Answers are grounded in this paper and cite their sources.
Three anomalies from preregistered experiments on a deterministic political-economy engine, every number recomputed from published data rows by a verifier that runs green in public CI: (1) a family-dependent protocol-compliance cliff (2 of 4 model families fail when the legal menu contains a single action, all comply on richer menus); (2) volume-confounded concentration metrics that invert the targeting signal; (3) an adversary-coverage bound where exhaustive policy enumeration keeps the share cap but breaches the HHI line and a preregistered downgrade fires. Verification: world-100-grid (verify.py, mutation journal, one-command compliance harness). Engine and preregistrations: alashi. Source repository: alashi-anomalies-note. This deposition is version 1 (single-model compliance case study); version 2 adds the four-family table.
No takes yet. Share an insight, caveat, or question.
Amir Zhakyshev (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: