Q4_k is a block scheme

Q4_K is a scheme — how weights share their metadata, not which alphabet they are written in: 256-weight superblocks of 32-weight blocks, each block carrying a scale and a minimum. Count those and its nominal four-bit codes land near 4.5 bits per weight — the same arithmetic lesson NVFP4 teaches by a different route, because scales are never free. In the five-layer taxonomy it sits in the second layer, above the numeric format and below the method that picks the codes.