PET Codebook v1 vs v2: Re-Coding Findings from the N=20 Mini-Pilot

Working paper documenting the re-coding of the 400 responses from the PET mini-pilot of 25 May 2026 using the refined codebook v2 (with three-indicator logic for D6 and trigger-based scale anchors for scores 4 and 5). The analysis isolates the codebook effect from model drift by holding the response dataset constant. Four key findings: (1) the central conclusion of the May memo — no persistent echo effect in phase C — remains fully intact under v2; (2) the scale compression problem is partially resolved (score 4 now used in four of five dimensions, score 5 still not used); (3) substantively new findings emerge for P2 and P4, which were hidden by scale compression under v1; (4) inter-coder reliability between v1 and v2 is solid (Spearman ρ between 0.74 and 0.86 across dimensions). Document is available in both English and German versions. The English version is the primary reference.