That black plastic tray in your recycling bin? The sorting machine at the plant can't see it. Carbon black soaks up the infrared light those sorters read by, so a black tray on a black belt reads as empty space and rides on to landfill. By industry estimates, 3 to 15% of the waste stream is lost this way. It isn't truly invisible, though. In another band, mid-wave infrared, that same tray has a clear signature. So we built Assay, a small neural net written from scratch that reads it and names the plastic the main sorter missed: PP, PE, PS, ABS, or PVC. (The sensor feed is simulated; the classifier and its checks are real code.) A good classifier isn't what makes a recycled bale sellable. Run our model on a stream of black plastic and it builds a bale that's 99% pure polypropylene, clean enough for the grade a buyer will pay for. Flip off one safeguard, leave the same model running, and the same stream ships a bale with almost 2% PVC in it. Rejected. Same model, remember. The difference is one rule: when the net isn't sure, it sets that piece aside for a person instead of guessing it into the bale, and anything that might be PVC gets pulled on sight. Across 40 test bales, the guarded version passed spec about 85% of the time. The unguarded one, about 22%. A system that oversells its own purity is useless to a recycler. So Assay abstains on what it can't prove, and signs a record of what's actually in every bale. #Recycling #PlasticRecycling #CircularEconomy #WasteManagement