Abstract
<p>Background: Computational research is reaching clinical journals faster than the infrastructure to evaluate it, while meta-research reads journal policy through automated agents assuming it is legible. We audit what journals document to evaluate such work and whether that policy is machine-legible.Methods: From a SNIP-stratified random sample of 150 MEDLINE-indexed clinical journals in six specialties, we hand-coded five editorial-capacity indicators, separating disclosure, deposit, and verification. Coding was checked against a timestamped archive, with blind automated cross-checks and independent second coding on indicator subsets (pooled κ = 0.69).Results: Requirements clustered into 50 policy units across the 150-journal frame. A data-availability statement was mandated by 32% by publication but only 6% at submission; none mandated code deposit or a runnable artifact. A statistics/methods editor appeared on 28% of boards, an AI editor on 5%. Half the author-instruction pages (75/150) were refused by all three standard HTTP clients, and one automated pass recovered 3 of 48 data mandates.Conclusions: Clinical journals document abundant means to disclose where data are but almost none to verify a computation runs—and that policy is barely legible to an automated audit. One problem seen twice: infrastructure not built to be read. Data and code are openly released.</p>