Does a sensor-and-rainfall model beat DC's post-rain advisory? A season-held-out skill audit of E. coli exceedance at Hickey Run and Watts Branch

Preregistered, blinded retrospective test of whether a penalized logistic model using USGS continuous turbidity and antecedent rainfall classifies single-sample E. coli exceedance (above 410 MPN/100 mL) better than the fixed post-rain advisory rules in circulation in Washington, DC (avoid contact 48 h or 72 h after ≥ 0.5 in of rain), at matched sensitivity, with leave-one-season-out validation, at Hickey Run and Watts Branch (Anacostia tributaries). Full plan: attached PREREGISTRATION.md (locked 2026-11-08, git tag prereg-v1).