What runs now, real money first, then every strategy version run from the first recorded market to the latest one, on the same markets. The shaded fan is ±2 standard deviations of luck alone, widened by what each market actually staked, so a run of big bets widens it more than a run of small ones. A curve that climbs out of the fan and stays out is showing edge; inside the fan it could be luck.
~/.config/polycoin/pubsub.env, the same one the alert app uses.
Edge vs luck is the average result per market divided by its standard error, the same statistic the lab's forward tests are judged on. Above 2 the result is hard to get by chance; between −1 and 1 it says nothing yet. The fan on the chart is the same measure weighted by each market's stake, so the two agree whenever every bet was the same size and part when they were not. Click a row to show its fan.
Each version minus the one it was built from, market by market. Both traded the same markets, so what they share cancels and only the change is left. A change near 0σ did nothing you can tell from luck; a version whose edge comes from an earlier change is riding on it.
Replays fill resting bids only after everyone already waiting at that price, and takers pay the fee at the ask they see. Versions that priced with a fitted blend, gold, and the spread bot can't be replayed and are left out. Markets missed while the recorder was down are filled in by the next full replay.