Item 108: The thin specification for evaluations, where no outside fact plays the world's role

Part A of the thin-legitimacy round states correct, sufficient and value over a finite set of worlds with her prior: the credence is about a fact, and correct compares it with the actual conditional. Her evaluations — what the fidelity score reads — have no outside fact to be correct about. Wanted: the same two-level structure for evaluations: what the thin properties become when the object of her later judgment is a valuation rather than a credence (candidates: correct as agreement with the valuation her program yields on the actual record; sufficient unchanged; value as a decision-theoretic statement about acting on the valuation), and the theorem that the thick conditions deliver them.

Deliverable shape: Lean in a contribution namespace, in the style of ThinLegitimacy.lean; an exact fixture. Acceptance check: python3 tests/run.py green; Lean audits clean. Context: projects/deference/rounds/2026-10-01-thin-legitimacy-and-effective-authority/REPORT.md Part A; the fidelity score's evaluator (Headline.evaluation); the after-compromise band. Consumed by: the round that restates the segment gate's "counts" on the thin level, so that the post's "illegitimate futures count as bad" has a thin-level reading. A solution ships: the three properties for evaluations with the satisfaction theorem, or the obstruction.