One row per surface, one cell per question. The 14 outlined cells leave the denominator entirely rather than counting as answers you were absent from — counting our outage as your bad result is the easiest way for a tool like this to lie. A zero on a row that answered is a finding; a zero on a row that failed is nothing at all.

Two fictional readings, each from 20 questions, differ by 10 percentage points. Their illustrative ranges overlap. Overlap alone does not prove that performance is equal or that a change is real. Check the original answers, usable counts, failed readings and collection conditions before drawing a conclusion. The chosen questions are a limited sample, not a random sample of all buyers.