                363
              ],
              "total_outliers": 2,
              "outlier_percentage": 0.5208333333333333,
              "z_scores": [
                7.9214091300964355,
                10.449893951416016
              ],
              "weight_norms": [
                2.7937510013580322,
                3.2416014671325684
              ],
              "mean_norm": 1.3906946182250977,
              "std_norm": 0.1771220713853836,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2119",
              "outlier_neurons": [
                363
              ],
              "total_outliers": 1,
              "outlier_percentage": 0.26041666666666663,
              "z_scores": [
                4.427882671356201
              ],
              "weight_norms": [
                1.6316750049591064
              ],
              "mean_norm": 0.7824981808662415,
              "std_norm": 0.19177943468093872,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2121",
              "outlier_neurons": [
                358,
                363
              ],
              "total_outliers": 2,
              "outlier_percentage": 0.5208333333333333,
              "z_scores": [
                6.758997440338135,
                13.026949882507324
              ],
              "weight_norms": [
                2.784144401550293,
                4.112109184265137
              ],
              "mean_norm": 1.3521441221237183,
              "std_norm": 0.21186578273773193,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2122",
              "outlier_neurons": [
                9,
                266,
                276,
                279,
                284,
                285,
                306,
                554
              ],
              "total_outliers": 8,
              "outlier_percentage": 0.6944444444444444,
              "z_scores": [
                3.212179660797119,
                3.245704412460327,
                3.1307146549224854,
                3.316779851913452,
                3.2148783206939697,
                3.2838239669799805,
                3.1412272453308105,
                3.164783477783203
              ],
              "weight_norms": [
                1.7544312477111816,
                1.7613235712051392,
                1.7376829385757446,
                1.7759358882904053,
                1.754986047744751,
                1.769160509109497,
                1.7398442029953003,
                1.7446870803833008
              ],
              "mean_norm": 1.0940423011779785,
              "std_norm": 0.20558904111385345,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2143",
              "outlier_neurons": [
                358,
                363
              ],
              "total_outliers": 2,
              "outlier_percentage": 0.5208333333333333,
              "z_scores": [
                6.859812259674072,
                10.17564582824707
              ],
              "weight_norms": [
                2.7563867568969727,
                3.4448232650756836
              ],
              "mean_norm": 1.3321459293365479,
              "std_norm": 0.20762096345424652,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2144",
              "outlier_neurons": [
                67,
                138,
                140,
                554,
                556,
                614
              ],
              "total_outliers": 6,
              "outlier_percentage": 0.5208333333333333,
              "z_scores": [
                3.020602226257324,
                3.1754822731018066,
                3.826287031173706,
                3.2961931228637695,
                3.650977611541748,
                3.5217597484588623
              ],
              "weight_norms": [
                0.3887397348880768,
                1.7699042558670044,
                1.9149746894836426,
                1.796811819076538,
                1.8758965730667114,
                1.8470927476882935
              ],
              "mean_norm": 1.0620598793029785,
              "std_norm": 0.22290925681591034,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2163",
              "outlier_neurons": [
                317
              ],
              "total_outliers": 1,
              "outlier_percentage": 0.26041666666666663,
              "z_scores": [
                3.0252268314361572
              ],
              "weight_norms": [
                0.36917099356651306
              ],
              "mean_norm": 1.0253673791885376,
              "std_norm": 0.21690815687179565,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2165",
              "outlier_neurons": [
                358,
                363
              ],
              "total_outliers": 2,
              "outlier_percentage": 0.5208333333333333,
              "z_scores": [
                3.6371476650238037,
                7.814241886138916
              ],
              "weight_norms": [
                2.133096933364868,
                3.1184256076812744
              ],
              "mean_norm": 1.2751353979110718,
              "std_norm": 0.2358885556459427,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2166",
              "outlier_neurons": [
                217,
                218,
                219,
                602
              ],
