Correlation QC common mistakes

Mistake	Fix
Correlating linear |S21|	Convert to dB first (20*log10(\|S\|))
Skipping x alignment	Interpolate or trim before Pearson r
Including the diagonal in the score	Exclude j == i — the diagonal is always 1 and inflates every score, and the smaller n is, the worse the inflation
Averaging first, then comparing to the average	Pairwise scores, then average the survivors
Declaring outliers from min/max summaries only	Summaries can hint; r needs the samples
Mixing S21 and S11 in one matrix	One trace type per run
Mixing lots / temperatures in one “batch”	Filter parameters first; QC one family
Deleting outliers from the project	Tag them; they are the audit trail
Publishing an average of everything	Average non-outliers only
n = 2	Need ≥ 3 traces

Shape screens and limit lines are complementary. A trace can pass correlation and fail a −3 dB insertion-loss spec.