A new Slop University research poster measures how much a pharmacy's mandatory human pill count adds beyond its dispensing robot's own fill-accuracy log, and finds the added value concentrated in a single failure type.
Across 42 pharmacy sites, catches beyond the dispensing robot's own log cluster almost entirely in one dosage-strength blind spot
Every pill a pharmacy’s dispensing robot fills is already weighed against what the prescription called for and logged automatically, before a human ever touches the bottle. Staff still count the pills again by hand before handover, on top of that log. The School of Emergent Priorities set out to measure how much of the second count’s value the first check hadn’t already banked.
Associate Professor Kwame Lindqvist and Associate Professor Casimir Beng matched robot-log and human-count records across 42 pharmacy sites over a ten-month window, coding every flagged discrepancy by error type and by which check caught it first. The human count adds relatively little on its own — a handful of percentage points beyond what the robot’s log already flags — but nearly all of that thin margin sits in one place: pairs of dosage strengths that look different on the shelf and weigh the same on a scale.
What surprised us wasn’t that the two checks overlap almost completely. It’s how neatly the overlap breaks down exactly where it doesn’t.
— Associate Professor Kwame Lindqvist, School of Emergent Priorities
“A system built to catch everything gets treated as though it does, right up until someone finally checks which parts of it are actually doing the catching,” said Associate Professor Casimir Beng, Lead of the Adaptive Metrics Lab.
The School of Emergent Priorities reads the result as part of a broader pattern in its work this year: how rarely a uniform safeguard’s value turns out to be spread evenly across everywhere it’s applied.
The full poster is available from the University’s research repository under an open licence, doi:10.5555/slop.1ruz61.