Why three agreeing indicators aren't always three votes

Pairvue scores a setup by counting how many of the indicators we run actually agree on a direction. More agreement, more confluence, a stronger read. That part hasn't changed. What we fixed recently is a quieter problem: not every indicator that agrees is telling you something new.

RSI, Stochastic and Bollinger Bands are three different formulas. They're also, most of the time, reacting to the same thing — price has moved too far, too fast. When all three flash the same direction, that isn't three independent opinions lining up. It's mostly one signal, said three ways. Counted at full weight each, that inflates confidence in a setup that's really only been confirmed once.

Grouping the basket into families

We run 12 indicators today, and we sorted all of them into five families based on what they're actually measuring, not what they're named:

• Trend — EMA, EMA Ribbon, ADX
• Momentum — RSI, Stochastic, MACD
• Volatility — ATR, Bollinger Bands
• Volume — Volume
• Market Structure — Support & Resistance, Chart Pattern, Candle Pattern

The families themselves were the easy part — standard technical-analysis groupings, mostly uncontroversial. The harder question was which pairs inside a family were genuinely redundant versus just neighbors on paper. So we measured it, rather than guessed: we logged every indicator's vote across thousands of historical windows and checked how often pairs actually agreed with each other. Bollinger Bands and Stochastic turned out to agree essentially every time we looked. RSI and Bollinger Bands weren't far behind. Volume and ATR — a momentum-adjacent pairing that isn't obvious from their names — showed the same pattern. Those are the pairs we now treat as an echo, not two votes.

What actually changed

When multiple indicators inside the same family agree on a setup, only the strongest one counts at full weight. The rest still count — they're not thrown out — but at a discount, because they're mostly confirming something the top vote in their family already said. An indicator in a different family agreeing carries no discount at all, because that's a genuinely separate read on the market: a trend indicator and a momentum indicator lining up is real, independent confluence in a way two momentum indicators lining up with each other isn't.

You won't see a new screen for this — it's not a display feature, it's a correction to how the confluence count itself is computed. The visible effect is that the score behind a setup better reflects how many independent things actually agree, not how many indicators happen to.

Why we're not calling this "improved accuracy"

We tested the change against historical data before shipping it, the same way we test everything. The direction held up — discounting redundant agreement measured better than either not discounting it at all or discounting it so hard it stopped counting — but the exact amount of discount is a judgment call we're not pretending is precision-tuned from a single small test. We picked a middle value on purpose rather than chase whichever number looked best on the data we happened to test it against, because that's exactly the kind of overfitting we've written about avoiding before.

So: this is a real fix for a real problem — redundant indicators were quietly inflating how confident a setup looked — not a claim that signals just got some specific percentage better. We don't have a number we'd stand behind, and we're not going to invent one.