Findings¶
Five empirical findings from the v2 investigation, each with a claim, evidence table, action, and related-work pointer.
- Finding 1 — Preamble content is load-bearing: outputs move measurably in either direction relative to a no-preamble baseline
- Finding 2 — Preamble effects are governed by rubric overlap, not by "expertness"
- Finding 3 — Bare enumeration captures most of the positive effect; expert framing is decorative
- Finding 4 — No "alignment vs capability" split, just preamble–evaluator overlap
- Finding 5 — Static-analysis tools cannot detect preamble effects on craft