The demonstration
An underwriting agent needs to know whether a building is occupied. No dataset can tell it. It spends ninety dollars, a verified person goes and looks, the evidence is proved against the server’s own clock and geofence, and the agent carries on with its queue. No operator, no dashboard, no phone call.
An underwriting agent is clearing a queue of 40 properties. For one of them, every data source disagrees about whether the building is still occupied. There is no API for the answer. Somebody has to go and look.
Why this is the product
An agent can read every record about a property in seconds and still not know whether the lights are on. Everything upstream of the physical world is solved and getting cheaper. The last hundred metres is not, and it is where a great many workflows still stop and wait for a person to be asked.
Anyone can dispatch a person. The hard part is coming back with something software can rely on: a place proved by fixes the server recorded, a time proved by the server’s clock, and a model’s reading of the photographs that is advisory and never releases money on its own. That is what makes the result an input to the next step rather than a photograph somebody has to look at.
Spending limits per Tag, per day, per month, per hour, per campaign and per Recipe — all enforced inside the transaction that charges the card. Unsafe work refused before anything is charged. An operator can pause a Corner on suspicion without stranding a single person already out doing the job.
The walkthrough above is scripted. The same sequence runs for real against a live database with npm run demo — a model decides what to do at each step, the money moves through Stripe, and the ledger has to balance at the end or the run fails.