benchmarks.json

{
	"schema_version": "1.0",
	"dataset": "OPP-115 consolidation-0.75 held-out validation/test split",
	"metric_source": "Stored Phase 3 published metrics bundled with the Space",
	"generated_from": [
		"deployment/demo-assets/phase-3-freeze/classifier_metrics.json",
		"deployment/demo-assets/phase-3-freeze/phase3_manifest.json"
	],
	"notes": [
		"The published Hugging Face model is the current PrERT-CNM v4 PrivacyBERT sequence classifier.",
		"A separate upstream/base PrivacyBERT held-out benchmark artifact is not bundled, so that reference row is shown without fabricated metrics.",
		"Rows are ranked only when stored test macro F1 and test accuracy are available."
	],
	"models": [
		{
			"model_id": "JayNightmare/PrERT-CNM",
			"display_name": "PrERT-CNM v4 PrivacyBERT",
			"model_family": "PrivacyBERT fine-tuned classifier",
			"status": "available",
			"role": "published_huggingface_model",
			"notes": "Current Hugging Face model prepared from the Phase 3 PrivacyBERT freeze.",
			"training_rows": 15645,
			"validation_rows": 1803,
			"test_rows": 2272,
			"metrics": {
				"validation_macro_f1": 0.849021,
				"validation_accuracy": 0.948419,
				"test_macro_f1": 0.894166,
				"test_accuracy": 0.957306,
				"bayesian_primary_score": 0.987538,
				"calibration_test_ece": null,
				"test_per_class": {
					"organization": {
						"precision": 0.982039,
						"recall": 0.968229,
						"f1": 0.975085,
						"support": 1920
					},
					"system": {
						"precision": 0.813333,
						"recall": 0.859155,
						"f1": 0.835616,
						"support": 71
					},
					"user": {
						"precision": 0.838816,
						"recall": 0.907473,
						"f1": 0.871795,
						"support": 281
					}
				}
			}
		},
		{
			"model_id": "mukund/privbert",
			"display_name": "PrivacyBERT upstream reference",
			"model_family": "PrivacyBERT base model",
			"status": "reference_only",
			"role": "external_reference",
			"notes": "No separate held-out OPP-115 benchmark artifact is bundled for the upstream/base checkpoint.",
			"metrics": {}
		}
	]
}