Q48 enumeration: every antigen, identifier and query this beat ran - including what was searched and rejected, and the three identifier traps found live

Rule 67 compliance: each test below was run through the identical code path as its controls, on 2026-09-20, with PubMed querytranslation read back. No parenthesised der(n) token and no unquoted multi-word phrase was used.

What each absence licenses, stated narrowly:

• #1-#3: no published functional test of CLEC12A dependence in KMT2A-rearranged leukaemia. Residual: 72 Europe PMC full-text co-mentions were title-checked, not read.
• #4: the enzymes that build CD15 are essentially absent from the KMT2A literature, which discusses the epitope 36 times.
• #5: no study reports whether HLA-DR expression on KMT2A-rearranged blasts differs from other B-precursor ALL.
• #6: a symbol-keyed literature sweep for TdT finds 2 of 39 papers, and the decisive one (PMID 34148065) is not among them.
• #7: CD65 is unresolvable in HGNC - the honest blank.

One instrument caveat that cuts against this beat's own framing. PubMed's counts are not like-for-like across these rows: CD15 expands to "lewis x antigen"[Supplementary Concept], CLEC12A and FUT4 and DNTT expand through ... protein human Supplementary Concepts, while FUT9, CD65 and CD65s are bare All-Fields string matches with no expansion at all. Ratios across differently-indexed terms overstate the gap. The in-graph regex counts, where no expansion happens on either side, are the cleaner measurement.

The brief notes that bounded rejection does not exist anywhere in this graph and is the larger half of a negative. These are this beat's.

• PMID 33097838, A trispecific killer engager molecule against CLEC12A effectively induces NK-cell mediated killing of AML cells (Leukemia, 2021). On-beat modality, not written as its own entry because the beat's CLEC12A therapeutic coverage is already carried by two clinical studies 12 and a preclinical gated design; a third preclinical modality adds little. Would change my mind: if a TriKE entered clinical trial in KMT2A-rearranged disease.
• PMID 37848261, Asymmetric anti-CLL-1xCD3 bispecific antibody, ABL602 2+1, with attenuated CD3 affinity (JITC, 2023). Same reason; noted because low-affinity-CD3 design is the bispecific analogue of the interferon-gamma attenuation strategy in the risk entry. Would change my mind: clinical data.
• PMID 34253594, Anti-CD19 CAR T Cells That Secrete a Biparatopic Anti-CLEC12A Bridging Protein (Mol Cancer Ther, 2021). Directly relevant to CD19-to-myeloid lineage switch in KMT2A-r disease and arguably belongs in this graph - but it belongs in the CD19/lineage-switch beat, not here, and per the brief I do not write into another beat's territory. Flagged to the coordinator as a gap.
• PMID 39133622 (tandem CD33-CLL1 CAR-T) and PMID 35261791 (PD-1 knockdown CLL-1 CAR-T case reports). Small series and case reports; the phase 1 data in 1 and 2 dominate them. Would change my mind: nothing short of a controlled comparison.
• PMID 33220589's CD123 conclusion. Read and deliberately left in gene-clec12a's counter-evidence rather than written up, because gene-il3ra already exists and owns CD123.
• The sialyltransferase that makes CD65s. ST3GAL4 (Q11206, 333 aa) and ST3GAL6 (Q9Y274, 331 aa) were resolved in UniProt and deliberately not asserted, because no source read on this beat attributes the VIM-2 epitope to either. Recording this unresolved is the correct outcome. Would change my mind: a paper attributing the VIM-2 ganglioside to a named sialyltransferase in myeloid cells.
• CLEC12B (Q2HXU8, 276 aa). Looked up as an identifier control; no leukaemia relevance found, no entry written.
• PMID 26571505 was named in the dispatch brief as background and then withdrawn by the coordinator mid-beat as belonging to another agent's beat. Not searched, not read, not cited.

Trap A - CLL1 is a gene synonym of two reviewed human genes, and the wrong one sorts first. gene:CLL1 AND organism_id:9606 AND reviewed:true returns COLEC10 (Q9Y6Z7, Collectin-10 / CL-L1, 277 aa) before CLEC12A (Q5QGZ9, 265 aa). Twelve residues apart - tighter than the SMC1A/SMC3 pair (16) that rule 66 calls the limit of the length defence. Add CLEC12B (Q2HXU8, 276 aa) and three unrelated proteins lie within 12 residues. verify-accessions.py's length assertion provides no protection here; only gene_primary does. CL-L1 and CLL-1 differ by one hyphen and name different proteins. 3

Trap B - CD15 is a PREVIOUS OFFICIAL HGNC SYMBOL of FUT4. An antigen with no gene has, historically, been the official name of an enzyme gene, and HGNC free-text search still returns FUT4 for the string. The obvious repair to the named-but-unwritten queue introduces this error while fixing the TdT one. 4

Trap C - the length band on this beat is pathologically crowded. HLA-DRA 254, CLEC12A 265, HLA-DRB1 266, CLEC12B 276, COLEC10 277. Five proteins, four of them unrelated, within 23 residues. By contrast FUT4 (530) and FUT9 (359) are 171 apart and DNTT (509) is isolated. A length check is a real discriminator for some genes on this beat and worthless for others, and nothing in the check tells you which case you are in. 35

And one near-miss worth recording as a process failure. A DOI for PMID 9376686 was written into a draft evidence block as 10.1093/glycob/7.7.973 before the record was fetched. The true DOI is 10.1093/glycob/7.6.835. It was caught by fetching the record before the file was written, not by a checker. This is the brief's rule - when you have the fetched record open, copy every identifier out of it - failing in real time, and it is logged here rather than quietly fixed.