Codex audit of 5 sample regimes after PR #6 auto-generation. Actionable items for v5.1.

• china/tang: 司礼监 (Ming-Qing eunuch agency) was misplaced in Tang architecture. Replaced with 中书舍人 (Zhongshu Sheren, Tang drafter). "运行超过 1300 年" overclaim clarified — only the three-省 framework persisted, not the specific Tang configuration.

• global/byzantine: Patriarch depicted as routine ethics veto — overstated. theokrator is literary, not a standard office title.
• global/roman-republic: Chart too linear. Assemblies (Comitia) should sit more prominently above consuls and senate. Censor placed too actively in day-to-day flow (role was periodic, not continuous).
• global/prussia: Compresses 1701-1918 into one snapshot. Ober-Kriegsrat may be invented. Kammerdirektor is too generic. Recommend splitting into prussia-frederick (18c) and prussia-bismarck (19c).
• global/ottoman: Nişancı miscast as 大法官 (should be pen-bearer/tuğra authenticator). Janissary Aga's direct-to-Sultan reporting path overcentralized — ignores parallel ulema channel. One source citation (Patrick Balfour) format nonstandard.

• A second-round review (originally attempted on a now-removed provider) did not complete.
• cc-mimo third-round review hit --max-turns 1 limit mid-output; re-run needed.

v5.1 plan: complete the second/third review rounds via Codex + opencode + Mimo on a dedicated branch, fix Byzantine/Roman/Prussian/Ottoman per above. (gemini is no longer used — see CHANGELOG v5.1.0.)

A full historical-accuracy audit of all 57 regimes (web-verified against standard scholarship) was completed and the findings fixed. This closes the v5.1 plan above.

• byzantine: dropped the literary "Theokrator" (→ Basileus/Autokrator); reframed the Patriarch from a routine veto to doctrinal + coronation-legitimacy authority under symphonia (not caesaropapism); fixed a mislabeled citation (《秘史》 = Procopius's Secret History, distinct from De Administrando Imperio).
• roman-republic: already conformed (assemblies above consuls/senate; censor noted as periodic) — no change needed.
• prussia: replaced the invented "Ober-Kriegsrat" with the real Generaldirektorium (1723); narrowed the 1701–1918 blend to a coherent Frederician 18th-c. snapshot.
• ottoman: Nişancı corrected from "chief justice" to chancellor / kanun-keeper; added the missing Şeyhülislam (ulema/fetva) channel; Kazasker set as chief judge.

• Citations (fabricated/misattributed): carthage (Aristophanes, Plutarch Life of Hannibal), sparta (Hansen→Cartledge), safavid (Qian Mu→Newman), meiji (Jansen/Hall conflation + dup title), zulu (Knight not "Greaves"; Wright fabrication→Morris), venice (Curtis→Lane/Norwich), zhou (《西周史论》→《西周史》), five-dynasties (新五代史/五代史记 dedup→Wang Gungwu), swiss (Riker 1962), viking (Snorri fabrication→Heimskringla), caliphate (Kennedy title align).
• Structure/anachronism: liao (林牙 civil not military; 敌烈麻都→夷离毕), joseon ("Three Censors"→"Three Offices/Samsa"), jin-jurchen (un-pin the founder from post-1125 organs), yuan (中书令→右丞相; 理问所 moved under 行省), russian (narrowed the impossible 1547–1917 blend to a Muscovite Tsardom snapshot), aztec (Tlacochcalcatl under the Tlatoani, not Tlacopan), roman-empire (Magister Militum independent, not under the finance ministry), ming (批红 = 秉笔太监, sealed by 掌印太监), eu (eurozone now 21 after Bulgaria, Jan 2026).

Traditional Chinese historiography legitimately retrojects later office names onto earlier dynasties; these are kept as the conventional reconstruction rather than "corrected": shang 太师, xia 司空/司徒/司马, jin (Sima) 中正 placement, north-south 六官 generalization, three-kingdoms 都督 apex, zhou 三公 (太宰 carries the real 冢宰 administrative function). khmer Bakong/Bakheng inscription wording is a minor, low-confidence ambiguity left as-is.