Synthetic Players

ClaimRegistered · verdict pass

P5-1b — between-persona dispersion checkpoint

Corrected between-prompt SDs 0.418–0.478 cleared frozen thresholds 0.309/0.234; retained as permissive.

Registered heuristic tolerance from protocol-nonmatched human SD references (ρ=.75)

P5-1b compared panel dispersion against thresholds mechanically implied from Dal Bó & Fréchette’s R=40 strategy-frequency SDs (0.4122 for δ=.50, 0.3116 for δ=.75; frozen ρ=.75 gave thresholds 0.3092 and 0.2337). Finite-opportunity-corrected plug-in SDs were 0.4182, 0.4784, 0.4408, 0.4323 — all clear.

Caveats made explicit after review: unanimous cells are point masses in the conditional bootstrap; a two-stage prompt+episode bootstrap (a persona-generator estimand) widens the intervals to 0.27–0.51; and the displayed equality of one lower bound with the human 0.4122 reference is a rounding coincidence, verified by independent recomputation. P5-1b is a permissive historical checkpoint, not evidence of human-variance equivalence.