Search protocol, screening log and bibliometric analysis for "Beyond identity: biometric traces as multi-attribute probabilistic evidence"

Supporting materials for the manuscript Beyond identity: biometric traces as multi-attribute probabilistic evidence. These files exist so that the manuscript's negative claim can be checked. The paper argues that four biometric research threads barely cite one another. A negative claim of that kind is worth only as much as the corpus it is computed on, so the corpus, every screening decision and the analysis code are deposited here. search_strings.csv — the 28 verbatim Crossref queries, with fields, filters, page size and page ceiling. screening_log.csv — the primary artefact: one row per screened record, 12,974 rows, each with its include/exclude decision and reason. recount_from_log.py — regenerates the PRISMA-ScR record flow from the log, so every count in the paper can be verified without trusting either the paper or the pipeline. It currently reproduces every cell of the manuscript's Table 3 with zero differences. citation_verification.json — machine verification of all 210 references against Crossref, DataCite, OpenAlex and Semantic Scholar: 200 registry-resolved, 10 grey-literature items cleared by a live official URL, 0 unresolved, 0 DOI mismatches. The seven screened corpora, the cross-thread citation counts, the negative-control corpus pairs, the three screening specifications, and the full pipeline in run order. Version 2 replaces the screening log. Version 1 carried a log inherited from a superseded pipeline that fetched Crossref twice with two query sets — the defect unified.py was written to remove after peer review found the paper reporting two irreconcilable cross-thread counts for one specification. That log held 12,975 rows and differed from the manuscript's table in five cells. Version 2's log is emitted from inside the screen itself, by the same call whose include/exclude decision the analysis acts on, so the log and the table are two views of one decision rather than two computations that happen to agree. The superseded summary file has been removed rather than kept alongside. A withdrawn result is included deliberately. An earlier version of this analysis reported a normalised observed/expected fragmentation ratio against a degree-preserving null. That estimator did not survive its own controls and the manuscript withdraws it; the raw counts, which do survive, are what the paper reports. controls.json records the controls that defeated it, because a withdrawn result is part of the evidence. Data are CC BY 4.0; code is MIT. See README.md and LICENSE.