# Human review and AI use

This note identifies what readers should review before relying on the project summary. It is a review aid, not an independent validation of every source or analysis.

Review the estimand and model assumptions alongside the reported interval; confirm that evidence-register records are not mistaken for independent studies; and consult the reproduction description for the boundary between supplied-file checks and unavailable full raw-source replay. The main manuscript and supplement contain the detailed methods, design-specific results, and limitations.

The author originated the ideas and research questions. Claude (Anthropic) and Codex (OpenAI) wrote and revised substantial portions of the paper. AI systems used in the project included Claude (Anthropic), ChatGPT and Codex (OpenAI), DeepSeek-V4 and DeepSeek-V4.1-Flash (DeepSeek), GLM-5.3 (Z.ai), Grok (SpaceXSI, formerly xAI), and Kimi (Moonshot AI). These systems assisted with source discovery and retrieval, data extraction, analysis code, statistical analysis, figures, and drafting and reviewing project materials. Their suggestions and cross-model reviews are not independent scientific validation. This disclosure does not imply that every source record or analysis received external review. The reported completion checks and replay tiers describe the checks actually performed; scientific judgment about assumptions, identification and generalizability still requires human review.