The pilot engagement
The Tier-Map Audit
A paid, fixed-fee, fixed-duration measurement of what your fleet can actually deliver.
What you get
- The per-camera Tier Map.
For every camera in the stratified sample: its Tier, its best measured accuracy on your held-out footage, the Lift where adaptation ran, and a plain-language capture explanation you can forward to the end site.
- Fine-tuned models on your worst clusters.
Where the Tier Map shows fixable cameras, we run the Adaptation Loop on the lowest-performing camera clusters and hand over the adapted models, co-owned per our standard IP terms.
- The runnable scoring script.
You can re-verify every number in the Tier Map yourself, on your own footage, without us in the room.
- A hardware-change shortlist.
The Tier 3 cameras, each with the physical reason it is capped.
How it is scoped
Fixed fee, fixed duration.
Typically a small number of weeks, agreed in the contract, with a hard end date. Always paid; we do not run free pilots.
A stratified sample, not a fleet crawl.
We sample roughly 50–200 cameras binned across site type, camera geometry, and lighting regime, so the sample predicts the fleet without auditing every camera by hand.
Pre-agreed acceptance criteria on your held-out footage.
Before we start, you state your Sellable Bar in writing: the accuracy threshold that makes the use case shippable for you, and how it is measured. The audit is accepted or not against footage we never trained on, from your fleet, verified against your records. There is a kill switch in the contract and it is defined before day one.
What we need from you
- Scoped read access to footage from the sampled cameras.
- The operational records we anchor ground truth to — for LPR, your billing or transaction records.
- Your Sellable Bar, in writing.
What happens after
The Tier Map seeds the expansion conversation. Audit, then adapt the fixable cameras, then ongoing per-camera measurement as the fleet and seasons change. You only pay to adapt cameras the measurement says are worth adapting.