The standard
How we measure 97.5%
An accuracy claim with no method behind it is decoration. Here is the audit protocol the number comes from, published so you can hold us to it — and so you can hand it to a reviewer who asks.
The audit protocol
Sample selection
On every study, a reviewer selects two fifteen-minute intervals per site for manual audit: one inside the identified peak hour, and one outside it. Peak intervals are where error concentrates, so the sample is deliberately weighted toward the hardest conditions rather than an easy average.
Independent manual count
Those intervals are counted by hand, movement by movement, by a reviewer who has not seen the automated output for them. The manual count is the reference; the automated extraction is what gets scored against it.
How the score is computed
Accuracy is scored per movement, not on the intersection total, because offsetting errors in opposite directions can make a bad count look perfect in aggregate. The reported figure is the volume-weighted mean absolute percentage agreement across all audited movements. Classification, where ordered, is scored separately as the share of audited vehicles placed in the correct bin.
The threshold, and what happens below it
A study passes at 97.5% or better on the audited sample. Below that, it does not ship: the reviewer works the full site rather than the sample, corrects it, and re-audits. A site that cannot reach the threshold because of the footage itself is flagged to you with the affected intervals named, and you decide how to handle it.
Recorded, not asserted
Every audit result is logged against the study, and the QA notes delivered with your workbook state which intervals were audited and what they scored. The published figure is the rolling result of those audits — which means it can move, and we publish it as measured rather than as a marketing round number.
For context
Agencies commonly require better than 95% agreement for count data to be accepted. Our threshold sits above that deliberately, because a study that barely clears an agency's floor is a study that invites a second look.
If you disagree with a count
Send us your manual count within sixty days. If an audit shows we fell short, we reprocess the affected sites at no charge or refund the credits — your choice.