can a model call the next minute?

Every minute, just after the candle closes: fetch sixty one-minute candles of a traded asset from a public feed; describe them as numbers only — the last twenty closes, percentage changes over 1, 5, 15 and 60 minutes, a relative-strength reading, recent volumes, the fifteen-minute range; ask the model one typed question: will the next close be higher? Store its probability. One minute later, store the real close beside it.

Accuracy rewards a model that says "up" whenever the base rate is above half, so it is not enough. The Brier score, the mean squared distance between the probability given and the outcome, punishes confident wrong answers and rewards honest uncertainty. A coin flip scores 0.25. Two baselines are scored on the same minutes: persistence (the next minute goes the way the last one went) and always up. A model has to beat all three to have said anything.

Nothing. A minute of price is mostly noise, and a decision model has no notion of time. The point is to measure that in public instead of assuming it either way, and to give sam a task with a clean, endless supply of labelled outcomes. If a model ever beats the baselines over a thousand settled minutes, the next question is whether it survives fees, and the answer to that is usually no.

After the first hundred minutes every model sits near or below the coin flip, exactly as expected. The live numbers are on the market page's scoreboard and update every minute.