Ranking test

Does the order of /setups mean anything? Every day since 2026-10-15 we take all setups that pass the default filter, and 5 days later compare the top 5 with the rest of that same day: no stops, no targets, so only the ranking is tested. The ranking stays as it is during the test. A day shows here once its 5 days are up.

For comparison, the backtest over 121 days: +1.79 ATR per 5 days direction-neutral (t = 2.88), +1.85% plain (t = 2.08), rank IC 0.001. Ranking code unchanged since the start.

Per day

No day has finished yet: the first result comes 2026-10-20 14:49 UTC.

Rules

• Every day at the first check after 12:30 UTC: every setup of /setups on 4-hour candles that passes the default filter (both directions, reward/risk at least 1.5), in rank order, at the entry /setups gives (the latest price).
• Exit after exactly 5 days: the close of the 15-minute candle that moment falls in. The return is signed by direction (a short gains when the price falls), before fees: fees are the same for every setup, so they don't change the difference.
• Per day: the average return of the top 5 minus that of the rest. Direction-neutral: each return minus that day's average of setups in the same direction, so a top that leans short in a falling market doesn't count as skill. Per unit of volatility: that divided by the coin's ATR, so a top of wilder coins doesn't win on size alone.
• One number per day; consecutive days overlap (5-day windows), so the t-value uses Newey-West standard errors (lag 4). A setup without a price 2 days after its exit drops out; a day needs 3 of the top 5 and 3 of the rest.
• Main measure, fixed in advance: direction-neutral per unit of volatility. First verdict after 60 counted days.
• Paper only, not trade advice. Our earlier backtest found the setups themselves no better than random entries after fees; this tests only whether the order adds something.

Data: /rank-test.json · The other paper tests: daily top 3 setups, trend following on the top 30