Cat ownership and oral health in US adults: preregistration of a secondary analysis of NHANES 2005–2006 and of a meta-experiment on AI-produced research

This deposit serves research integrity: it is the frozen, timestamped preregistration of a study that is also a meta-experiment on research produced almost entirely by an AI agent. The substantive study asks a deliberately implausible question — is living in a household with a cat associated with a higher prevalence of untreated dental decay among US adults? — using public data from the National Health and Nutrition Examination Survey (NHANES) 2005–2006, with survey-weighted modified Poisson regression, prespecified confounders, sensitivity analyses and equivalence testing. The meta-experiment preregisters how the production process itself will be measured: total and active time, human interventions, agent errors (type, severity and who detected them) and the share of text and code written by the AI agent (Claude, Anthropic), against a prespecified threshold for "produced mostly by AI". The study responds to the documented rise of formulaic NHANES-based papers (Suchak et al., PLoS Biol 2025; doi:10.1371/journal.pbio.3003152). Status: the protocol was frozen before any analysis. No association between the exposure (household cats or dogs) and any dental outcome had been computed at the time of deposit; the blind is lifted only after formal registration. The protocol is deposited here first because OSF was out of service (2026-10-16 to 2026-10-19). Formal registration on OSF will follow and will cite this DOI and the commit hash below. Contents: the preregistration (Markdown, authoritative; PDF rendering) and an archive of the project repository at git tag prereg-v1.3. The archive contains the protocol, analysis scripts (Python), process logs (human decisions, agent errors, time), the blind feasibility report, archived CDC codebooks and SHA-256 checksums of the NHANES files used. The raw NHANES files are not included; they are public and re-downloadable with scripts/01_download.py. Full texts of cited articles are not included for copyright reasons. Git tag: prereg-v1.3Commit: 4a7438cf1e489ec46a998b4b73c4f1c469485de4 (2026-10-17 08:11:38 UTC−3)SHA-256 of prereg/osf_prereg_draft.md (= prereg_v1.3.md): b3a8886a4f8154d0e263aaacb68650e012c0f6486d30ffaf850e0cc6f8e197ba Licenses: code in scripts/ under the MIT License; all other content under CC BY 4.0. Author: Diego Halabi (Universidad Austral de Chile). The AI agent drafted the protocol under the author's direction; every methodological decision was taken by the author from options offered by the agent. The agent is not an author.