Method & provenance. site records only: camera-observed events, verification ledger, staffing calendar; ground truth structurally unreachable complexity adjustment: indirect standardization - expected catches = Σ facility per-prep-type rate × technician case mix first-pass = preps with zero alerts of any kind (incl. false alarms and voluntary redraws: a strict, honest definition) company goals shown are demo defaults (first-pass ≥ 96.0%, assists ≤ 2.5/100 preps, rework ≤ 1.0%) - set per site policy complexity mapping: High = PN/complex compounds · Medium = reconstituted/multi-vial · Routine = liquid/RTU/syringe
Mock 6-month study (2026-04-05 → 2026-10-05, 3 hoods, alerting world). Recognition metrics are measured, never invented: flawless = zero alerts + clean verification; streaks are consecutive flawless preps in timestamp order; clean double-checks = independent second-pharmacist checks on high-alert preps that passed. Volume is deliberately excluded from goals.