What
Start with one flagged event, not the whole fleet. Pick a refuelling record that looks wrong, then pull the three raw streams that sit behind it: the GPS tracking trace, the engine state log, and the fuel-level telemetry. Do not read the summary report yet. Summaries are already averaged, and averaging is exactly what hides the problem you are trying to find. Work from the raw timestamps first, then build a single reconciled timeline where every entry carries the same clock. This is the practical heart of fuel monitoring cost control, because a fuel number is only as trustworthy as the timeline it sits on.Position comes next. Does the GPS trace place the asset inside a known fuel location boundary at the moment the level rose? Compare the rise against your geofence alerts for that yard, depot, or station. If the level climbed while the trace shows the vehicle on a highway, either the boundary is drawn wrong or the event belongs to a different asset. Watch for signal latency here, since a delayed alert can fire after the vehicle has already left, pulling the event into the wrong location window. Tank tilt on uneven ground can also produce a level change that looks like a fill but is just fluid moving inside the tank.Engine state is the second match. Was ignition on, or engine running, during the rise? Many vehicle telematics units report accessory-on as engine-on, which is why idle fuel burn often reads fifteen to twenty percent high. That same misreporting can make a parked vehicle look like it was idling through a fill. Check whether the device logged a voltage dip or reboot around the event time, because reboots and sensor recalibration are common causes of false fuel drop alerts. A theft story built on a rebooted probe will not survive a supplier conversation.Now the time window. A clean fuel stop sits inside one shift and one geofence dwell. If the refuelling period straddles a shift boundary, or falls outside the dwell window, the cost lands on the wrong driver or the wrong day. A late geofence alert is the classic culprit: the truck leaves the yard at six, the alert fires at six eleven, and the fill is attributed to the next shift. Nothing was stolen. The record is simply wrong, and it will stay wrong until someone aligns the alert timestamp against the raw position data.Run this check on every flagged event, and on a small random sample each week so you are not only looking where alerts point you. Compare four things: position at the time of the level change, engine state during that same minute, the shift and dwell window it falls into, and the device health signals around it. If position, engine state, and window all agree, treat the event as real and escalate. If they disagree, fix the pipeline before you accuse anyone.Set a threshold for escalation rather than adjusting numbers by feel. A small variance inside sensor tolerance can be noted and monitored. A discrepancy that changes which shift, driver, or location owns the fuel, or that repeats across the same asset, is worth investigating as a data fault first. Where whole reporting periods are missing, flag the gap instead of letting the reporting layer average it away, because a silent average makes a broken week look normal.The same reconciliation logic extends across mixed assets. Generators, trailers, and tanks report on the same platform as vehicles under IoT asset monitoring, and each class has its own dwell and engine-state definitions. A generator has no odometer, so the match relies on runtime and load rather than distance. Keeping one reconciled timeline per asset class keeps the comparison honest.All of this rests on one assumption: compliance logs depend on accurate timestamps and location data. If the timestamps drift, the fuel figures built on top of them cannot be defended during an audit or a supplier dispute. GPS Controller pairs fuel performance monitoring with the underlying location and telemetry record, so the evidence chain stays intact from raw signal to reported cost. A clean reconciliation record shows the position, the engine state, the time window, and the device health in one place, with gaps marked rather than smoothed. That is the record you want in hand before anyone questions your numbers.