Synthetic Players
Paper PDF

AnalysisPost-adjudication

Finite-opportunity variance correction

Corrected between-prompt SDs 0.418–0.478; plug-in between-shares 85.5–96.1% with bootstrap intervals.

50,000 bootstrap replicates · 16 personas × 6 episodes per cell

Recordv11 latent-propensity view

Separates real between-prompt variation from finite-opportunity measurement noise (six episodes per cell): corrected SDs 0.4182 / 0.4784 / 0.4408 / 0.4323, between-shares 0.855 / 0.961 / 0.888 / 0.902. The v11 companion adds the fixed-panel Dirichlet–Jeffreys latent-propensity view (share medians 0.631–0.705), created after Round 9 identified that resampling unanimous six-episode cells as point masses is a conditional statement, not a complete uncertainty statement. Together they are the two bracketing views quoted throughout the paper.

Linked in the research graph

Everything connected to this item, in both directions.

Claims

Figures

Review rounds

Manuscript versions

  • v11 — the post-freeze addendumSupersededAdds the fixed-panel Dirichlet–Jeffreys latent-propensity sensitivity — the uncertainty view that turned “85–96% dominant” into “63–71% medians, prior-dependent dominance” — as an explicit, declared addendum to the frozen v10 (22 pages; margin line numbers removed)..