Tightening the measurement: when you need a number you can defend

Reviewed June 22, 2026 7 min read

Most MPG tracking is personal and a 3 to 5 percent error doesn't matter. Some isn't: warranty disputes, tax filings, fleet reporting, regulatory submissions. This page is the second case. It walks through what changes when you need a number that survives scrutiny, and where the additional effort is worth it.

When the casual method is not enough

The fill-up method described on the calculation guide is accurate to about 1 to 2 percent on a 3 to 5 fill-up average. That is more than enough for tracking your own MPG over time and comparing to the EPA sticker. It is not enough if the number will be used in a context where someone else will scrutinize it: a warranty claim, a tax filing, a fleet report to a client, a regulatory submission.

What changes between the casual case and the defensible case is not the basic math (still MPG = D ÷ V). It is the controls around the measurement: how the fuel is measured, how the distance is verified, how the conditions are documented, and how the data is analyzed. The next sections walk through each.

Error analysis: uncertainty in the inputs and the output

For a quotient (MPG = D ÷ V), the relative error in the result is the quadrature sum of the relative errors in the inputs:

Relative error in MPG = √[(σₖ/D)² + (σₛ/V)²]

where σₖ is the absolute uncertainty in the distance and σₛ is the absolute uncertainty in the fuel volume.

Where the error actually comes from

For a typical personal fill-up, the error budget breaks down roughly like this:

  • Odometer: 0.1 to 0.5 percent (depending on stock vs. non-stock tire size).
  • Fuel pump: 0.5 to 1.0 percent (legal tolerance is the main source; display resolution is a small contributor).
  • Fill-level consistency: 0.5 to 1.5 percent, dominated by the operator's habit of topping off or not.
  • Fuel temperature: 0.1 to 0.5 percent, depending on how far the fuel temperature is from the 60°F reference.

The pump and the fill-level consistency dominate. Tightening the fill procedure (always first-click, always same station) usually moves the result more than any other single change.

Systematic vs. random error

Distinguish two kinds of error, because the fixes are different:

  • Systematic (bias): A consistent over-read or under-read. Examples: non-stock tire size, a pump that runs hot. The result is consistently off in the same direction, and the fix is calibration.
  • Random (noise): Variation around the true value from one fill-up to the next. Examples: pump-to-pump differences, traffic variation, cold-start variation. The fix is averaging across more fill-ups (√n reduction).

For full background on the factors in the error budget, the factors guide walks through each one; the math guide covers the propagation formulas in detail.

Validation: cross-checking the result

For defensible work, no single source of truth is enough. The result should agree with at least one independent measurement within the combined error budget. Three cross-checks that actually catch errors:

  • Compare with the trip computer (expect 3 to 8 percent difference, trip computer is usually optimistic). If the difference is outside that band, the fill-up procedure or the odometer is suspect.
  • Compare with the EPA combined rating for the specific trim (expect within 10 to 15 percent, lower in real conditions). A result that is dramatically above the EPA rating, say 20 percent or more, usually means a measurement error, not a super-efficient driver.
  • Compare fill-up against GPS distance for a long drive (expect within 1 to 2 percent). This is the most reliable cross-check for the odometer specifically, and it catches non-stock tire-size issues that the casual method cannot.

The calculator on the homepage runs the EPA-rating comparison automatically and flags results outside the plausible band.

Control charts: detecting real change vs. noise

Once you have a baseline (10 to 20 fill-ups under consistent conditions), the next question is whether a new fill-up represents a real change in efficiency (an aging sensor, a dragging brake, a fuel-quality change) or just the noise of measurement. The standard quality-engineering tool for this is the control chart, adapted from the same Shewhart / Western Electric rules used in manufacturing SPC.

The full application to fuel tracking, with the I-MR chart construction in Python, the control-limit formulas, and the interpretation rules, is on the SPC guide. The short version: a single fill-up 2 to 3 MPG below the baseline is normal noise; the same deviation sustained across 5 to 10 fill-ups is a signal worth investigating.

When to use a different method

The fill-up method is not the right tool for every defensible use case:

  • Warranty or tax disputes: For an authoritative number that an insurer, manufacturer, or tax authority will accept, get a controlled test from a credentialed professional (a dealership service bay, an emissions test center, or an independent automotive testing lab). The fill-up method's result is an estimate, not a measurement, and the difference matters in a dispute context.
  • Regulatory submissions: For EPA or CARB filings, the relevant test cycle is the one the regulator mandates, not a fill-up log. The standards are documented in the EPA vs WLTP vs JC08 guide.
  • Fleet-wide reporting: For more than a handful of vehicles, a purpose-built fleet platform with telematics (Geotab, Samsara, Verizon Connect) is more efficient than manual fill-up logging. The per-vehicle accuracy is similar; the labor cost is the difference.

The accuracy checklist

For a defensible number, the protocol needs to be reproducible. The minimum that survives scrutiny:

  • Same station, same pump, same side of the pump, for the full measurement period.
  • Always fill to the first click. Document the click as the fill level; never top off.
  • Adequate distance per fill-up: 200 miles minimum, 300+ preferred.
  • Normal driving conditions for the driver and route; note unusual conditions (long trip, towing, idling) and exclude them or report them separately.
  • Photographs of odometer and pump display at each fill-up, with timestamps.
  • Documented environmental conditions: temperature, terrain, traffic.
  • Load and weight documented, consistent across the measurement period.
  • Cross-check against at least one independent source: GPS distance, the EPA rating for the trim, or a controlled test for the specific vehicle.

For the math behind the analysis (weighted average across fill-ups, confidence interval, error propagation), the MPG formula math page has the formulas. For the process-control method of detecting real change, the SPC guide is the next read.