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
- v1 literal:
- v3 reduced flat-semantic overconfidence by weakening
15prior 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