Supplementary Data for: Spatial Rule-Based Triage for CNN-Assisted Gram Stain Classification

Three supplementary tables providing S1 — CNN training configuration and optimised hyperparameters, including backbone architecture, loss settings, training protocol, and software/hardware environment S2 — Full spatial rule engine parameter configuration selected by threshold sweep, covering context window geometry, confidence tier thresholds, slide eligibility criteria, and per-rule firing thresholds S3 — Per-rule patch outcome counts from the spatial rule engine, reporting patches affected, improved, and worsened for each rule across the test set