Screener presets, backtests and survivorship bias — an honest guide

Each one-click preset in our screener shows a badge like 59% in profit · 7d · 22 sig. — the historical win rate of that preset's conditions replayed over daily candles. Most screeners don't show this at all; the ones that do rarely explain the fine print. Here is ours, in full.

How the numbers are made

• We replay only the historically reproducible subset of a preset's conditions — things derivable from price history (RSI, moving averages, regimes, patterns, z-scores). Conditions that need live data with no archive (funding rates, exchange flows, composite score) are skipped and listed; the badge shows ≈ when the stat is simplified.
• Signals are causal (only data up to the signal bar decides an entry) and non-overlapping (a 7-day cooldown after each sample), so one trending week can't be counted five times.
• Below 8 signals we print "low data" instead of a percentage — a preset that fired twice does not have a 100% win rate, it has two anecdotes.

Why some presets say "live data · no backtest"

Presets like Derivatives squeeze or Multi-pillar A+ filter entirely on live derivatives, on-chain or composite data. There is no historical archive to replay them against, so a causal backtest is impossible by design — and we would rather show you no number than a fabricated one. Validate these forward: run the scan, note the candidates, track what happens.

The bias you can't remove

Our test universe is the current top coins by market cap. Coins that collapsed or were delisted are excluded retroactively — which flatters every "buy the dip" stat, because the dips that never recovered aren't in the sample. This is survivorship bias, it affects essentially every historical crypto statistic you have ever seen, and it is why the tooltip under every badge says the win rates skew optimistic. We would rather you know.