Auditing evidence for 24-h low-dissolved-oxygen warnings before model comparison: code and public-source derived evidence

This release provides analysis code, tests, public-source derived data and provenance for auditing whether dissolved-oxygen early-warning model comparisons rest on an interpretable evidence base. It separates (1) observational interceptability, governed by observed low-DO run length relative to monitoring cadence, from (2) repeated representation of the same low value in overlapping 24-h sliding targets. The constraints arise at different stages and cannot compensate for one another. This is a pre-model evidence audit; it does not claim universal predictability, model-performance improvement, causal mechanism or management effectiveness. Restricted Heshui, Guangdong, CNEMC/national and other permission-unconfirmed monitoring records and their row-level derivatives are excluded.