evidence

Comparator v3 Excerpts

Comparator v3 Excerpts

  • v1/v2/v3 progression:
    • v1 literal: 0/20/29/5
    • v2 flat semantic: 37/2/5/10
    • v3 dimensional: 25/18/6/5
  • v3 reduced flat-semantic overconfidence by weakening 15 prior full-agreement fields
  • v3 false-positive candidates: 0
  • v3 validation recommendation: safe_for_limited_calibration_use

Caution:

  • dimensional normalization remains provisional
  • literal disagreement remains real
  • unresolved dimensional matches remain high