Regional stock status inference shaped by source dependence in inland fisheries

Abstract Freshwater fisheries evidence databases often contain many extracted records from a single publication, yet regional syntheses usually treat each row as independent evidence. Using a verified dataset of 279 records from 61 sources on West African inland fisheries, we test whether the choice of counting unit changes regional stock-status inference. Only 115 records from 27 sources carried ecological classifications, and record concentration reduced their effective source number to 16.2. Counting records, fully exploited fisheries dominated (45.2%); giving each publication equal total influence, overexploited classifications dominated (39.9% vs 29.1%), although cluster-bootstrap intervals overlapped [paired difference: insert estimate and interval]. Combined collapsed and overexploited classifications rose from 41.7% of records to 63.9% of source-balanced profiles. In simulations, multiplying rows in one category inflated record-weighted error and wrong-dominant selection (99.4% at a tenfold ratio), whereas equal-source estimates remained stable; this follows by construction and illustrates the mechanism rather than independently validating it. Indicator reporting was sparse (66.1% of complete records reported none of four indicators), and 28 of 35 mapped locations rested on a single source. Source-balanced estimands and explicit reporting of source depth are a simple safeguard for evidence syntheses with clustered observations.