A Calibrated Bayesian Search for Potential Chemical Technosignatures in Polluted White Dwarf

Abstract: We present a meteorite-calibrated Bayesian framework for searching archival abundance records for chemical technosignatures--operationally, compositional patterns better explained by an idealised "processed" template (endmember) than by the empirical distribution of natural rocks. We fit a multi-modal natural-composition reference using 3,493 whole-rock meteorite analyses, and for each of 697 star-paper abundance sets--spanning at least 397 distinct objects once Gaia-designated repeats are consolidated--we compare the Bayesian evidence for (i) natural material and (ii) a mixture of natural material with a fixed siderophile-enriched template, parameterised by a Ca-normalized mixing fraction alpha. Strong support for the processed template is uncommon: in the photospheric compilation (atm) 8/697 records have BF > 10 (4/697 have BF > 100), and in the diffusion-adjusted steady-state subset (acc ss; 148 records spanning at least 94 objects) 6/148 have BF > 10. We report the highest-evidence candidate records and infer the fraction of records detectably favoring the mixture model, with posterior medians pi-tilde = 0.011 (atm) and pi-tilde = 0.041 (acc ss). We calibrate the analysis with end-to-end injection-recovery experiments matched to each record's coverage and censoring. The calibration shows that discrimination is driven mainly by chemical information, typically requires greater-than-or-similar-to 5 detected elements for decisive support, and--for the siderophile template--is strongest for exact five-element panels that include Fe, Mg, Cr, and Ti together with Ni, Si, or Na. These results constrain the detectable incidence of the tested processed-composition class in current data and set observational requirements for future multi-element surveys and expanded template families.