May and June can share an average while differing in run coverage and measurement completeness. Equal averages do not show equal conditions or performance.
Four comparison areas
| Area | Records to review together |
|---|---|
| Spread | Run-level readings and median |
| Tail events | Count outside the limit and longest estimated duration |
| Record coverage | Measured, gapped and missing runs |
| Mix | Route, product and run type distribution |
Check comparability first
Put run counts and measured runs side by side. If route or product mix changed, explain that with the result; do not compare raw averages alone.
A data gap is not performance. Do not treat an unmeasured run as normal. Count it separately; match the device file, download record and run schedule to investigate when the gap began.
Two distributions, one average. In a hypothetical example, two months can have the same average; one may have clustered readings while the other has a few high readings and many low ones. Median and spread make the difference visible.
Season and route effects. Comparing the same month across years may reduce seasonal effects; still record route, load and product mix. This matching alone does not prove what caused a change.
A concise comparison note. For each month, use the same format for run count, measured runs, gaps, event count and longest estimated duration. Qualify the result if data coverage differs.
Next step
If a difference clusters on one route or product group, review those run records. Product assessment or action depends on product requirements and the authorized quality process.