              "total_outliers": 4,
              "outlier_percentage": 0.3472222222222222,
              "z_scores": [
                3.0621910095214844,
                3.151977300643921,
                3.188354253768921,
                3.0198326110839844
              ],
              "weight_norms": [
                1.8419299125671387,
                1.868815302848816,
                1.879707932472229,
                1.829246163368225
              ],
              "mean_norm": 0.924994170665741,
              "std_norm": 0.2994377911090851,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2185",
              "outlier_neurons": [
                141,
                317
              ],
              "total_outliers": 2,
              "outlier_percentage": 0.5208333333333333,
              "z_scores": [
                3.005260944366455,
                3.345065116882324
              ],
              "weight_norms": [
                0.428337424993515,
                0.36667728424072266
              ],
              "mean_norm": 0.9736661911010742,
              "std_norm": 0.1814580261707306,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2186",
              "outlier_neurons": [
                252,
                338,
                828,
                914,
                946
              ],
              "total_outliers": 5,
              "outlier_percentage": 0.4340277777777778,
              "z_scores": [
                11.505152702331543,
                4.314464569091797,
                12.327495574951172,
                9.297687530517578,
                3.366377115249634
              ],
              "weight_norms": [
                4.306219577789307,
                2.2771012783050537,
                4.538273811340332,
                3.6833016872406006,
                2.009563446044922
              ],
              "mean_norm": 1.0596158504486084,
              "std_norm": 0.28218692541122437,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2187",
              "outlier_neurons": [
                147,
                358,
                363
              ],
              "total_outliers": 3,
              "outlier_percentage": 0.78125,
              "z_scores": [
                3.2084083557128906,
                3.8869893550872803,
                9.003568649291992
              ],
              "weight_norms": [
                2.062852144241333,
                2.2326014041900635,
                3.5125303268432617
              ],
              "mean_norm": 1.2602583169937134,
              "std_norm": 0.2501532733440399,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2188",
              "outlier_neurons": [
                326,
                651,
                742
              ],
              "total_outliers": 3,
              "outlier_percentage": 0.26041666666666663,
              "z_scores": [
                3.1055855751037598,
                3.0717461109161377,
                3.0304436683654785
              ],
              "weight_norms": [
                1.6304196119308472,
                1.6237001419067383,
                1.6154987812042236
              ],
              "mean_norm": 1.0137479305267334,
              "std_norm": 0.19856856763362885,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2207",
              "outlier_neurons": [
                141
              ],
              "total_outliers": 1,
              "outlier_percentage": 0.26041666666666663,
              "z_scores": [
                3.137122869491577
              ],
              "weight_norms": [
                0.4656246602535248
              ],
              "mean_norm": 1.142965316772461,
              "std_norm": 0.215911403298378,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2208",
              "outlier_neurons": [
                147,
                489,
                723,
                731,
                1111
              ],
              "total_outliers": 5,
              "outlier_percentage": 0.4340277777777778,
              "z_scores": [
                3.3600692749023438,
                3.3963823318481445,
                8.018890380859375,
                3.6322925090789795,
                3.479583501815796
              ],
              "weight_norms": [
                0.2726581394672394,
                1.847255825996399,
                2.9245359897613525,
                1.9022349119186401,
                1.8666459321975708
              ],
              "mean_norm": 1.0557255744934082,
              "std_norm": 0.23305098712444305,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2209",
              "outlier_neurons": [
                358,
                363
              ],
              "total_outliers": 2,
              "outlier_percentage": 0.5208333333333333,
              "z_scores": [
                4.950748920440674,
                6.248614311218262
              ],
              "weight_norms": [
                2.431596517562866,
                2.7228353023529053
              ],
              "mean_norm": 1.3206568956375122,
              "std_norm": 0.22439830005168915,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2230",
              "outlier_neurons": [
                252,
                299,
                715,
                793,
                798,
