Data supporting an auditable sparse-evidence workflow for phage receptor-binding-protein prioritization and Gp17-mediated magnetic capture of Yersinia enterocolitica O:3

Abstract

Version 3 expands the existing dataset to Figures 1–7 and Supplementary Figures S1–S2, with available sources, current captions, provenance mappings and SHA-256 hashes. V2 history is preserved. No experiments were added, and the frozen 827-candidate analysis, G_pre1/G_pro1, weights, random seed and released software are unchanged. The dataset supports an auditable workflow for phage receptor-binding-protein prioritization and Gp17-mediated capture of Yersinia enterocolitica O:3. It includes frozen rankings, component ablations/baselines, 10,000 weight perturbations, identifier-masked and strict context-blind rankings, S4/S5A/S5B, the verified-evidence registry, ELISA and magnetic-capture records, and processed-matrix/retail-sample table transcriptions. Only Q9T0Z9/Gp17 has complete in-house wet-lab evidence. Ranking robustness evaluates internal stability and implementation reproducibility, not independent predictive accuracy or generalization. Figure 3A's current full-precision derivative uses the same 36 technical wells, the unrounded PBS mean and the existing one-site model. Display remains 44.38 ± 2.98 μg/mL (estimate ± SE), R² = 0.9904. Original workbooks and historical rounded results are retained. Figure 3B uses raw-record-supported 0.1860; historical Prism 0.1855 remains archived with derivation unestablished. Numerical and plot-geometry comparisons are separate evidence; geometry does not prove byte-identical export inputs. The new derivative is outside the frozen software release's 18 historical tests/51 output hashes. Source limits are documented: Figure 2 lane labels derive from historical assemblies, not an authenticated loading ledger. Figure 4 scales are 5 μm in A–C and 10/5/5/10 μm in D(a–d); calibration does not verify specimen identity or unrecorded control blocking. Figure 5's 87.2% is a historical count index without recoverable volume normalization or independent-repeat definition. Figure 7 uses M+6 sample lanes; the wider seven-lane source is retained, but the seventh identity and omission reason are unrecorded. S1/S2 map to assembly components; independent acquisition originals and sample keys remain unavailable. Figure 7 is qualitative low-input evidence, S1 pilot qualitative concordance, and S2 amplified-readout compatibility, not a formal detection limit. See 06_image_manifest/FIGURE_COVERAGE.csv, V3_FIGURE_PROVENANCE_STATUS.md and Figure3_source_precision_conclusion.md, plus 00_README. Software remains separately archived: https://doi.org/10.5281/zenodo.22625899, v1.0.0, commit 604ee08829ac5ee628f4aa0c98fbb5a21e3c3e17. Dataset licensing does not replace software/third-party licences.