Answers are grounded in this paper and cite their sources.
Pre-registration of the main analysis of a study on whether precipitation phase (rain versus snow) changes event streamflow delivery in 582 snow-affected US catchments (CAMELSH, 2003-2024). Status of the data at deposit (2026-10-24). Event outcomes (7-, 30- and 90-day quickflow delivery fractions for 522,575 events) were computed on 2026-10-23 and sealed in a read-only file (sha256 6040c8d4b5c28c71d0ddb3d1996d7916597c68511f0c745c8233bc4eeaddf46e). The only access was an outcome-blind minimum-detectable-effect script, which used absolute pair differences and reported only the spread of a sign-flip null. No snow-versus-rain contrast has been computed or examined. Hypotheses. RQ1: does the 90-day delivery fraction differ between snow- and rain-dominated events once input rate, antecedent wetness, snow storage, season and recent temperature are held fixed (H1a: snow higher; H1b: no difference)? RQ2: is any phase effect concentrated in the 7-day fraction? RQ3: how much of the published rain-melt contrast is reproduced by negative controls? RQ4: can the event-scale effect account for the annual snow-fraction elasticity of runoff ratio? Design. Primary: same-storm spatial contrast (Design B). A snow-dominated catchment is paired with a rain-dominated catchment in the same HUC2 region, storm week and 24 h window, matched on area, elevation range and climate (3,863 pairs in 452 storm weeks). Supporting: a near-threshold wet-bulb band with within-catchment matching (Design A; 1,086 pairs). Also a dose-response regression (Design C). Design A was demoted by the pre-specified feasibility rule: corrected minimum detectable ratio 1.38 against a 1.15 limit, after SNOTEL validation of the SNODAS phase classifier gave an attenuation factor of 0.29. Contents. The v1 analysis plan; amendments A1-A10 with dated approvals; all step-02 specifications; the Design B and main-analysis specifications (02f, 03); the gate G1 decision memo; outcome-blind step-02 results; and a claims register (53 numbers, re-checked from saved outputs with zero differences). The sha256 of the code snapshot at commit bd32252 is listed in the README; the full code will be deposited with the paper. AI use. An AI agent (Claude, Anthropic) wrote and ran the analysis code under the author's direction, inside a written protocol with decision gates. Every analysis was specified before it was run, and every reported number is traced in the claims register. The author made all scientific decisions.
No takes yet. Share an insight, caveat, or question.
Zeeshan Asghar (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: