q-trust-codebert / reports /benchmark_comparison.json
KRISHNAPURI's picture
add reports/benchmark_comparison.json
a36c9b3 verified
Raw History Blame Contribute Delete
4.06 kB
{
"seed": 42,
"comparisons": [
{
"model": "discovery/CryptoCodeDetector",
"metric": "F1 (held-out real code)",
"n": 2415,
"qtrust": {
"baseline": "qtrust (rules+AST+ML ensemble)",
"precision": 0.952,
"recall": 0.953,
"f1": 0.9525,
"accuracy": 0.9271
},
"baselines": [
{
"baseline": "rules-only (static layer)",
"precision": 0.9794,
"recall": 0.5127,
"f1": 0.673,
"accuracy": 0.6178
},
{
"baseline": "majority (always crypto)",
"precision": 0.7673,
"recall": 1.0,
"f1": 0.8683,
"accuracy": 0.7673
},
{
"baseline": "random coin-flip",
"precision": 0.7686,
"recall": 0.4895,
"f1": 0.5981,
"accuracy": 0.4952
}
],
"best_baseline": "majority (always crypto)",
"relative_gain": 0.097
},
{
"model": "discovery/CryptoCodeDetector",
"metric": "F1 on adversarial holdout (132 \u00a729 cases)",
"n": 132,
"qtrust": {
"baseline": "qtrust (rules+AST+ML ensemble)",
"precision": 0.8,
"recall": 0.4,
"f1": 0.5333
},
"baselines": [
{
"baseline": "rules-only (static layer)",
"precision": 0.7778,
"recall": 0.35,
"f1": 0.4828
}
],
"best_baseline": "rules-only (static layer)",
"relative_gain": 0.1046
},
{
"model": "discovery/AlgorithmPurposeClassifier",
"metric": "accuracy (held-out real triples)",
"n": 294,
"qtrust": {
"baseline": "qtrust (context-aware)",
"accuracy": 1.0
},
"baselines": [
{
"baseline": "family-prior-only (no context)",
"accuracy": 1.0
},
{
"baseline": "majority purpose",
"accuracy": 0.483
}
],
"best_baseline": "family-prior-only (no context)",
"relative_gain": 0.0
},
{
"model": "migration/PQCRecommender",
"metric": "family-correct rate (real triples)",
"n": 294,
"qtrust": {
"baseline": "qtrust (purpose-aware catalog)",
"accuracy": 0.7007
},
"baselines": [
{
"baseline": "always-ML-KEM-768",
"accuracy": 0.0782
}
],
"best_baseline": "always-ML-KEM-768",
"relative_gain": 7.9565
},
{
"model": "risk/QuantumExposureModel",
"metric": "MAE (held-out real TLS hosts)",
"n_train_hosts": 221,
"n_eval": 56,
"qtrust": {
"baseline": "qtrust (calibrated, host-held-out)",
"mae": 0.0
},
"baselines": [
{
"baseline": "constant-mean predictor",
"mae": 7.4412
}
],
"best_baseline": "constant-mean predictor",
"relative_gain": 1.0
},
{
"model": "monitoring/CryptoAnomalyDetector",
"metric": "false-positive rate (held-out real hosts)",
"n": 56,
"qtrust": {
"baseline": "qtrust (baseline+zscore)",
"fp_rate": 0.0
},
"baselines": [
{
"baseline": "always-alert",
"fp_rate": 1.0
}
],
"best_baseline": "always-alert",
"relative_gain": 1.0
},
{
"model": "vendor/SupplyChainRiskModel",
"metric": "ranking agreement (real NVD vendors)",
"n": 16,
"qtrust": {
"baseline": "qtrust (5-layer propagation)",
"kendall_tau": 0.7176,
"ndcg@5": 0.9722
},
"baselines": [
{
"baseline": "random ordering",
"kendall_tau": -0.0695
},
{
"baseline": "uniform scores",
"kendall_tau": 0.0
}
],
"best_baseline": "uniform scores",
"relative_gain": 0.7176
}
],
"summary": {
"comparisons_run": 7,
"models_beat_best_baseline": 6,
"mean_relative_gain": 1.5537
}
}