Siril - Register and stack without re-normalization?

I have a sequence of fits files which were all taken with the same settings and exposure. It’s from a CMOS and the images have integer pixel values from 0 to around 30000 (I think saturation would set in around 65000).

I am trying, and failing, to register and stack the images to form a simple summed output that is NOT renormalized. The registration process always includes a line like “normalizing input data to our float range [0,1]” even though I have no where checked a box indicating I want it to normalize.

I am using the “image pattern alignment” method for registration.

What am I doing wrong?

Nothing, by default Siril works with 32bits images, so it converts image to the [0, 1] range.
In Siril preferences, you can choose to stay in 16bits.

If you are using simple summation, it will always renormalize in the [0,1] range (if siril is set in 32b) or [0,65535] (if set in 16b).
If you really want no output normalization, you can use Average stacking with rejection and choose No rejection. For this method, unless you check the Output Normalization box, no output normalization will occur. If you also want to have no input normalization (will be faster), choose No Normalization in the combo.