Override Preservation Pilot
A controlled trial of whether removing throughput pressure restores meaningful human override in a deployed decision-support system — with the result published either way.
Our recurring finding is that formal override rights are preserved while the practical ability to exercise them is not: staff may disagree with a system, and are measured on throughput that disagreement reduces. Override rates fall towards zero and the institution records this as alignment.
The hypothesis
If override decline is driven by workload pressure rather than agreement, then removing the pressure should restore it. If override remains near zero once time pressure is removed, the mechanism is something else — deference, interface design, or genuine agreement — and our recommendations to date are wrong in an identifiable way.
Design
- A partner institution running a deployed decision-support tool, with baseline override data.
- A defined cohort of staff whose throughput measure is suspended for the trial period.
- Measures agreed in advance: override rate, override accuracy on subsequent review, and decision time.
- A stopping rule defined before the trial begins.
- Publication of the result whether or not it supports our position.
Status
In development. We are in discussion with prospective partner institutions and are not yet naming them. A recommendation that has never been tried inside a real institution is a hypothesis, and we would rather test ours than repeat it.
