Ground truth · a number that does not exist yet

Six Months Behind the Sun

The Sun reached the maximum of Cycle 25 in October 2024. Nobody could say so until May 2025, and for eleven monthly updates before that the official answer was a different month and a smaller number every time. This page recomputes all of it from SILSO's own files, fetched when you loaded it.

Where the Sun is, in two currencies

Both boxes are true and they disagree, because they are different statistics. The left one is smoothed and therefore late. The right one is current and therefore noisy. Mixing them, putting tonight's monthly mean over the smoothed peak, is the single most common way to get a sunspot number wrong, so this page never does it.

The anchor: make it reproduce their published numbers first

SILSO published an article titled "Solar Cycle 25 reached its maximum in October 2024", which states:

the maximum value of the monthly smoothed sunspot number for Solar Cycle 25 was reached in October 2024 and is 161 (see figure 1), versus 159 for September and 157 for November 2024.

They derived those from their smoothed file. We derive them from their monthly file, by applying the 13-month weighted mean ourselves, and then check the result against every smoothed value they ever published. If our smoother did not agree with them on the months they have published, it would have no business speaking about the months they have not.

The control: can this comparison fail?

Agreeing with SILSO three thousand times proves nothing unless the comparison is capable of disagreeing. So plant a wrong value. The number you set below is added to that month in the monthly data array, before anything runs, and every calculation on this page then reads the doctored array without being told. Nothing is appended to a result list. If the comparison stays green with an error planted in it, the comparison is worthless and you have caught this page being worthless.

The frontier, and the six months that do not exist yet

SILSO's smoothed file does not stop at its newest value. It carries rows for six more months with the sentinel -1.0 in the value column, which means "this cannot be computed yet", not "zero sunspots". Here is that edge of the file, with what the smoother can already say about each blank month.

Watch the maximum move

This is the part that turns a caveat into a measurement. Drag the control to choose how much monthly data has arrived. With data through month T you can smooth as far as T minus 6, so the maximum you would have announced is the maximum of everything up to there. The chart draws the smoothed curve you could see at that moment, and the readout is the announcement you would have made.

The scoreboard, and the reason most of it refuses to answer

Now score the forecasts. Pick an observed series and a published forecast. Most pairings will not give you a number, and that is the finding rather than a limitation: published Cycle 25 forecasts are stated in at least three different statistics, and a percentage between two different statistics is not a forecast error, it is a category mistake with a decimal point. Where this page refuses, it prints the number it would have printed if it had gone ahead, so you can see what the mistake was worth.

The honest headline is not "solar forecasting fails". The 2019 consensus panel, five and a half years out, missed by a wide margin. NOAA's own experimental revision of 25 October 2023, about a year out, gave a range and a window that the observed maximum landed inside on both axes. Select those two in turn and the page will show you exactly that. The same goes the other way for the forecast that missed high: McIntosh and colleagues put Cycle 25 at 233 in 2020, and the same authors revised it down to 184 in September 2022 once the event their method waits for had happened. Both statements are on the scoreboard, because scoring the first without the second would imply they never moved.

The forward claim: an answer that is already three quarters decided

The smoothed value for the first blank month is not a mystery. Twelve of its thirteen terms are already published, and they carry eleven and a half of the window's twelve units of weight. Only one term is missing, and it enters with a weight of one half out of twelve, so the whole remaining uncertainty in a number that will be printed by the Royal Observatory of Belgium is one twenty-fourth of a single month's sunspot count.

The second claim is longer and simpler: the announced maximum of Cycle 25 has now held still for more than a year, and this page claims it will still be the same month at the same value, to within 0.5, on every future run of its verifier. Both claims are scored on every run and the outcome is appended to research/six-months-behind-the-sun/record.jsonl whether they hold or not.