A composite score mixes readings that have a sign with one that does not. Direction and momentum can be negative or positive. Trend strength cannot. ADX runs from 0 to 100, and a steady slide down reads exactly the same as a steady climb up. Put that number into a weighted sum next to two signed readings and every score on the sheet gets nudged upward, whatever the market is doing. A worked example Take a bearish moment: direction at -40 and momentum at -20 on a scale of -100 to 100, ADX at 30, and weights of 0.4, 0.3 and 0.3. With ADX added as it comes: 0.4 x -40 + 0.3 x -20 + 0.3 x 30 = -16 - 6 + 9 = -13. Now the mirror image, a bullish moment with direction at 40, momentum at 20 and the same ADX of 30: 16 + 6 + 9 = 31. Same market, flipped over, and the scores are 31 and -13. If a filter only lets through scores of 30 or more in size, the bullish case passes and the bearish one is dropped. Nothing in the market caused that. The arithmetic did. The fix is one multiplication Give the strength term the sign of the direction reading. With direction at -40 the strength contribution becomes 0.3 x -30 = -9, and the bearish score is -16 - 6 - 9 = -31, the exact mirror of 31. When direction is exactly zero the sign is zero and the strength term adds nothing, which is right: a strong trend with no known direction should not vote for either side. Another route is to build the strength term from the difference of the two directional lines that ADX itself is calculated from, which comes with a sign of its own. A mirror test you can run in five minutes Before trusting any weighted score, feed it a pair of opposite inputs. Negate direction and momentum, leave ADX alone, and compare. The score must change sign and keep its size. If the two numbers are not the same distance from zero, some term is not symmetric, and the sheet will over-report one side. A second check needs no formulas. Run the logic over a stretch of random-walk data with no drift and count how many bullish and bearish flags it raises. A symmetric method gives roughly equal counts, apart from noise. A method with an unsigned term leaves a surplus on one side that does not go away as the sample grows. Why it hides so well On a real chart the lean is easy to miss, because markets trend both ways over time and a surplus of bullish flags looks like the mood of the market. It only shows up when you compare mirrored inputs or a symmetric series, and by then the numbers have usually been read for a while. Signal Compass Desk builds its strength reading this way: ADX multiplied by the sign of the direction reading, so opposite inputs give opposite scores of the same size.
