Composite Signal Scores and Why They Decay

A score out of ten is supposed to make the decision simpler. In practice it makes the decision unexaminable, because a seven can be built from four different combinations of conditions and the number does not say which one it was. The breakdown sitting on orb trading signals mlstpodcast lists the components separately for that reason. Four trading conditions summed into one figure, say relative volume, gap size, trend alignment and the width of the opening range, produce a value that cannot be argued with and cannot be diagnosed either, which is a poor trade for the small convenience it buys.

Aggregation Destroys the Diagnosis

When results fall off, the useful question is which condition stopped working. A composite cannot answer it. The score went from seven to six, and the reason is somewhere inside an addition that was performed and discarded. Keeping the four inputs beside the total costs one line in the log and preserves the only information that will matter later.

Weights Are Almost Always Invented

Most scoring schemes weight their components by feel, and the weights then look precise because they are numbers. Volume counting for three and trend alignment for two is a claim about relative importance that nobody measured. It can be measured, by comparing outcomes when each condition is present alone, but that requires the components to have been recorded separately, which the composite prevented.

Correlated Inputs Get Counted Twice

Conditions in the same family move together, so a score can double count without anyone noticing. Relative volume during the first fifteen minutes and the width of the range are not independent, since heavy volume tends to build wide ranges. Adding both as separate points makes a volatile morning score highly on what is essentially one observation, which is exactly the kind of session where a breakout rule performs worst.

Thresholds Beat Sums

A set of independent gates is harder to fool and easier to repair. Each condition either passes or it does not, and the trade requires all of them, or a stated number of them. When results decay, the pass rate of each individual gate can be checked against its own history, and the broken one is visible immediately. Nothing gets averaged away.

If You Keep the Score, Keep the Parts

There is nothing wrong with a headline number for ranking candidates in a hurry during premarket. The failure is storing only that number. Write the components into the log alongside it, and the score remains a convenience rather than becoming the sole record. Six months on, that difference decides whether the intraday review can explain a decline or only observe one.