                843,
                853
              ],
              "total_outliers": 7,
              "outlier_percentage": 0.607638888888889,
              "z_scores": [
                3.6941657066345215,
                3.1229217052459717,
                3.9711530208587646,
                3.111565589904785,
                3.424854278564453,
                3.259798765182495,
                3.0462286472320557
              ],
              "weight_norms": [
                1.8310163021087646,
                1.7133229970932007,
                1.8880839347839355,
                1.7109832763671875,
                1.7755300998687744,
                1.7415237426757812,
                1.6975219249725342
              ],
              "mean_norm": 1.0699079036712646,
              "std_norm": 0.20602984726428986,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2231",
              "outlier_neurons": [
                363
              ],
              "total_outliers": 1,
              "outlier_percentage": 0.26041666666666663,
              "z_scores": [
                3.510470151901245
              ],
              "weight_norms": [
                2.2871758937835693
              ],
              "mean_norm": 1.3612122535705566,
              "std_norm": 0.26377198100090027,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2252",
              "outlier_neurons": [
                316,
                661,
                736,
                815,
                839,
                942,
                1027
              ],
              "total_outliers": 7,
              "outlier_percentage": 0.607638888888889,
              "z_scores": [
                3.2526912689208984,
                3.417073965072632,
                3.306257486343384,
                3.0526297092437744,
                7.109450340270996,
                7.056085586547852,
                3.029418468475342
              ],
              "weight_norms": [
                1.8308838605880737,
                1.868377447128296,
                1.8431016206741333,
                1.785252332687378,
                2.710561990737915,
                2.698390245437622,
                1.7799581289291382
              ],
              "mean_norm": 1.0889859199523926,
              "std_norm": 0.2280874103307724,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2255",
              "outlier_neurons": [
                244
              ],
              "total_outliers": 1,
              "outlier_percentage": 0.26041666666666663,
              "z_scores": [
                3.056356430053711
              ],
              "weight_norms": [
                0.6583805680274963
              ],
              "mean_norm": 0.5920038223266602,
              "std_norm": 0.021717606112360954,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2256",
              "outlier_neurons": [
                62,
                268
              ],
              "total_outliers": 2,
              "outlier_percentage": 0.5208333333333333,
              "z_scores": [
                3.048487663269043,
                3.0232150554656982
              ],
              "weight_norms": [
                0.7006304264068604,
                0.6999441981315613
              ],
              "mean_norm": 0.617854654788971,
              "std_norm": 0.02715306170284748,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2257",
              "outlier_neurons": [
                206,
                250
              ],
              "total_outliers": 2,
              "outlier_percentage": 0.78125,
              "z_scores": [
                3.219963312149048,
                3.752326488494873
              ],
              "weight_norms": [
                0.9235143065452576,
                0.9379115700721741
              ],
              "mean_norm": 0.8364334106445312,
              "std_norm": 0.027044065296649933,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2258",
              "outlier_neurons": [
                60
              ],
              "total_outliers": 1,
              "outlier_percentage": 0.78125,
              "z_scores": [
                3.0868079662323
              ],
              "weight_norms": [
                0.6551735401153564
              ],
              "mean_norm": 0.5127619504928589,
              "std_norm": 0.0461355522274971,
              "analysis_method": "structural_analysis",
              "architecture_confidence": 0.8999999999999999
            },
            {
              "layer": "onnx::MatMul_2259",
              "affected_neurons": [
                0
              ],
              "total_affected": 1,
              "num_extreme_weights": 1,
              "threshold": 0.08830882608890533,
              "max_weight": 0.09035700559616089,
              "total_outputs": 1,
              "analysis_method": "structural_analysis"
            }
          ]
        },
        "why": "Neurons with weight magnitudes significantly different from others in the same layer may indicate tampering, backdoors, or training anomalies. These outliers are flagged when their statistical z-score exceeds the threshold. Thresholds are adjusted based on structural analysis of the model architecture.",
        "timestamp": 1777246651.9158192
      },
      {
        "name": "Path Exists",
