Quality and audit
99.97%+ is a measurement. Here is how it is produced.
Accuracy that cannot be reproduced is an assertion. Ours comes out of the same system that runs the queues, is sampled continuously rather than at contract review, and is reported to you on whatever cadence you set.
How accuracy is measured
The disagreements are the measurement.
Because every plate gets four independent reads, the system already knows where humans diverge. That divergence — combined with supervisor adjudication and independent sampled audits — is what produces a defensible accuracy figure rather than a self-reported one.
Continuous sampling
- Audits run continuously against live production output
- Sampling rate set per contract, not per convenience
- Audits assignable to any reviewer or queue
- Your own team can assign audits against ours
Per-reviewer scorecards
- Accuracy, throughput and escalation rate per person
- Trend over shifts, weeks and months
- Visible to the reviewer, not just to management
- Used for coaching before it is used for anything else
Error categorisation
- Errors classified by type, not just counted
- Character confusion separated from jurisdiction error
- Capture-quality issues attributed back to the gantry
- Feeds targeted retraining rather than blanket refreshers
Reporting to you
- Accuracy against contract threshold, on your cadence
- Aging and oldest-pending against service levels
- Reason-code trending with gantry and time-of-day attribution
- Complete action history, exportable on request
Training
New reviewers do not learn on your transactions.
- Structured onboarding before floor access. Reviewers train on historical and constructed sets, not live production, until they clear an accuracy gate.
- Jurisdiction training is its own discipline. Specialty series, university and military plates, and out-of-state lookalikes are taught and tested separately from character entry.
- Transactions flagged for training stay in the record. A flagged transaction is used for coaching without being pulled out of the production audit trail.
- Retraining is targeted. If a reviewer’s errors cluster in one plate type or one confusion pair, that is what gets retrained — not the whole curriculum.
- Supervisors are promoted from the floor. Adjudication is done by people who have keyed plates at volume, because that is the only way to judge a marginal call quickly.
Independent verification
You can sample our work yourself, at any time, without notice. Assign audits against our reviewers through the platform, or pull the raw action history and check it against your own records. We would rather be audited than trusted.
See the report before you buy anything.
Send a sample file. We will return it reviewed, coded and scored, with the quality report attached.