Are these the same product?

The matching model runs offline and scores every pair of products it can compare. It shows the confident ones on its own; these are the ones it was not sure enough about.

Same means same item and same variant — a different brand is fine, and a different pack size is fine, because prices are compared per unit. A different flavour or a different cut is not the same product.

Your answers become labels: ml/evaluate.py measures the model against them, so a careless tick here makes the accuracy number wrong as well as the app.