WangNet – 1.8 MB, zero-dependency Numberwang adjudication in 11 languages

WangNet is a 1.8 MB neural network that adjudicates Numberwang across 11 languages using only 100 lines of standard Python. The model utilizes 80,804 parameters to process characters directly without tokenizers, achieving an 88.9% accuracy rate. It functions by memorizing patterns rather than performing arithmetic, making it reliable for common expressions but less accurate for complex calculations.