#3712 row A2: the Q4_K universe is read from each file's tensor header (apr tensors --json), never a *Q4_K* filename glob; the judge recomputes membership #3763

#3712 row A2 (owner aprender-62, approved by the cop 2026-09-21): the Q4_K universe is read from each file's tensor header, not from its filename. Stacked on #3738 (row B1), which is stacked on #3721 (row A).

The requirement (the #3712 issue author's done_when 1 clarification, quoted verbatim)

done_when 1 clarified (issue author, cop aprender-04, 2026-09-21 18:15Z), after aprender-c3 asked whether …-q4k.apr (no underscore) counts as Q4_K. Under the operator's rule ("tests are dervived from surface and SHACL enforced … it is hand coded not derived", #3745), filename is not evidence of quantization. A file is in the Q4_K universe iff its header says so: the dominant weight-tensor dtype read by apr's own header reader (GGUF file_type / per-tensor ggml types; APR tensor dtypes) is Q4_K, with Q4_K_M/Q4_K_S included. A *Q4_K* filename glob is refused as the universe definition. Every candidate model file on the host (*.gguf, *.apr) is read, and the receipt lists each file with its measured dominant dtype, so a mislabelled file is visible both ways: a q4k name holding fp16 is excluded, and an unlabelled file holding Q4_K is included.

Design, as approved by the cop

• Candidates. Every *.gguf / *.apr under the ladder's model dirs is a candidate, selected by extension only. The patterns glob (*q4_k* / *q4k*) is removed, and a ladder that still declares one is refused.
• Membership. A file is in the universe when Q4_K is a most-frequent dtype among its tensors of two or more dimensions. Ties count as members, since including too much is the strict side.
• Reader. The dtypes come from apr tensors --json, apr's per-tensor header reader. It reads GGUF (IQ4_XS included) and APR. apr inspect --json is unusable for this: its quantization field reported Q6_K for a Q4_K_M file, None for every .apr, and it fails outright on IQ4_XS. That defect is filed separately and is not fixed here.
• Receipt. It lists every candidate with its dtype histogram. The judge recomputes membership from the histogram:
  • a Q4_K-header file missing from the inventory is a FAIL (excluded by name);
  • a non-Q4_K file in the inventory is a FAIL (included by name);
  • an unreadable candidate is a named FAIL.
• GPU. The header read runs through a no-GPU wrapper (CUDA_VISIBLE_DEVICES= empty), not the fleet GPU lock, so a held lock can't stall the inventory.

Measured on lambda (apr tensors --json, tensors of two or more dimensions, by count):

By bytes, that same Q4_K_M file is Q6_K, which is why bytes is the wrong measure.