Keep factual scope and everyday language in natural rewrites

#2
by ciaochris - opened

Keep factual scope and everyday language in natural rewrites

Live verification showed the model retaining ceremonial praise and introducing
a first-person actor into an impersonal pilot report. Explicit editorial
examples and a scope check target those observed outputs.

Confidence: medium
Scope-risk: narrow
Tested: Unit regressions and live writing examples
Directive: Evaluate prompt changes against facts and voice, not detector scores

ciaochris changed pull request status to merged

Sign up or log in to comment