Every label in the tank, checked against the forecast, one hour at a time.

Between Oct 4, 9:19 PM and Oct 20, 6:07 PM Eastern, 2,581 of 3,820 point-hours had both a station observation and a forecast issued about 24 hours earlier. For those, the planner ran on the earlier forecast; where it said forecast-permitted, the label limits an airport weather station can measure were checked against what the station observed.

236 of 370checked tank-mix hours held up (64%).

• All three in the tank236 held · 134 broke
• Folex 6 EC alone527 held · 232 broke
• FreeFall SC alone283 held · 174 broke
• Super Boll alone480 held · 192 broke

Every broken tank-mix hour broke on wind: the station measured 10.3 mph to 16.1 mph where the labels allow at most 10 mph (6 wind clauses across the three labels). A day-ahead forecast is a plan, not a measurement: re-run the planner close to spray time, and measure wind at the boom as the labels require. See every hour that broke.

6 tank-mix clauses cannot be checked from a station (temperature inversions, rain expected within hours, the night-low tank rule) and are never counted as held. 1270 hours had no observation, no forecast from a day earlier, or no wind reading, and are left out. Stations: Southwest Georgia Regional Airport (KABY), Americus Souther Field Airport (KACJ), Douglas, Douglas Municipal Airport (KDQH), Moultrie, Moultrie Municipal Airport (KMGR), Moody Air Force Base (KVAD). Recompute with python tools/checkback.py --recorded data/nws_recorded on the recording; full detail in eval/results.