Stable fractions for each depth label permit either sign of the ComCat shallow-share change between 2000-2012 and 2013-2025

We tested how the group used for a stable depth assumption affects a shallow-share comparison in 12,848 global ComCat events with preferred M >= 5.5. We reused a published snapshot and compared 2000-2012 with 2013-2025 at depth <= 15 km. Treating exact 10, 33 and 35 km records as one group with a stable unknown shallow fraction gives a change of +1.43 to +7.62 percentage points. Stable fractions for each label instead permit -3.63 to +12.68 points. The labels change in composition. A continuous drift of 4.31 points removes the positive lower bound in the pooled model; a finite witness uses 86 late flagged records. All bounds assume correct classifications outside the three labels. These are deterministic sensitivity results, not measured true depth changes or confidence intervals. Source errors, mixed origins and magnitudes, selected periods and untested stability assumptions limit inference. The package includes code, aggregate counts, bound witnesses, a drift curve, provenance and a separate count audit. External event records remain outside the publication.

earthquake depth · ComCat · partial identification · sensitivity analysis · catalog bias