After The Forecast

30 September 2026, Edition No. 4

A trace was counted as rain. At RAF Lakenheath, a METAR P group of P0000 reports a trace: rain too little to measure. The feed we read stores it as 0.005 inch, 0.127 millimetres, which clears the tenth of a millimetre at which an hour counts as wet. Edition No. 4 therefore scored 30 of its 42 wet hours on reports that measured nothing, and its wet week of 17 to 23 August included 3.4 millimetres that were never measured: 18.8 millimetres rather than the 22.2 it printed.

From Edition No. 6 a trace counts as a dry hour, and the method note says so. Edition No. 4 has not been rebuilt, because the rain section it was built with has since been replaced; its rain timing scores rest largely on those trace hours and should not be cited. The Netherlands editions are not affected: De Bilt's gauge reports millimetres, not traces.

25 August 2026, Edition No. 1

The edition was rebuilt now that 13 August has closed, as the note on this page promised.

The heat section is no longer provisional. It was published against the 33 degrees the day was expected to reach, with 18 of its 24 hours reported. The day finished at exactly 33 degrees, so the expectation held and not one figure in the forecast grid moved. The provisional marking has been removed from the page.

The common sample grew, and every score with it. The edition now rests on 960 station hours over 40 complete station-days at three stations, where the published version had 504 hours over 21 station-days at two. The period did not change: what changed is how much of our own collected record of Apple Weather and Foreca is available for it, and every ranked model is still scored on exactly the same hours as every other. The temperature and wind scores in the standings therefore differ from the version published on 13 August.

The runner-up day in the method note moved from 29 July to 4 August, at 32.9 degrees. De Bilt's record for 4 August was incomplete when the edition was first built and is complete now, which makes it the hottest closed day of the period after the one the section leads on.

Two sentences were corrected rather than recomputed. The edition described 13 August as the start of a heatwave; by KNMI's definition the spell that followed was not one, and the page now says only that the heat arrived that day. Its closing paragraph also promised that rain would be added once enough independent radar data existed. Rain has returned, from a source that was not available when that sentence was written.

25 August 2026, rain returns

Edition No. 1 removed its rain sections in August because they were scored against Schiphol's present-weather reports, a truth source that exists at only a handful of Dutch fields. The note said rain would wait for the independent radar record. It has not: rain is now scored against the rain gauge at De Bilt, the station the edition already used for temperature.

The gauge is a better answer than the one promised. It measures what reached the ground at the station being forecast, it is independent of every model on the page, and it reports every ten minutes, so an hour is a measurement rather than an observer's note. The radar record remains a rolling 35 days, which cannot carry a reporting period without being copied aside first. No rain figure from the version published in July should be cited; the figures in Edition No. 3 rest on a different and better measurement.

13 August 2026, Edition No. 1

The edition was rebuilt with four changes, all of which move published numbers.

The event day moved to De Bilt. It was read from Schiphol before. De Bilt is KNMI's reference station and reports every ten minutes rather than twice an hour, so the observed daily maximum is measured more closely. It became available to us only once we began collecting the KNMI station feed directly.

Apple Weather and Foreca joined the ranking, and the UK Met Office model left it. The two commercial services are the forecasts most readers actually use, and we now hold enough of our own collected record to score them. Because that record begins in mid-July, the common sample every row rests on is shorter than before: 720 station hours over 18 dates, where the previous version had 1,560 over 38. Every ranked forecast is still scored on exactly the same hours as every other.

The rain sections were removed. The previous version ranked rain against Schiphol's present-weather reports. That truth source exists at only a handful of Dutch fields and cannot carry a national claim, so rain is held back until the independent radar record can support a full period. No rain figure from the earlier version should be cited.

The heat grid now runs to eleven days. It stopped at seven, the depth of the standings. It now keeps reaching back while any model still had the day inside the band and ends where none of them did, which is the point the section is making.