Café Intelligence
See the bar going wrong while it is going wrong
Every shot, judged per group head, with a channeling alert in seconds rather than on tomorrow's report.
The problem
One failing group on a three-group machine is the most common real fault in a café — and a machine-level average hides it behind two groups that are fine.
Judged per group head
Three bad shots out of the last five on one group is the equipment, not the barista. Two is a barista having a moment, and alerting on it teaches everyone to ignore alerts.
- Channeling is distinguished from merely fast — one sends you to the grinder, the other to the basket
- A dumped shot is recorded, because waste is a number the manager wants
- Consistency is reported beside the average: 28s ±1 and 28s from 20 and 36 are different bars
A replayed buffer changes nothing
Bar bridges dedupe on their own shot id. A reconnect after an outage does not double the day's numbers, which is the quiet failure that makes a café dashboard untrustworthy.
Shots against the till
Reconciliation compares what the machines pulled against what the point of sale rang up. A sale with no shot is a till error; a shot with no sale is waste or theft. Merging them would erase the question.
Everything here is in the API
This area is cafe.* — the same operations the console calls.