What did the sky do?

Sol writes its sunset call down before the evening it speaks about. Every bake that reads the cloud forecast is kept as its own file, named by the instant it was issued and never rewritten, so a call can be read back later without anyone being able to move it. The scoreable part of that archive is published at /history/calls.json: for each city and each evening, the last call issued at least an hour before that evening's sunset. The horizon is the point. A call made in the last minutes before a sunset is a nowcast and would flatter the record, so it is not kept, and a later issue can never displace a real forecast. Day files, which the next bake overwrites, are not read at all.

The rule, in words. An evening is scored only if three things are true: Sol has a call on file for that place and evening, the call was issued at least an hour before that sunset, and you recorded yours after it. Both sides sit on the same four rungs — 1 little colour, 2 some colour, 3 good colour, 4 exceptional colour. The score of one evening is the difference: what you saw minus what was called. Zero is exact; one rung either way is close, and close counts the exact ones with it; two or more is a miss. The direction is kept, because the two failures are not the same thing: Sol over-called an evening when it promised more colour than the sky gave, and under-called it when the sky beat the call. Both are counted below, and every miss is named by its date, its place and its direction.

Which call an evening is scored against. Sol bakes a fixed fleet of cities, so a call exists for those places and no others. When you stand somewhere else, the page uses the nearest fleet city and only if it is within 75 kilometres — near enough to be under one evening's weather — and it prints that city and the distance beside the evening. Further out, the evening is still yours to keep; it simply is not scored, and the journal says so rather than borrowing a call from a city over the horizon.

The misses appear here by date, as plainly as the hits.

What this cannot establish. Not that Sol is accurate. You choose which evenings to record, so the sample selects itself. You can see the call before you rate the sky, so nothing here is blinded and your word may be pulled toward it. One person's "good colour" is not another's — the four words are not calibrated between observers, and one sky is rated once, by you, from one spot. Nothing is pooled: your evenings are yours alone, and this page writes no shared collection. Under twenty scored evenings the page prints counts and refuses to print a rate, because a handful of nights cannot separate a working forecast from a lucky one. A public accuracy figure waits on a reviewed observation set that does not yet exist, and nothing here should be read as one.

Photographs. Not in this version, and not by accident. The site is static: there is no server to receive a file and no storage configured to hold one, so an upload offered here would be a promise nothing keeps. A picture pushed into the record instead would fill the small space your journal lives in and take the rest of your evenings down with it. And publishing other people's photographs asks for consent, moderation and the stripping of the location buried in every camera file — real work, done properly or not at all. So this version takes your words, and pairs them with the sky Sol paints for that place and that minute.

Where an evening is kept. Signed out, in this browser's own storage alone, which means it is gone if you clear the site's data or move to another device. Signed in, on your account, readable by you and by nobody else. Either way every evening in the journal carries a Remove of its own, and it takes the record with it. Publication is opt-in per evening and off unless you turn it on; today it only marks the record, because there is nothing published to join yet, and no shared collection is written by this page.