Some evenings the sky puts on a show. Vivid tells you which ones.

Most weather apps hand you a score out of 100 and leave you to guess what it means. Vivid's headline is a percentile: "82% of the past year's evenings here ended duller than tonight promises." The same sky that's ordinary in Santa Fe would be the evening of the month in London, so every place is ranked against a full year of its own history. A 70 means the same thing everywhere: a top-third evening for wherever you're standing.

Every forecast comes with its reasoning spelled out in plain sentences — which factor helped, which hurt, and what would have to change for tonight to get better.

Vivid scores each evening with a physical model. Cloud is weighted by altitude, because high cloud holds color and low cloud blocks it — and by layering, because sunset color is light from below: a solid mid-level deck shades the high cloud above it, and a full overcast's own underside goes leaden where broken cloud would take color. The score is also capped when cloud along the light path would stop the sun from reaching your sky at all. Eight independent forecast systems are scored separately — ECMWF's physical model and its newer machine-learning one, the UK Met Office, DWD's ICON, NOAA's GFS, Environment Canada, Météo-France, and JMA — then blended by track record, with the weights following each center's published verification standing: ECMWF's two models first, the UK Met Office next, and so on down. A blend is steadier than any single run. The ranking comes from scoring the past 365 evenings at your coordinates from the ERA5 weather archive.

The week view ranks each evening the same way: a bar per day, scored out of 100. The solid stretch is what every system agrees on, the pale stretch is the range they differ over, and a caret above marks the blended forecast inside it — so a wide pale stretch is an unsettled evening that could land on either side of the caret.

The caret is a weighted average, not the midpoint of the range, which is why it often sits off-center: the forecast systems with the better track record count for more, so when they read brighter than the rest, the caret rides high even though most of the pack sits lower. It is deliberately not a show of hands. The ranges are wide because cloud cover is the least reliable thing in weather forecasting, and because the score is deliberately sensitive to it — fifteen points of low cloud can be the difference between color and gray. Tapping a day says in plain terms how firm that evening is and whether it's worth committing to yet.

It has real limits. Cloud forecasts are the weakest part of weather prediction, and no scoring model can beat the forecasts it is built on. Past about day three the honest answer is a range, and that is what Vivid shows. Some caveats worth stating plainly: the model weights are priors from each system's general track record, not yet measured against sunsets; the year of history is scored without the light-path check, so in places with a chronically cloudy west the rank runs a little low; the history is scored from a weather archive while tonight comes from live forecast models, two records that don't see cloud identically; and because tonight's number averages several systems, it sits closer to the middle than any single sky will — on split evenings the extremes are understated by design. The model was tested retrospectively against a decade of photographs, which is encouraging but a small sample — formal calibration against rated evenings is ongoing.

A promise. When the forecast systems disagree, the page says so instead of hiding it behind one confident number.

It also isn't a business. There's no account and no ads.

I kept missing good sunsets and catching bad ones. The apps I tried put their useful features behind a paywall, and their multi-day forecasts kept promising skies that never arrived. So I built the version I wanted, one that ranks tonight against where I'm actually standing and tells me why, down to the hill to my west.

Before trusting it, I tested the model against ten years of my own camera roll — the evenings I actually stopped and photographed. The evenings I photographed most were the ones it scored highest, including the single biggest sunset in the whole library. It's young, and it will be wrong some evenings.