What
The cleanest answer is to stop letting any single signal close a fleet maintenance task. Build a sequence in which three separate events must all land before the job is considered done. Start with the geofence. Set a configurable minimum dwell time, meaning the vehicle must remain inside the service area for a defined stretch before the geofence alert is even eligible to contribute to a completion status. A vehicle that rolls past a bay or pauses briefly in a yard will never satisfy that threshold. This single change removes a large share of false positives, especially in crowded depots where vehicles pass service areas all day. Layer two is the ignition-off signal. Capture it only while the vehicle is still inside the geofence, and treat it as corroborating evidence, not proof. A driver who parks and kills the engine has clearly stopped, but stopping is not repairing. Keeping this event separate in the compliance log matters later, because it gives your audit trail a defensible timestamp that does not depend on a single location fix. Remember that some OBD-II GPS trackers sample engine runtime roughly once a minute, so short idle windows after service may not register at all. Design your rules with that limitation in mind rather than assuming every second is captured. Layer three is manual driver confirmation. The technician or driver who actually performed the work signs off against the specific service order. Tie that confirmation to the work order number, not to the location. This is the step that separates real service from a vehicle simply sitting in the right place. If parts were unavailable and no work happened, the location and ignition events still occur, but the manual layer stays open. That is exactly the behavior you want. Now combine the three layers into one sequential gate so no single input can flip the task to complete. The order should be geofence entry with minimum dwell time satisfied, ignition-off captured inside that same geofence window, and manual confirmation recorded against the service order. All three must be present and time-ordered. If any one is missing, the task stays open and surfaces on a discrepancy list. Then run a daily morning review before dispatch. Compare the compliance log against the manual confirmation record and look for three things: tasks marked complete with no manual sign-off, manual sign-offs with no matching geofence or ignition event, and events whose timestamps drift apart by hours rather than minutes. That review turns silent gaps into visible exceptions you can act on before the day starts. For the audit trail itself, record who confirmed the work, when they confirmed it, what work was performed, and the identifiers of the geofence and ignition events the confirmation is attached to. That set of fields is what lets you reconstruct a service event months later without guesswork. Threshold tuning comes next. If tasks are being marked complete for vehicles that only passed nearby, raise the minimum dwell time. If legitimate short services keep getting flagged as unverified, lower it slightly. Adjust in small increments, watch the discrepancy count for a week or two, and stop tuning once false positives and false negatives both settle at a level your team can manage by exception rather than by hand. There is a point where tuning stops helping. If your platform cannot accept a manual confirmation step alongside automated GPS telemetry, no threshold change will close the verification loop, because the workflow simply has no place to store human sign-off. That is the breakpoint where internal fixes stop working. Finally, protect the manual layer with accountability and routine training. Under time pressure, drivers skip sign-offs. Make confirmation part of the closeout routine, check it in the morning review, and treat repeated omissions as a process issue rather than a personal one. Verification only holds when all three layers stay in place.