Synthetic Players

Provenance

Manuscript genealogy — v1 to the canonical release

Fifteen versions, each prompted by a named review round; earlier drafts preserved byte-exact, never rewritten.

The title itself records the corrections: “Empirical Deviations from Nash Equilibrium…” (v1) → “Passing Marginal Checks Can Be Cheap… Weak Incentive Response” (v2) → adding “Coarse” (v3) → “Persona Mixtures and Imprecise Incentive-Response Estimates” (v5–v7) → “Treatment-Response” (v8, final). The v10 scientific text freeze is externally timestamped; v11 is its first declared addendum.

VersionStatusWhat changed
v1 — the refuted prototype paperSuperseded“Empirical Deviations from Nash Equilibrium in Classic Game Theory Games” — 40 experiments, 11 claims, all initially called supported.
v2 — first corrected draftSuperseded“Passing Marginal Checks Can Be Cheap: Persona Mixtures and Weak Incentive Response…” — p13 demoted from confirmatory to suggestive (family p = 0.0525 ± 0.0050), δ-matched and fivefold comparator language retired, and a 17-row v1→v2 correction ledger appended.
v3 — the registered reanalysis queueSupersededAdds “Coarse” to the title, retires the “Level-Matching Is Cheap” hook, and registers the A.7 queue of five zero-call reanalyses with precommitted reporting rules — the blueprint for the submission analyses.
v4–v5 — verification and readinessSupersededv4 was verified by the Round 3 anonymous clone; v5 adopted its corrections (all three family constructions reported, registration language scoped to confirmatory claims, reviewer-role disclosure, drafts committed to history, five contributions cut to three) and was declared “ready for full scientific review” by Round 4..
v6 — first Explore Science PDFSupersededThe first formatted review PDF (14 pages), scored 92/100 Platinum with 13 minor issues..
v7 — the Round 5 integrationSupersededIntegrates all thirteen Explore Science issues: the five-stage architecture table, protocol glossary, X2 operational definition, corrected Figures 1 and 5, represented-treatment language, and the corrected p13 power statement.
v8 — treatment-response retitleSupersededEleven Round 6 recommendations, including the title change from “incentive-response” to “treatment-response” and the p≈0.075 attainability boundary added to Figure 5.
v9 — freeze candidateSupersededThe text-freeze candidate: archived-panel vs latent-propensity separation, “cheap” defined as evidentiary economy, count reconciliation integrated.
v10 — the scientific text freezeSealed recordThe frozen scientific text (21 pages, tag paper-text-freeze-v10, OpenTimestamps-anchored).
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)..
v12 — near-arXiv preprintSupersededExtends the replay perimeter to all of Phase 3 (320 runs + 20 baselines), adds the independent bootstrap-coincidence audit, full leaning-strata table, matched-lattice entropy reanalysis, and decoding audit; the abstract leads with the latent-propensity posterior.
v13 — dependence-aware candidateSupersededAdds the global-multiplicity limitation, provider-failure accounting, P5-2 prompt-cluster bootstrap and Dirichlet–Jeffreys sensitivities, and interval-bearing Figure 3.
v14 — arXiv candidateSupersededFixes the three Round 12 release blockers and incorporates the Round 14 middle-path restorations and precision corrections.
Final — canonical releaseFinalThe canonical 19-page, five-figure release: PDF, byte-identical Markdown, and minimal PDFLaTeX arXiv package, all SHA-256-pinned, independently recompiled and page-image-compared in CI, OpenTimestamps-anchored, and published with the project site on 2026-08-01 (PR #13)..

Historical PDFs and hash manifests live in docs/paper/; exact v2/v3 sources in docs/paper/history/. v4/v5 exist as verified states of the working draft rather than separate files; the v9 PDF survives as its recorded hash in the Round 8 review.