SST™ Engine
Statistical Signature Transfer
Statistical Signature Transfer™ (SST™) is a core sub-engine of ASH™. It measures the statistical "fingerprint" of human reference texts — sentence-length distribution, punctuation rhythm, Zipf compliance — and then systematically shifts the input text's profile to match.
How It Works
Reference Profile Extraction
SST™ builds statistical profiles from a corpus of authentic human writing across 5 genres: academic, web, journalism, fiction, and technical documentation.
Divergence Scoring
Computes Jensen-Shannon divergence between the input text's statistical distribution and the target human profile — identifying the largest "AI giveaway" features.
Profile Alignment
Applies targeted edits to align the input distribution: adjusting sentence-length variance (burstiness), punctuation frequency, and clause-length patterns.
Convergence Verification
After each transformation pass, SST™ re-measures divergence. Editing stops when the score falls below the configured threshold (default: JS-div < 0.042).
Key Metrics
Use Cases
- Shifting AI-uniform sentence lengths to human-like variable rhythm
- Matching punctuation density of target publication style
- Aligning Zipf frequency distribution for natural vocabulary spread
- Statistical profile matching for domain-specific content (legal, medical)
Try SST™ Engine Now
Powered by our proprietary technology stack. Available directly in the TextHumanize tools and via REST API.