●Evaluation methodology

AI Detector Evaluation Framework

A transparent framework for reporting performance by modality, dataset, transformation, model version, and error type.

AUROC
Score discrimination
FPR
False positive rate
FNR
False negative rate
p50/p95
Latency distribution

Current public results status

ZeroTrue does not publish one universal accuracy number on this page. Versioned results should be reported by modality with dataset details, thresholds, false positives, false negatives, and known limitations.

Methodology

The minimum information required for a useful detector benchmark

Datasets

  • β€’ Named public datasets
  • β€’ Representative held-out samples
  • β€’ Compression and editing variants
  • β€’ Separate test sets by modality

Evaluation

  • β€’ Model version recorded
  • β€’ Thresholds fixed before scoring
  • β€’ Repeatable test runs
  • β€’ Ambiguous samples retained

Reporting

  • β€’ AUROC, precision, and recall
  • β€’ False positives and negatives
  • β€’ Latency distribution
  • β€’ Known failure cases

Real-World Generalization

Compression Robustness

Tests should include recompression and conversion patterns common in social media and messaging apps.

Novel Generator Handling

New generator families and model versions should be added to held-out test sets before broad performance claims.

Adversarial Resilience

Evaluation should include common evasion, editing, paraphrasing, noise, and obfuscation transformations.