One Event, Many Records: Provenance-Consistent Inference from Gravitational-Wave Catalogs

Abstract Scientific archives can contain several valid analyses of one physical event. Combining those records as independent observations changes a likelihood even when each record is correctly cited and versioned. We develop a provenance-consistent representation that separates the physical event, the detector data and the analysis product, and test it on frozen public GWOSC catalogue snapshots. GWTC-1-confident, GWTC-2.1-confident and GWTC-3-confident contain 11, 54 and 35 records respectively, but their union contains 90 distinct canonical event names rather than 100 events. Ten events appear in both the first and second snapshots, with revised primary-mass summaries for all ten. These counts describe catalogue membership, not a new estimate of the number of independent astrophysical signals. A separate Gaussian experiment uses the observed multiplicity pattern to quantify the consequence of treating repeated records as independent. A nominal 95% interval then has 92.64% analytic coverage, and a row-weighted estimator has an effective sample size of 83.33 despite containing 100 records. A singular-covariance formulation preserves the original likelihood under exact record replication; its quadratic form agrees with a direct event-level calculation to (5.68 \times 10^{-14}). Additional controls show why arbitrary deduplication can select stale analyses, why prior correction alone does not repair repeated use of the same data, and why record-level test splits can leak physical events. The resulting registry preserves useful analysis alternatives while making their evidential relationships explicit. Keywords: gravitational-wave catalogues; scientific provenance; repeated evidence; likelihoods; catalogue versions; reproducibility.