Eight matches is not evidence
A scan matched 8 times and price rose after 6 of them. That feels like 75%. Here is why it says almost nothing, and what to compare it with before believing it.
The tempting sentence
You build a scan: some mix of conditions on a chart. You look back and find it matched 8 times. Price was higher ten candles later in 6 of those 8 cases. That is 75%, and it is very tempting to write "this setup wins three times out of four".
The sentence is true about those 8 cases. The trouble starts when it is read as a statement about the next 8.
What a coin would do
Toss a fair coin 8 times. How often do you get 6 or more heads? There are 256 possible sequences of 8 tosses. Of those, 28 have exactly 6 heads, 8 have exactly 7 and 1 has all 8. That is 37 out of 256, or about 14%.
So a rule with no edge at all, a pure coin, shows "6 out of 8 or better" about one time in seven. If you try seven different scans, you should expect one of them to look this good by luck alone. Most people try far more than seven.
Compare with what any candle did
The coin is not even the right comparison. In a rising market, price is higher ten candles later more often than not, whatever came before.
So count the base rate. Take every candle in the same period, not only the matches, and ask the same question: how often was price higher ten candles later? Suppose the answer is 55%. Then the scan's 75% has to be measured against 55%, not against 50%. With a base rate of 55%, a scan with nothing special about it shows 6 or more out of 8 about 22% of the time. That is better than one time in five.
A match rate only means something as a gap above the base rate, and the gap has to be bigger than luck can explain.
How wide the uncertainty is
A rough rule of thumb for a percentage near 50%: the uncertainty is about 1 divided by the square root of the number of cases, in each direction.
With 8 cases the square root is about 2.8, so the range is roughly plus or minus 35 points. A measured 75% is consistent with a true rate anywhere from about 40% to 100%. With 100 cases the range is about plus or minus 10 points. With 400 cases it is about plus or minus 5.
This is why small samples swing so much. Going from 8 cases to 400 does not make the setup better. It makes the measurement narrow enough to read.
Matches at the same moment are one event
A screener looks at many markets at once. If a scan matches in 5 crypto markets in the same hour, that looks like 5 cases. It is much closer to 1.
Crypto markets move together. When the largest one drops, most of the others drop with it. Five matches in the same hour will usually all rise or all fall together, so they tell you about one moment in the market, not five separate tests of the idea.
A quick honest count: group matches by date, and count the number of separate dates. Eight matches on four dates is a sample of about four.
What to do with a small sample
Treat it as a question, not an answer. Widen the test: more history, more markets, other timeframes. If the gap over the base rate shrinks as the sample grows, it was probably luck. If it holds, that is more interesting, and still not a promise.
A larger sample does not prove a setup either. Markets change, and a pattern that showed up for two years can fade. A bigger sample only removes one specific way of fooling yourself.
The Candlesview screener shows how many matches a scan had and what price did afterwards, next to what any candle did over the same period, so the gap can be seen directly.
To take away
- A fair coin shows 6 or more out of 8 about 14% of the time.
- Measure a setup against what any candle did, not against 50%.
- Matches in many markets at the same moment count as roughly one case.
Try it
Keep reading
A backtest that made 18% sounds good until you learn the market rose 30% over the same months. The comparison with holding is the first check, not the last.
21 November 2026Testing ideas · 5 minMixing timeframes without looking into the futureA daily trend filter on an hourly chart is a sound idea that is easy to test wrongly. The rule is simple: a higher-timeframe candle only counts once it has closed.
19 November 2026Order flow and derivatives · 5 minWhat a liquidation map can and cannot tell youThe bright bands on a liquidation map are estimates built on assumptions about leverage. They show where pressure may sit, not where price is going.
18 November 2026Education only. Nothing here is financial advice or a recommendation to buy or sell anything; examples are arithmetic, not results. Trading involves a high risk of losing money.