How we work out every number

Last updated 18 September 2026

This page exists because we ask you to trust figures about your own money. You should be able to check where they came from.

What we do not claim

We do not publish an accuracy percentage. To do that honestly you need a set of quotes where somebody independently established the correct price, and a published protocol for measuring yourself against it. That dataset does not exist in this industry, and we have not built one. Any figure we quoted would be unverifiable.

If we ever do run a blind validation against known-good invoices, we will publish the protocol next to the result.

What we do instead

Every number in the app renders with three things attached: where it came from, how many observations sit behind it, and how old it is. If our pipeline cannot attach those, the number does not appear at all.

When we refuse to answer

Observations we hold	What the app does
Fewer than 5	Says “we can't assess this”. No verdict, no range, no hint.
5 to 19	May say a line looks typical or high, at low confidence, but will not show a numeric range.
20 or more	Shows the typical range, with the sample size next to it.

We never widen a range to cover for thin data. A wide range still reads as a verdict; a refusal does not.

We need more evidence to accuse than to reassure

This is deliberate and it is asymmetric. Telling you a fair quote is a rip-off is far more damaging than missing a genuine overcharge — it costs you your relationship with an honest tradesperson and costs us your trust, in one move. So a low-confidence benchmark can raise a question, but it can never issue a verdict.

Four things we do that are easy to get wrong

We group the quote into jobs before judging it

Published benchmarks are whole-job totals. A real quote splits one job across a parts line and a labour line. Comparing a single parts line against a whole-job benchmark would systematically tell you your parts were cheap.

We account for work that shares a strip-down

Brake pads and rotors on the same axle share the wheel-off and caliper teardown. A water pump sits behind the timing cover. Adding two published job totals together overstates the real cost — so we model the overlap, and the app asks the business whether the shared labour was charged once or twice.

We treat parts markup as a sliding scale

Markup in the trades is tiered and gets proportionally smaller as parts get dearer — roughly 3× on a very cheap part and about 1.7× on a very expensive one. A flat rule would flag every inexpensive part as gouging.

We do not average across things that vary wildly

A timing belt is a different job on a small sedan and on a full-size pickup — published estimates for the same operation differ by a factor of three to seven between vehicles. A national average for “timing belt” is not a benchmark; it is a number that is wrong for nearly everybody.

Benchmarks get old, and we say so

The best independent survey of UK garage labour rates we can find was published in April 2023. That is still the best available evidence, so we use it — and we lower our confidence, show you the date, and tell you on the number itself that the underlying survey is three years old. We would rather be visibly uncertain than quietly out of date.

Our sources

Where a source does not publish a sample size, we show “sample size not published” rather than implying one. We are in the process of licensing or replacing several of the commercial sources above; where a licence does not permit us to use a figure, we remove it rather than use it quietly.

What we get wrong

Our coverage is uneven. It is strongest for US auto repair and weakest outside it. We have almost no Canadian data, which is why Canada is not a launch market. Some jobs have no benchmark at all and the app will tell you so. If you think a verdict is wrong, tell us at admin@quoteshield.site — corrections from people who work in the trades are the most useful mail we get.