{
  "schema": "deepfakepolicy.detector-benchmark-protocol.v1",
  "version": "1.0.0",
  "published_at": "2026-09-02",
  "status": "protocol_published_dataset_freeze_pending",
  "public_ranking_available": false,
  "purpose": "Evaluate operational deepfake detector performance without presenting deployment smoke tests as accuracy evidence.",
  "modalities": ["image", "video", "audio"],
  "truth_classes": [
    "authentic_capture",
    "fully_synthetic",
    "partial_identity_or_content_manipulation"
  ],
  "minimum_samples_per_primary_class_and_modality": 200,
  "quality_conditions": [
    "highest_quality_available_original",
    "platform_recompression",
    "resize_or_crop",
    "screen_capture_or_screen_recording"
  ],
  "required_metrics": [
    "balanced_accuracy",
    "macro_f1",
    "sensitivity",
    "specificity",
    "false_positive_rate",
    "false_negative_rate",
    "calibration_error",
    "abstention_coverage",
    "latency",
    "cost_per_asset"
  ],
  "reporting_requirements": [
    "dataset_license_and_provenance",
    "asset_hash_manifest",
    "documented_ground_truth",
    "generator_and_capture_family_separation",
    "detector_version_threshold_and_run_timestamp",
    "quality_condition_breakdowns",
    "bootstrap_confidence_intervals",
    "failed_and_abstained_request_counts",
    "reproducible_correction_log"
  ],
  "publication_gates": [
    "dataset_frozen_before_scoring",
    "licenses_and_consent_reviewed",
    "leakage_review_completed",
    "all_primary_strata_meet_minimum_sample_size",
    "all_detector_failures_retained_in_denominators",
    "results_reproduced_from_locked_inputs"
  ],
  "excluded_from_v1": [
    "text_authorship_detection",
    "real_time_live_stream_detection",
    "identity_attribution",
    "claims_that_a_low_score_certifies_authenticity"
  ],
  "contact": "deepfakepolicy@proton.me"
}