Every parameter is a decision you made from data, and every decision from data can be fitted to noise. Counting them honestly is the start of any robustness work: a 20/100 moving average crossover with a 2 ATR stop and a 5% volatility target has four parameters, not one.
Parameters are not evil; they are unavoidable. The question is whether the result depends on the exact value. A rule that works for lookbacks of 40 to 120 is describing something real; a rule that only works at 63 is describing your sample.
When you count, include the invisible ones: the universe you picked, the start date, the bar size, and the fact that you tested this idea after seeing the last three years of price action.
Related: parameter-sensitivity, parameter-plateau, degrees-of-freedom, overfitting