As extra data point, this ordering choice seems to be consistent with Python syntax used by TACO:
csr = pt.format([dense, compressed]) # Compressed sparse row
csc = pt.format([dense, compressed], [1, 0]) # Compressed sparse column
The online TACO utility keeps the level format and dimensions nicely grouped together, though, avoiding the ambiguity. We could perhaps do the same in the syntactical format (e.g. (i,j) → (j:dense, i:compressed))?