A new Slop University paper reports a supermarket loyalty chain's pre-registered attempt to replicate a capability-drift index that once forecast an exam-integrity workaround, finding no signal in the new setting and no change to the automatic promotional-terms clause built on the original claim.
A pre-registered replication finds no lead time for a transplanted assessment-integrity forecasting tool, and no change yet to the loyalty terms clause built on it.
An index that once told the University when an exam-integrity workaround was coming has spent the past thirty weeks telling a supermarket chain nothing at all, and the chain has kept using it anyway.
A new paper from the School of Continuous Improvement and the School of Emergent Priorities, led by Lecturer Marit Osayande with Senior Lecturer Torun Ezeigwe and Postdoctoral Fellow Petra Umbile, re-ran the index unchanged on a discount-grocery chain’s loyalty-app data, testing whether it could anticipate shoppers finding a way to stack a manufacturer’s coupon with the chain’s own loyalty discount at the self-checkout. The pre-registered test, run across 46 store clusters over thirty weeks, found no stable lead time at all.
What we’ve shown is where the index stops working. What comes next is working out why any of us expected it to keep going.
— Lecturer Marit Osayande, Convenor, Living Dashboard, School of Continuous Improvement
The method translates a campus tool into supermarket terms: loyalty-app feature adoption for exam-tool capability, a flagged checkout override for a reported workaround. Neither substitution moved the needle; an unregistered follow-up found the override log tracking something else instead, how recently a store’s self-checkout kiosks were installed.
The University counts a transplant that fails cleanly among its more useful contributions to the field, on the reasoning that a false lead ruled out is still ground worth clearing. “A finding that only works once is still data,” said Dr Petra Umbile, one of the index’s original authors. “It’s just data about the finding, not the domain.” The chain’s own promotional-terms revision clause, adopted on the strength of the original finding, has not been amended.
The full paper is available from the University’s research repository under an open licence, doi:10.5555/slop.nhgtil.