Define the window
I define the observation window and the comparison rule before I look at what happened next. A window can describe returns, price relative to pivots, the order of level crossings or a combination of structural features.
Compare like with like
Normalisation can make different price scales comparable, but it can also discard information. Document whether the method preserves volatility, elapsed time, range and market regime. A visually close match may be economically different.
Separate matching from outcomes
Only data available at the evaluation cutoff should influence match selection. Examine subsequent prices after selection. Changing the comparison after seeing the outcome introduces hindsight bias.
Inspect the distribution
Show several matches, their dates and the spread of subsequent paths. A central path hides disagreement. Repeated or overlapping windows can overstate the effective sample size.
Challenge the interpretation
Compare against a simple baseline and repeat the evaluation on unseen periods. Report how sensitive the results are to the window length, threshold and available data.