0002 — RW reads the run ending on the hour, not HH:50

Status: accepted (2026.11.0)

Context

DWD runs RW every 10 minutes, each run the hour ending at that :x0. Up to 2026.10.1 the integration read the HH:50 run. Nothing recorded why; it is RADOLAN's traditional hour, and SF is still only published at :50 (OpenData has no other SF minute).

Only a clock hour can be called "previous hour" (see 0001) and line up with HA's hourly statistics. The question was whether the :00 run is as good as the :50 one.

scripts/compare_rw_runs.py compares every RW run on OpenData (48 h) with radar-only RS _000 for the same hour, cell for cell. If the sub-hourly runs used fewer gauges, they would sit closer to RS. They do not: over 285 wet runs (2026-10-10..12) every run minute departs from RS alike (mean |RW-RS| 0.53-0.58 mm, 58-62 % relative, 76-78 % of wet cells adjusted). Every run minute is published at ~25.7 min (288 runs, 2026-10-13).

Decision

rw reads the run ending on the hour. SF stays at :50, so rw and sf windows are now 10 minutes apart.

A total stored before 2026.11.0 has a last_release at :50, off the new grid. Restoring it maps it once with counted_on_grid() onto the :00 release before the first window that starts after it, so counting resumes there: the 10 minutes in between are left out once, rather than 50 counted twice (counted_until reads 10 minutes ahead until the next release).

Consequences

Rerun scripts/compare_rw_runs.py before relying on the quality comparison for anything new; DWD may change how the runs are adjusted.