I looked at the PyPOTS 0.1.1 source code, and it appears for SAITS and the Transformer (maybe for others, I only checked these so far) in their fit methods, they create a DatasetForSAITS (https://github.com/WenjieDu/PyPOTS/blob/201fe6f3b8d7bfe5181924f7850947f8d99c5afa/pypots/imputation/saits/model.py#L386) which applies MCAR missingness at a default rate of 0.2: https://github.com/WenjieDu/PyPOTS/blob/3f0d09c7de4ca7349371a59bb4be786378e106e2/pypots/imputation/saits/data.py#L16

If I'm not mistaken, this means that the missingness already imposed on the data by the YAIB code (i.e. the amputated values) will not be imputed, but rather treated as genuinely missing by the PyPOTS code, and so only the new MCAR missing values it imposes will be predicted on.