Tell a real fuel loss from daily noise — with a certificate.
Fuel-tank numbers never balance exactly — the question is whether a discrepancy is noise or a leak. at1 wetstock reconciles daily dips, deliveries and pump totals with a model-coupled residual plus a CUSUM change-detector, separating a real leak or meter drift from ordinary scatter, and issues a certified report: loss rate, confidence interval, and estimated onset date.
- leak vs noise
- a model-coupled residual separates a real loss from daily scatter
- loss rate + CI
- not a yes/no — a rate with a confidence interval
- onset date
- CUSUM change-detection estimates when the loss began
- certified report
- a signed, re-checkable verdict, not a spreadsheet guess
Reconcile the tank
Daily dips, deliveries and pump totals should balance. The unexplained residual — the part the model can't account for — is where a leak or meter error hides.
Leak or meter drift
A steady loss reads as a leak; a slow bias reads as meter drift. The model tells the two apart instead of lumping every discrepancy into 'shrinkage'.
When did it start
A CUSUM detector watches the residual for a change in trend and estimates the onset date — so you know not just that there's a loss, but roughly when it began.
The workflow
# reconcile a tank's history (dips, deliveries, pump totals) -> certified report at1 wetstock reconcile tank_07.csv -o wetstock-report.json # read the verdict: loss rate + confidence interval + onset date at1 wetstock report wetstock-report.json # re-verify the certified report independently at1 wetstock verify wetstock-report.json
Honest scope
This is a statistical reconciliation — it flags and sizes a loss with a confidence interval, it does not physically locate a leak or replace a certified calibration. Detection power depends on how clean and regular your daily readings are: sparse, erratic or mis-entered dips widen the interval and delay onset detection. Small losses need more days to separate from noise. The right check is your acceptance test on real data — run it over a tank with a known historical loss and see if the report matches. Validated on synthetic tank histories; partner acceptance on real forecourt data is pending.