How to read a scan's hit rate

Every scan page here has a panel that replays the query across the last 250 sessions and reports what its matches did next. The panel answers one question — has this setup been worth a look? — but only if you read it correctly, and the two most common misreadings both flatter bad scans.

What the columns measure

For each past session where the scan matched, the replay records the close-to-close return over the next 1, 5 and 20 trading days, then aggregates. Three windows because setups have natural lifespans: a hammer at support is a claim about the next few days, a golden cross is a claim about the next few months, and grading either on the other's clock produces noise. Read the column that matches how long you actually hold.

Misreading one: hit rate without payoff

A 65% hit rate sounds like an edge. It is not — not until you know the size of the wins against the size of the losses. A scan that is right 65% of the time for +1.2% and wrong 35% for −3.1% loses money at a hit rate that sounds excellent. The inverse is just as real: breakout scans like the 52-week high breakout often show hit rates near or below 50% and remain interesting, because the setup's premise is a few large continuations paying for many small failures.

So read the pair, never the percentage alone: how often it worked, and how much working and failing each paid. A scan is only as good as that product.

Misreading two: the average without the count

The other trap is a beautiful average on eleven matches. Eleven matches in 250 sessions means the scan fires once a month; one lucky cluster — a single result week, one index rally — can carry the whole average. Before trusting any number in the table, look at how many signals produced it. A modest average across three hundred matches says more than a spectacular one across a dozen. High-frequency scans like volume shockers sit at the other pole: plenty of signals, weaker per-signal meaning, and the table shows you that trade-off instead of letting you assume it away.

What this replay deliberately is not

The panel calls itself a sketch, and the word is chosen. Close-to-close returns assume you entered at the close the signal appeared — plausible for an end-of-day workflow, generous for anything else. There are no transaction costs, no slippage, no position sizing. And universe membership is measured as of today, which leans survivorship-positive: a stock that cratered out of the liquidity tier since is not in the replay to hurt the numbers.

None of that makes the sketch useless. It makes it a filter: a scan that cannot look good under these friendly assumptions has nothing to offer the unfriendly ones, and you have found that out in ninety milliseconds instead of ninety trades. The longer walkthrough covers the method end to end.

The habit worth building

Before acting on any scan — the prebuilt ones here, something translated from Chartink, your own idea at eleven at night — open the panel and ask the three questions in order. How many signals? What did the holding window I actually use pay? Does the payoff survive the hit rate? Thirty seconds, and it is the difference between screening and folklore.