Sapien Engineering

Killing a shop-floor exploit with a mirror, not a lockout

A manufacturer's workers were gaming a commission scheme through a gap in their shop-floor software. We stopped it in a day, not by locking the system down, but by making the gap visible on a screen.

Anonymized case study. A manufacturer; name and shop-floor system withheld. Small scope, pure judgment. Engagement led by Ryan Luccioni, principal.

The situation

Machinists state per phase Shop-floor system job tracking Status board live, color-coded Engineers update state API read engineers walk over the moment someone games the wait state
The accent-outlined box is the whole fix: a plain screen that makes hidden state visible.

Between two phases of a job, a worker had to signal "waiting," an engineer had to come approve, and then work continued. Two things were going wrong. Engineers were slow to come over, and the sharper workers had found an exploit: after approval, they'd keep themselves in the "waiting" state while actually working, then flip to the next phase and immediately back to "waiting," making a real phase look like it took two seconds. That dropped their average phase time, which inflated their commission. The cheating worked because it was invisible.

The fix

The reflex fix is a lockout: enforce the state transitions, validate everything, block the workaround. That's a multi-week project, everyone hates it, and it breaks the honest workers along with the cheaters.

Instead: a service account on the shop-floor system's API and a plain status board, literally HTML and JavaScript, that shows every machinist's live state, color-coded, on a screen the engineers can see. Nothing about the underlying system changed. The gap just stopped being invisible. Engineers now walk over the moment someone's sitting in "waiting" while clearly working, and the behavior stopped.

Why it's the right fix

The problem was never technical. It was an incentive being gamed through a blind spot. So the cheapest correct move targets the blind spot, not the software: make the state visible and let the existing incentive correct itself. A lockout would have cost weeks and punished everyone; a mirror cost a day and punished only the thing that depended on the dark.

How we knew it worked: it was the first automation of ours that made a group of people annoyed with us. Make a hidden thing visible, and whoever was using the dark gets loud. That's the mechanism working, not a side effect.

Outcome

← All work