Local shuffles left uncertain tensor similarity excess for shallow Mw 5.8 or larger GCMT pairs within 100 km and 1-7 days in 1982-2019

We tested full moment tensor similarity for shallow Global CMT earthquake pairs in 1982-2019. The fixed sample has 5,821 events with scalar-moment Mw at least 5.8 and centroid depth below 70 km. The primary set has 960 pairs within 100 km and 1 to below 7 days. Mean signed tensor cosine was 0.5684. A shuffle within 2-degree cells, depth and magnitude groups, and five-year periods gave an exact mean of 0.5519. The excess was 0.0165, with an upper-tail Monte Carlo p value of 0.0700 and a conditional three-year block interval of -0.0143 to 0.0551. One-degree cells changed the excess to -0.0109; one-year periods gave -0.0125. The planned 50 km check gave an excess of 0.0594 and p = 0.0005. An added check with both 50 km pairs and one-degree cells reduced that excess to 0.0110, with p = 0.1540. Equal-event weights also changed the primary evidence. These tests show sensitivity to local reference groups and pair weights. They do not establish no temporal relation, causal triggering, or stress rotation. Sparse shuffle groups, fixed spatial cells, sequence dependence, and unmodelled catalog error limit inference. The package gives exact selected source records, pair data, all draws, code, figures, an independent audit, and offline repeat instructions.

earthquakes · Global CMT · moment tensor similarity · conditional shuffle · spatial mixture · null result · reference sensitivity · reproducibility