Short answer
Reconciliation compares what a tank should hold, from deliveries and dispensing records, with what it measurably holds, and accumulates the gap over weeks. The US EPA reports statistical reconciliation detecting a leak as small as 0.2 gallons per hour, a rate no single-event threshold could ever see.
Key takeaways
- The US EPA inventory-control threshold is an overage or shortage of at least 1.0% of flow-through volume plus 130 gallons, a deliberately coarse figure.
- Statistical inventory reconciliation, using the same daily readings, detects a leak at least as small as 0.2 gallons per hour or a 150-gallon release within a month.
- Nothing about the sensor changed between those two sensitivities; only the length of the window and the statistical treatment did.
- Diesel volume changes with temperature, about 0.00085 per °C, so an uncorrected reconciliation can manufacture a discrepancy that never left the tank.
- A slow repeated theft is invisible per event by construction, so per-event thresholds and reconciliation are two instruments answering two different questions.
01Two published sensitivities, one set of readings
The most instructive pair of numbers in fuel release detection sits in the same US EPA document. For simple inventory control, the EPA describes a discrepancy worth investigating as an overage or shortage of at least 1.0% of flow-through volume plus 130 gallons. That is a coarse trigger, and deliberately so, because the arithmetic is done by hand from stick readings and delivery tickets. But the EPA also describes statistical inventory reconciliation, which works from broadly the same daily figures and detects a leak at least as small as 0.2 gallons per hour, or a 150-gallon release within a month. The gap between those two sensitivities is large, and nothing about the tank or the measuring device accounts for it. What changed is the length of the window and what is done with the numbers inside it.
02Why a longer window buys sensitivity
The mechanism is not mysterious. Every reading carries an error, and errors of the kind that dominate tank measurement, thermal drift, probe noise, a slightly tilted tank, an operator reading a stick, are as likely to be positive as negative. Add many such readings together and their errors partially cancel, while a genuine loss adds in the same direction every single time. Signal accumulates linearly and noise accumulates far more slowly, so the ratio between them improves as the window lengthens. That is the entire trick. A loss of 0.2 gallons per hour is utterly invisible in a single reading and completely unambiguous after a month, without any improvement in the instrument. Which is why the same probe supports two very different claims depending on whether it is being read as an alarm or as a ledger.
03What reconciliation actually requires you to have
Reconciliation is a subtraction, and it needs both sides of it to exist. On one side is the book figure: the opening measured volume, plus every delivery received, minus every issue dispensed. On the other side is the closing measured volume. The gap between them is the discrepancy, and its usefulness depends entirely on the completeness of the middle terms. A missing delivery note becomes an apparent gain. An unrecorded issue to a generator becomes an apparent loss identical in form to theft. So most reconciliation programmes fail on record-keeping rather than measurement, and the practical prerequisite is that every movement of fuel into or out of the tank produces a record with a quantity and a timestamp. A sensor that reads the tank perfectly cannot compensate for a delivery that nobody wrote down.
04Temperature will manufacture a discrepancy for you
There is one physical effect large enough to dominate a reconciliation if it is ignored, and in the Gulf it is not a small effect. Diesel expands as it warms, and the published volume correction factors quantify it: 1.0000 at 15 °C, 0.9958 at 20 °C and 0.9873 at 30 °C, which works out to roughly 0.00085 per degree Celsius. On a 20,000-litre tank a 15-degree swing between a cool morning delivery and a hot afternoon reading is about 255 litres of apparent volume change with no fuel having moved. Measurement Canada is explicit about the remedy, instructing that the uncompensated meter reading be multiplied by the volume correction factor for the measured temperature. If a reconciliation is run without that step and without recording a temperature, its discrepancy figure is a temperature history as much as a loss history.
05Why a slow theft is invisible per event by construction
Put the reconciliation numbers next to the instrument numbers and the reason for two instruments becomes arithmetic rather than opinion. Technoton publishes an inaccuracy of not more than 0.2% for a stationary tank, and on our own catalogue the OMNICOMM probe is listed at plus or minus 0.5%. OIML places certified dispensers in accuracy class 0.5. On a 20,000-litre tank, half a percent is 100 litres. A theft of twenty litres, repeated twice a week, is therefore below the noise floor of every single reading it occurs in, not because the sensor is poor, but because a small fraction of a large tank is a small fraction. Over three months, however, the same pattern removes roughly 500 litres, which is five times the noise floor and cannot be explained away.
06What a reconciliation programme looks like in practice
The programme is more clerical than technical, which is why it is often skipped. Take a measured volume at the same point in the operating cycle every day, ideally at a consistent time so temperature varies less between readings. Record every delivery and every issue with a quantity, and require a document reference for each. Record a fuel temperature alongside each measured volume and apply the correction factor. Compute the discrepancy daily but judge it weekly and monthly, because the daily figure is mostly noise by construction. Then plot the cumulative discrepancy rather than the daily one: a genuine loss appears as a line with a persistent slope, while measurement noise appears as a line that wanders around zero without going anywhere.
