Our numbers
Every figure we publish, with its conditions
A number without its conditions is a claim. Here is each one, how it was measured, and what it doesn't tell you. Quote them with the conditions.
97.7%
Our model picks the right serial on 97.7% of the drive labels it can read, within one character.
| Result | 209 of the 214 labels where the serial could be read |
|---|---|
| Labels | 241 human-reviewed labels from real storage devices, each with a visible serial |
| Method | Five-fold cross-validation, grouped by physical device, so no device is tested on a model that has seen it |
| Correct means | Within one character of the true serial |
| Where | Our lab, on an offline bench, not a customer floor |
| Measured | September 2026 |
87.4%
On the same labels, the serial it picks is an exact match 87.4% of the time.
| Result | 187 of the same 214 labels |
|---|---|
| Conditions | As above |
100%
Against an asset list, matching is exact: a serial is matched only if it is on the list.
| What it is | A design property, not a measured rate |
|---|---|
| Applies when | Your client supplies an asset list for the job |
| Every record | Shown to the operator, and can be corrected before the certificate is issued |
Under 1 second
From pressing capture to the serial being presented, on every deployment.
| Clock | Starts when capture is pressed; stops when the serial is presented |
|---|---|
| Applies to | All deployments |
| What it is | Our own check, not a published benchmark |
What we haven't measured yet
- A timed job of hundreds of devices on a customer's floor. Every figure here comes from our own bench.
- Accuracy on your labels. Labels vary by manufacturer and condition; a trial on your own drives will tell you.
When we measure something new, it goes on this page first, with its conditions.
Test it on your own drives
The app is free on Google Play, with 500 scans a month. Or tell us about your operation.