Twenty fridges, built from statistics
The first study compares dishes planned by home cooks with recipes suggested by AI models for the same kitchens. We had planned to design those kitchens with focus groups. They are now built from official statistics instead: national household budget surveys tell us what households in Italy, Germany, Poland and the United Kingdom buy, and WRAP’s measurements of household food waste tell us which of those foods most often end up in the bin.
The result is twenty “fridge-and-pantry” scenarios, five per country, each with 19 foods and a leftover, some of them past their best or close to their use-by date. The rules are fixed in advance and the code will be released with the benchmark. You can try them in the new demo: pick a region, open a fridge stocked with the same rules, and see how much of the food at risk your own dish would rescue.