@foone Is this yet another probabilistic LLM victim, where the dataset somehow had a majority of AUSGANGs being 12V, while the outputs where 5V or just maintained whatever was the input digit?
Or is this chip and its labelling too old for that particular can of large-language-worms?