An interactive game that sorts 1,840 everyday objects into the City of Toronto's 8 waste bins, with Jev playing against you. Drag an item into a bin, see where Jev put it and how confident it was, then see what the City's rules say. How it uses Jev: every item is pre-scored offline in one pass, with a choice question (8 options) plus a noul question for whether the item needs a depot trip. All answers are embedded in the page, so the demo runs offline: no server, no API key, no live API call. The whole 1,840-item list ran in 58 seconds for $0.065, about $0.000035 per item. Scored against the City's published answer key: Jev 71.2%, an obvious keyword rule 39.5%, guessing the busiest bin 16.2%. That 31.8 point gap over the keyword rule is what shows this is real judgement rather than word matching. At 0.95 confidence or above it is 94.4% accurate; below 0.5 it is 43.4%. The best result is the qualifier round. 38 objects appear twice with different qualifiers that land in different bins, like sales receipts (matte paper recycles, thermal does not) or coffee cup lids (non-black recycles, black does not). Jev got all variants right for 23 of 38. A keyword rule got 0 of 38 and structurally cannot do better, having no concept of a qualifier. Where it fails: when the City's answer is that it does not accept an item at all, Jev finds that answer only 8.7% of the time and sends those items to a depot instead. It prefers to hand you an action over delivering a refusal. The same bias appeared in a separate health project, where it barely used its lowest category. At 71.2% it is not deployable on its own, but at high confidence it is accurate enough to queue for a human, and every answer carries a reason code that can be checked against the source data. Answer key: City of Toronto Waste Wizard open data, Open Government Licence. Not affiliated with or endorsed by the City.