Back to the auditor

Detection dimensions

TrueLLMs aggregates 12 independent signals into a single weighted verdict. Each dimension is a pure TypeScript function with documented thresholds and a known failure mode. Click any row for the deep write-up.

  1. 01Logprobs Fingerprintweight 17%
  2. 02Tokenizer Boundary Probeweight 15%
  3. 03LLMmap Active Probingweight 15%
  4. 04Model Equality Testing (MMD)weight 12%
  5. 05Inter-Token Rhythm Fingerprintweight 8%
  6. 06Cache Hit Detectionweight 8%
  7. 07Canary Prompt Behaviorweight 7%
  8. 08Context Window Probeweight 6%
  9. 09Sparse-Token Stress Testweight 5%
  10. 10Stylometric Analysisweight 3%
  11. 11Latency Distributionweight 2%
  12. 12Self-Identification Probeweight 1%
  13. 13Refusal Boundaryweight 1%