The problem with P&L
A profit and loss curve is an outcome, not an explanation. It tells you that the last two months went well. It cannot tell you which of the six things you look for was responsible — or whether you'd have done just as well ignoring three of them.
This matters because your confluences are not free. Each one you insist on filters out trades. If a confluence isn't earning its place, it's costing you setups for nothing. And if one of them is genuinely carrying your results, you want to know so you can weight it more heavily and stop taking setups without it.
What you actually need to measure
To attribute performance to a confluence, you need three things recorded per trade:
- Which confluences were present, recorded before the outcome was known.
- The outcome — R is usually more useful than raw dollars, because it normalizes across position sizes.
- Enough trades that the split isn't noise.
The first is where nearly everyone falls down. If you tag a trade after seeing the result, you'll unconsciously tag the winners more generously. The record has to be made at the moment of decision or it isn't evidence.
The basic read
Once you have the data, the first pass is simple: for each confluence, compare the trades where it was present against your baseline.
A confluence that's worth keeping shows a meaningfully better win rate or expectancy than your overall average. One that tracks your average is neutral — it's along for the ride. One that's below your average is actively costing you, and it's usually one you'd have sworn by.
Look at expectancy rather than win rate alone. A confluence that wins less often but produces much larger winners can easily be your most valuable one, and win rate will hide that completely.
Four traps that make the numbers lie
Sample size. A confluence present in eight trades tells you almost nothing. Resist reading a 75% win rate off six setups — that's four wins.
Correlation between confluences. If you only ever take order-block entries during the London killzone, you cannot separate the two. They'll show identical numbers because they're the same trades. Untangling this needs setups where one is present without the other, which means occasionally taking a setup that breaks your usual pattern — or accepting that the two are measured as a pair.
Survivorship in your own records. If you only log the trades you take, you learn nothing about the setups you passed on. The no-trades are data too: a checklist that keeps you out of losers is working, and you can only see that if you record the ones you skipped.
Hindsight edits. Going back and adding a confluence you "definitely saw" after a trade wins is the fastest way to turn a dataset into a flattering story. If you do correct a record, correct it because you misrecorded it — not because you know the outcome.
From measurement to stacks
The single-confluence view is where you start, not where the value is. What usually matters more is which combinations outperform: a liquidity sweep on its own may be unremarkable, while a sweep plus displacement plus a discount entry is where your real edge lives.
Combinations need more data than single factors, so this is a question for a few hundred setups rather than a few dozen. But it's the question worth building toward, because it's the one that describes your actual edge rather than its ingredients.
What to do with the answer
Three honest responses to the data:
- Promote what pays. Weight it more heavily, and treat its absence as a reason to pass.
- Demote what doesn't. Drop it to optional, or cut it. You'll take more setups and lose nothing.
- Leave the ambiguous alone. Most of your list will sit in the middle with too little data to judge. That's a normal, honest result — not a failure of the method.
The goal isn't a perfect checklist. It's a checklist that gets less wrong every quarter, edited by evidence rather than by whichever video you watched last.