Synthetic Players

Adversarial record

Fourteen review rounds

External critique changed this paper's central claim — and the full trail is preserved: reviews, disposition matrices, reviewer errors, and role disclosures.

The correction record is part of the result. Reviews diagnosed the missing family control that demoted p13, the seat-vs-episode dependence defect, the boundary-uncertainty gap that produced the latent-propensity view, and the prior sensitivity that demoted P5-2's Bayesian corroboration. Reviewer roles are disclosed — including one reviewer's expansion from critique to analysis specification, and one reviewer's retracted fabrication accusation that became a permanent artifact-identity rule. Preservation rule: sealed research files are never edited in response to review; corrections are additive.

RoundReviewer · date · scoreManuscript arc
Round 1Multiple AI reviewers (ChatGPT, Gemini, Grok R1)v1 → v2
Round 2Adversarial methods reviewer + two editorial reviews · 2026-07-29v2 → v3
Round 3Independent reviewer, blobless anonymous clone · 2026-07-29v4 → v5
Round 4Independent reviewer, ran the capsule end-to-end · 2026-07-29v5 → v6
Round 5Explore Science (external) · 2026-07-29 · 92/100, Platinumv6 → v7
Round 6Claude (repository review)v7 → v8
Round 7Claude (independent, commit-pinned)v8 → v9
Round 8Claude (mechanical freeze-integrity review) · 2026-07-30v9 → v10 freeze
Round 9Explore Science (external) · 2026-07-30 · 97/100, Platinumv10 → v11 addendum
Round 10Explore Science (external) · 2026-07-30 · 96/100, Platinumv11 → v12
Round 11Explore Science (external) · 2026-07-31 · 96/100, Platinumv12 → v13
Round 12Claude (deep review, commit-pinned)v13 → v14
Round 14Final Claude review · accepted into the final candidatev14 → final

There is no Round 13 in the archive — the numbering jumps from 12 to 14. Author-side response plans are archived separately from external review text.