        "status": "passed",
        "message": "Path exists",
        "location": "/opt/sas/model-gli-text/models/class-edge/tokenizer.json",
        "details": {
          "path": "/opt/sas/model-gli-text/models/class-edge/tokenizer.json"
        },
        "timestamp": 1777246652.934864
      },
      {
        "name": "Path Readable",
        "status": "passed",
        "message": "Path is readable",
        "location": "/opt/sas/model-gli-text/models/class-edge/tokenizer.json",
        "details": {
          "path": "/opt/sas/model-gli-text/models/class-edge/tokenizer.json"
        },
        "timestamp": 1777246652.934904
      },
      {
        "name": "File Type Validation",
        "status": "passed",
        "message": "File type validation passed",
        "location": "/opt/sas/model-gli-text/models/class-edge/tokenizer.json",
        "details": {},
        "timestamp": 1777246652.9349384
      },
      {
        "name": "Template Extraction",
        "status": "passed",
        "message": "No Jinja2 templates found in file",
        "location": "/opt/sas/model-gli-text/models/class-edge/tokenizer.json",
        "details": {
          "file_type": "tokenizer_config"
        },
        "timestamp": 1777246653.1820912
      }
    ],
    "files_scanned": 4,
    "assets": [
      {
        "path": "/opt/sas/model-gli-text/models/class-edge/LICENSE",
        "type": "unknown"
      },
      {
        "path": "/opt/sas/model-gli-text/models/class-edge/README.md",
        "type": "metadata",
        "size": 1505
      },
      {
        "path": "/opt/sas/model-gli-text/models/class-edge/model.onnx",
        "type": "onnx",
        "size": 131130181
      },
      {
        "path": "/opt/sas/model-gli-text/models/class-edge/tokenizer.json",
        "type": "jinja2_template",
        "size": 3583864
      }
    ],
    "has_errors": false,
    "scanner_names": [
      "metadata",
      "onnx",
      "jinja2_template"
    ],
    "file_metadata": {
      "/opt/sas/model-gli-text/models/class-edge/LICENSE": {
        "license_info": [
          {
            "spdx_id": "Apache-2.0",
            "name": "Apache License 2.0",
            "confidence": 0.8,
            "source": "file_header",
            "commercial_allowed": true
          }
        ],
        "copyright_notices": [],
        "license_files_nearby": [
          "/opt/sas/model-gli-text/models/class-edge/LICENSE"
        ],
        "is_dataset": false,
        "is_model": false,
        "risk_score": 0.0,
        "scan_timestamp": 1777246575.0347283,
        "content_hash": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30"
      },
      "/opt/sas/model-gli-text/models/class-edge/README.md": {
        "file_size": 1505,
        "license_info": [
          {
            "spdx_id": "Apache-2.0",
            "name": "Apache License 2.0",
            "confidence": 0.8,
            "source": "file_header",
            "commercial_allowed": true
          }
        ],
        "copyright_notices": [],
        "license_files_nearby": [
          "/opt/sas/model-gli-text/models/class-edge/LICENSE"
        ],
        "is_dataset": false,
        "is_model": false,
        "risk_score": 0.0,
        "scan_timestamp": 1777246575.0542111,
        "content_hash": "9cdf7eecd3159ae7da6debef8e872a6dd3c77624ef99b94c213979fe7175bcff"
      },
      "/opt/sas/model-gli-text/models/class-edge/model.onnx": {
        "file_size": 131130181,
        "file_hashes": {
          "md5": "20298350a5e7fe666dd69ced6481d792",
          "sha256": "5289497ae11bb612806e56eefa168c6338e3e5d9f548e8012f1e10560e19fa8f",
          "sha512": "709fd6baee90e7e7ab1b1cc502ffbbd2133d857211bdaacc74fc7fae76ba02e207af9e458c990873c59ac38fd9d8df090a4994a7e410d5a8d7228f994660d13c"
        },
        "license_info": [],
        "copyright_notices": [],
        "license_files_nearby": [
          "/opt/sas/model-gli-text/models/class-edge/LICENSE"
        ],
        "is_dataset": false,
        "is_model": true,
        "risk_score": 0.0,
        "scan_timestamp": 1777246652.0719705,
        "ir_version": 7,
        "producer_name": "pytorch",
        "node_count": 1694,
        "layers_analyzed": 48,
        "anomalies_found": 33,
        "content_hash": "5289497ae11bb612806e56eefa168c6338e3e5d9f548e8012f1e10560e19fa8f"
      },
      "/opt/sas/model-gli-text/models/class-edge/tokenizer.json": {
        "file_size": 3583864,
        "ml_context": {
          "frameworks": {},
          "overall_confidence": 0.0,
          "is_ml_content": false,
          "detected_patterns": [],
          "optimization_hints": [],
          "file_type": "tokenizer_config",
          "is_tokenizer": true,
          "confidence": 2
        },
        "license_info": [],
        "copyright_notices": [],
        "license_files_nearby": [
          "/opt/sas/model-gli-text/models/class-edge/LICENSE"
        ],
        "is_dataset": true,
        "is_model": false,
        "risk_score": 0.0,
        "scan_timestamp": 1777246653.1862533,
        "content_hash": "b6e09abceac75851eae141e7d51224a9da835c0c03684282fe2d5c42c745754f"
      }
    },
    "content_hash": "c8e2d48cb9deffda2dc9d8958861a2dd2b3519601536517aca66248c99e80602",
    "start_time": 1777246574.8321774,
    "duration": 78.36124658584595,
    "total_checks": 16,
    "passed_checks": 16,
    "failed_checks": 0,
    "success": true
  }
}