I spent a long time tuning the wrong thing.

Like most people trading smart money concepts, I judged a setup by how it looked — how cleanly price swept the liquidity, how hard the market structure shift came, how wide the fair value gap sat. Cleaner setup, better trade.

Then I split 73 signals on BTCUSD 1h by attribute and measured each group at a 2R target.

Setup quality: +0R for scores above 50, −0.02R for scores below. Timeframe agreement: +0.24R with all three aligned, −0.09R with two. Zone position: a spread of 0.15R, which on these sample sizes is nothing.

Then the fourth column.

Stops within 1.5 ATR of the entry recorded an expectancy of +1R at a 2R target, across 15 signals. Stops beyond 1.5 ATR recorded −0.28R across 58.

Same model, same symbol, same three labels on the chart.

Why it happens

It is not arithmetic. Expectancy is measured in R, so a wide stop is not being punished for being wide — the target scales with it.

The reason is what the distance tells you about the setup. The stop sits beyond the sweep, so its distance is really a measure of how long the market took to shift after taking the liquidity. A shift arriving two bars after the sweep is a decisive rejection of the level. One arriving twenty bars later is the market wandering back and printing a similar shape on the way.

Same three labels. One is a reversal. The other is drift that happened to look like one.

The setup above

Swept low, structure shift up, entry at the gap the move left behind. Quality 79 out of 100 — high by any reading. The panel grades it low anyway, because signals sharing its profile have a negative record on this symbol. A setup can look excellent and still belong to a group the numbers do not support.

Two things I got wrong, in case you are logging trades too

I used to count every signal the model produced, including the ones where price never came back to the gap. But a fair value gap entry is a limit order. If it never fills, it is not a losing trade — it is no trade.

And when a stop and the best price of a trade fall inside the same bar, I now assume the stop came first. You cannot know the real order. Taking the favourable reading makes every backtest look better than the account does.

What I would not claim

Fifteen signals is thin. Enough to apply carefully, not enough to call settled, and it has to keep updating rather than be measured once. Across all 73 the model reached 1R on 52% of signals, 2R on 33% and 3R on 29%, with expectancy strongest at a 2.5R target.

Educational analysis, not financial advice. All figures describe recorded historical behaviour on the chart above and do not predict future results.