The confluence factors and their weights

The nine base factors

Each factor is a signed measurement rather than a checkbox, which is why factors that disagree subtract instead of simply being omitted. That is the mechanism behind a common surprise: a textbook pattern in a poor location grades in the eighties rather than the nineties, because the location factor is actively pulling against the pattern factor.

• Pattern — the setup’s own quality as detected.
• Trend — alignment with the prevailing direction on the graded timeframe.
• Multi-timeframe agreement — whether the charts above agree.
• Location — where the entry sits relative to structure.
• Extension — how far price has already travelled, penalised past fitted percentiles.
• Volume — confirmation from participation.
• Momentum — penalised beyond fitted percentiles rather than rewarded without limit.
• Structural confirmation — a recent structural event, decaying in influence as bars pass.
• Volatility regime — distance from the fitted centre of normal volatility for that market.

Weights differ by asset class

Crypto carries three additional factors that only exist in that market: funding, open interest and long/short positioning. Using them elsewhere would mean inventing them.

Forex excludes the volume factor entirely. Retail forex feeds report tick counts, and a tick count is not volume — treating it as one would import a confident-looking number with nothing behind it. That exclusion is permanent rather than pending a better data source.

Indices and commodities borrow the equity profile. That is an approximation, and it is documented as one rather than presented as a fitted result.

Redistribution

When a factor cannot be computed, its weight is given away explicitly to the remaining factors rather than left as a silent hole. The weights that actually run therefore always sum to one, which is asserted under every dropout combination at build time.

Measured versus declared

The thresholds inside each factor are fitted per asset class and timeframe from real candles. The weights between factors were declared rather than fitted until engine 1.9.0, and still are for gold and indices.

Since 1.9.0, crypto, forex and stocks grade under fitted tables. They were fitted to a stricter question than profit: whether the market played into the setup — reached the first target before the stop — more often than the same trade taken at a random moment of the same month, so that the distance to the target cancels and only the timing the analysis added remains. Under that test the declared tables were slightly worse than random, and their A grades were the worst-timed setups they issued.

The fitted tables carry negative weights: trend alignment, multi-timeframe agreement, location and structural confirmation count against a setup in every fitted market, momentum and volume in some. A negative weight does not mean a counter-trend trade is wanted for its own sake; it means that by the time every confirmation has arrived the move is mature, and the record favoured the earlier entry. A table shipped only after its ranking beat random on quarters it was never fitted on, and was checked again on bars fetched after the weights were frozen.

Frequently asked

How many confluence factors does a crypto read use?

Up to twelve: the nine base factors plus funding, open interest and long/short positioning. Forex uses eight, because the volume factor is excluded — retail forex tick counts are not volume, and treating them as such would import a number with nothing behind it.

Are the factor weights learned from trade outcomes?

For crypto, forex and stocks, since engine 1.9.0, yes — from replayed setups judged on whether the market reached the first target before the stop more often than the same trade at a random moment, and confirmed on unseen data before shipping. Gold and indices still grade under declared weights, because no fitted table has cleared that test for them. The thresholds inside each factor are measured per asset class and timeframe everywhere.