07Using both instruments, for what each is good at
The two approaches are complements rather than rivals, and confusing them is the common error. A per-event threshold answers a question about a specific moment: something large happened, here, now, and someone should look. It is the right instrument for a bulk siphon and the wrong one for a slow drain, and tightening it to reach the slow drain only converts measurement noise into alerts. Reconciliation answers a question about a period: over these weeks, more fuel left this tank than the records account for. It cannot tell you when or who, and it is not a real-time control. Run the threshold for large events, run the reconciliation for the pattern, and treat a reconciliation shortfall as the trigger for an investigation rather than as its conclusion.
| Property | Per-event threshold | Reconciliation over weeks |
|---|---|---|
| Question it answers | Did something large happen just now? | Did more fuel leave than the records explain? |
| Published sensitivity to work from | Bounded by sensor accuracy: 0.2%–1% | EPA: as small as 0.2 gal/hour over a month |
| Finds a slow repeated drain? | No, each event is below the noise floor | Yes, the losses accumulate in one direction |
| Finds a single bulk siphon? | Yes, that is what it is for | Yes, but only after the window closes |
| Tells you when and where | Yes | No, it gives a period, not a moment |
| Main failure mode | Threshold set below instrument noise | A missing delivery note or issue record |
| Needs temperature correction? | Rarely, over minutes | Yes, 0.00085 per °C dominates a long window |
Figures and sources
- US EPA inventory-control threshold for a book-versus-measured fuel discrepancy: overage or shortage ≥ 1.0% of flow-through volume plus 130 gallons (Source: US EPA, Release Detection for Underground Storage Tanks, Internal Methods, as of 2026-08-31)
- US EPA statistical inventory reconciliation sensitivity for slow repeated losses: detect a leak at least as small as 0.2 gal/hour, or a 150-gallon release within a month (Source: US EPA, Release Detection for Underground Storage Tanks, Internal Methods, as of 2026-08-31)
- Diesel volume correction factors per API MPMS Ch. 11.1, and the apparent change they imply: VCF 1.0000 at 15 °C, 0.9958 at 20 °C, 0.9873 at 30 °C ≈ 0.00085 per °C (Source: Measurement Canada (ISED), Volume correction factors, diesel fuel, as of 2026-08-31)
- Measurement Canada instruction that temperature correction is a required reconciliation step: multiply the uncompensated meter reading by the VCF for the measured temperature (Source: Measurement Canada (ISED), Volume correction factors, diesel fuel (PDF), as of 2026-08-31)
- Capacitive fuel level sensor accuracy for a stationary tank, the noise floor a daily reading sits on: not more than 0.2% stationary; 1% for moving vehicles (Source: Technoton, Fuel level sensor DUT-E product specifications, as of 2026-08-31)
- OIML accuracy class and maximum permissible error for the dispensing side of the reconciliation: accuracy class 0.5; MPE 0.5% (line A) and 0.3% (line B) (Source: OIML R 117-1:2007(E), Tables 1 and 2, as of 2026-08-31)
- OMNICOMM fuel level sensor accuracy and temperature rating on the ENTEK.AI catalogue: ±0.5%; −40 °C to +85 °C; RS-485 Modbus RTU (Source: ENTEK.AI product catalogue, as of 2026-08-31)
- NIST bulk-delivery test draft floor, which bounds how precisely a delivery quantity itself is known: at least the volume delivered in 1 minute at maximum rate, and never less than 180 L (50 gal) (Source: NIST Handbook 44 (2023) §3.31, N.3 Test Drafts, as of 2026-08-31)
Questions
What is inventory reconciliation for a fuel tank?
It is a subtraction repeated over time: opening measured volume, plus deliveries received, minus issues dispensed, compared with the closing measured volume. The gap is the discrepancy. Judged over weeks rather than days, it reveals losses far smaller than any single reading could resolve.
Why can reconciliation find a 0.2 gallon per hour loss when an alert cannot?
Because a genuine loss adds in the same direction every day while measurement errors are as likely positive as negative and partly cancel. Signal grows linearly with the window and noise grows far more slowly, so sensitivity improves with time, without any change to the sensor.
Do I need a temperature reading to run a reconciliation?
For any window longer than a day, yes. Diesel volume changes by roughly 0.00085 per degree Celsius, so on a 20,000-litre tank a 15-degree swing is about 255 litres of apparent change with no fuel moving. Measurement Canada instructs that the uncompensated reading be multiplied by the volume correction factor for the measured temperature.
Should I replace my fuel alerts with reconciliation?
No, run both, because they answer different questions. A threshold tells you something large happened at a specific moment and someone should look now. Reconciliation tells you that over a period more fuel left than the records explain, without saying when or who. Neither substitutes for the other.