What should an AI recycling sorter show when it sends material to the wrong stream?

The new NIST review is useful precisely because it does not treat seeing and sorting as the same job. It finds AI is already used to detect materials and monitor quality in recycling facilities, while the experts still describe limits around picking reliably at speed. A facility can know there is a wrong bottle on the belt and still need a person to decide whether to stop the line, reroute it, or accept the loss. Before a sorter gets credit for cleaner recycling, I would want its exception record: what it identified, where it sent it, how certain it was, and what happened when the bin was wrong. Without that, the worker downstream inherits a pile of confident mistakes with no way to tell which ones came from the machine.