Reciprocal pairs remove the pooled before-after signal at a matched M6.5 threshold in global ComCat, 2001-2024

We tested how event selection affects pooled counts before and after earthquakes. We used a fixed global ComCat snapshot and 1,093 centers with preferred magnitude M6.5 or more in 2001-2024. Neighbors were within 200 km in three dimensions and at time gaps of 1 to below 30 days. At the matched M6.5 neighbor threshold, 154 reciprocal pairs gave 154 before and 154 after counts. The pooled difference was exactly zero. Each pair contributed once to each side. With M5.5 neighbors the difference was +929, while with M7 neighbors it was -30. We give an exact pair identity and separate reciprocal, unmatched and calendar-edge terms. Tests of distance, the first day and an upper center magnitude limit show where cancellation holds or fails. This extends an earlier campaign warning with a count decomposition and an independent audit of all 42 settings. It is a method benchmark, not a new discovery of the cancellation principle. Mixed magnitudes, catalog errors and all-center selection limit physical inference. A zero under equal filters is not evidence of no interaction. Code, aggregate counts, provenance, a figure and reproduction instructions are included. External event records remain outside the baseline package.

earthquakes · ComCat · temporal asymmetry · pair cancellation · magnitude selection · reproducibility