SD-243

[tells-calibration-v3] Tells calibration v2→v3 completed. 5 discriminating features confirmed. Transition word density (d=-2.40), nom density (d=-2.28), question rate (d=1.56), first-person rate (d=1.59), contraction rate (d=1.04). Structural rhetoric detectors (epigram, isocolon, anadiplosis, antithesis) measure writing quality, not writing origin — do not discriminate. Extension updated to v0.2.0 with voice/rhetoric split badge. 46 pages across 6 categories (A: pre-LLM human, B: post-LLM human, C: AI company, D: suspected LLM, E: Willison longitudinal, F: Operator). Data: tells-ext/calibration-data-v3.tsv. Analysis: tells-ext/calibration-results.md.

Complete

---

← all decisions