V I S O R

Strategies · intermediate · 8 min

Confluence

Confluence is the practice of requiring several signals to agree before taking a trade: a supply zone that lines up with a fair value gap, during the London killzone, on a bullish change of character. The intuition is that if four independent things all point the same way, the setup is stronger than any one of them alone. It feels like stacking evidence. Sometimes it is. Often it is the opposite, and the difference is subtle enough that it deserves its own lesson.

The case for confluence

There is a genuine statistical idea here. If two signals are truly independent and each carries real information, requiring both does raise the quality of what remains — you filter out cases where only one fired by chance. This is the same logic as asking for a second opinion: two independent doctors agreeing is stronger than one. When it works, confluence trades less often but with a higher hit rate on what it does take.

Note the load-bearing words: truly independent and each carries real information. Both are usually false in practice, and that is where confluence turns from a filter into a trap.

Trap one: the signals aren't independent

Most retail "signals" are different names for the same thing. A supply zone, an order block, and a fair value gap frequently mark the same impulse — of course they "agree," they are three descriptions of one event. Stacking them is not three independent votes; it is one vote counted three times. The apparent confluence is an illusion created by redundant indicators, and it makes a setup feel far more confirmed than the evidence supports.

Trap two: confluence is a multiple-testing engine

This is the deeper problem, and it connects straight to the Multiple Testing lesson. Every condition you add is another knob you can tune, and confluence rules are usually discovered by tuning them: you look back at the winning trades, notice they mostly happened near a zone, in a session, after a CHoCH, and you add those as requirements. Each added condition throws away the trades that would have contradicted it. Do this enough and you can make almost any past into a string of winners — you have not found an edge, you have drawn the target around the arrows.

The tell is the trade count. Genuine confluence trades rarely. A four-condition rule that still fires 200 times over five years has almost certainly been fitted, because four real, independent filters would have thinned the sample far more. And a rule that fires only eight times can't be trusted either — there is not enough evidence to tell skill from luck, as the mean-reversion verdict showed at exactly 8 trades.

How Visor keeps you honest

The robustness gates are built for precisely this failure. A tuned confluence rule tends to look spectacular in-sample and fall apart out-of-sample — the overfitting signature. And the deflated Sharpe gate explicitly penalises how many variants you tried: the more conditions you searched to build the rule, the higher the bar its result must clear to count. That is the statistically correct response to confluence-by-tuning — it does not forbid stacking signals, it just refuses to be impressed by a result you reached after trying twenty combinations.

Confluence, done honestly, means choosing genuinely independent conditions before you look, keeping them few, and accepting that a real filter should leave you with few, high-quality trades — then putting the result through the same gates as everything else. Confluence as usually practised — pile on agreeing indicators until the past looks perfect — is one of the most reliable ways to fool yourself in this entire track.

What to read next