I'd like to get both the LogNormalized counts and the z-scaled counts out of the pipeline. The metro map of the pipeline shows the algorithms used on each step but it is not clear to me what count matrix is used on each step, so it's hard for me to understand what I should expect adata.X to contain on each step. I've seen the celltypist template calculating log normalized counts, but I believe they are not reused afterwards. Or is this step happening in all cell type annotation methods for some reason? Sorry for asking multiple questions. I'm happy to provide pull requests improving the documentation once I know the answer if you believe documenting this can be helpful to others

The adata.X stays the raw counts (as provided in the input) throughout the entire pipeline, and each tool comes with its own normalization procedure. It's easier to re-compute these whenever they are needed rather than having compicated case structures (if a certain step before has performed the normalization or if it has been deactivated for some reason) So for now it's probably easiest for you if you run the normalization yourself on the pipeline output, and I can think if there is a fairly elegant way of including it in the pipeline

It's nice to be able to rely on .X are the raw counts. It's easy to normalize when I need it, thanks!