Goal — To make patterns in markets visible: when something recurs regularly, currently on Bitcoin and gold — and in which timeframes trading actually happens. The benefit runs both ways: you see when the market is alive, and you learn what you no longer need to watch. Explicitly not a trading bot and not investment advice.

Challenge — A pattern that worked in hindsight usually only works in hindsight. The difference between structure and chance is measurable, but the measurement costs more than the pattern and almost always comes out against it. That is precisely why it is rarely done — and why patterns that were never examined persist.

Current state — Live, in 3 layers. The first shows patterns by weekday and hour of day: which hours of which days are loud and which are quiet. The second is a test lab with an overfit gate stricter than the result it judges — it is the actual subject of the project, because it decides what gets passed on at all. The third, delivery as a message to the operator, is built; what it delivers is the gate’s decision alone, and so far the gate has let nothing through. 7 hypothesis groups are closed, alongside several purpose-built counter-probes. What survives is made visible whether or not it is tradeable: a sub-page shows volume seasonality as a picture, with an explicit note that no direction follows from it. One finding held and is explicitly not a trading idea: the major crypto assets share one weekly clock. live

What part AI plays — In the evaluation none: this is statistics, and there is no model in this system that predicts anything. In the building, throughout: the 3 layers, the overfit gate and the counter-probes are developed by AI agents — including the checks that rejected the project’s own results.