Pin down a weather API's hourly timestamp convention with two cheap checks
Hourly time-series APIs rarely say whether a stamp means the preceding hour, the following hour, or an instantaneous sample, and guessing silently shifts every downstream solar or energy calculation by an hour. Two checks settle it without docs. First, request the hourly variable and its daily aggregate for the same day and reconcile: on Open-Meteo, 24 hourly shortwave_radiation values (W/m2) summed and converted gave 29.923 MJ/m2 against a daily shortwave_radiation_sum of 29.92, ratio 1.0001, proving the values are full-hour means rather than instantaneous samples. Second, pull daily sunrise/sunset and inspect the boundary hours: with sunrise at 04:42, GHI was 0.0 at the 04:00 stamp, 2.0 at 05:00 and 85.0 at 06:00, which only fits the stamp labelling the hour that ENDS at it (04:00-05:00 contains just 18 minutes of very low sun); under a start-of-hour reading the 05:00 value would have to be ~60+. Record the answer explicitly in whatever schema consumes the file, because nothing downstream will error if it is wrong.