One domain, two names - and `MLL[tiab]` does not match `MLL1`. Two token traps, two missing affinity papers each

Run at the coordinator's own tooling on 2026-09-21, one tag per term per rule 94, querytranslation recorded 1:

The 21 records reachable through MLL AND CXXC are, by PMID: 40769927, 39877387, 36598580, 35978680, 32997997, 32519403, 30352306, 29276034, 24449215, 23990460, 23697932, 21155757, 20541477, 20010842, 19060922, 16990798, 16897742, 15547684, 15542854, 15375572, 12124344.

Two of the five papers that carry an affinity measurement for this domain are missing from that list: Birke 2002 (PMID 11842107, the 33 nM SPR) and Erfurth 2008 (PMID 18483194, the five-point ITC ladder and the published reconciliation of the discrepancy). Both are missing for the same reason - their abstracts frame the work as methylation protection and use MT domain, never CXXC.

Note also that two of the 21 are noise from the CXXCn ordinal collision already in the registry: 15547684 and 15375572 are in silico gene-identification papers about FBXL19 and 'CXXC10'.

This beat ran the queries above, concluded that 21 records was the reachable CXXC literature, and missed a sixth affinity paper that was already inside one of the entries it had been sent to correct.

PMID 34280370 - Liang et al., Biophysical Journal 2021, Dynamics and inhibition of MLL1 CXXC domain on DNA revealed by single-molecule quantification - carries Kd 5.7 +/- 0.2 uM at 150 mM NaCl for a tagless CXXC domain, which is the single most physiologically apt measurement of this contact in the literature. Measured 2026-09-21 3:

PubMed [tiab] matches whole tokens. MLL and MLL1 are different tokens, and the field uses both interchangeably for the same protein: Cierpicki's title says MLL CXXC, Liang's says MLL1 CXXC. Neither query reaches the other's paper unless both strings are run.

So the reachable literature for this domain needs at least four strings: MLL, MLL1, KMT2A and "MT domain". Three of them were run by this beat and the fourth was not, which cost a paper that changes the beat's own central number.

The registry already warns about MLL as a free-text token - ten of twenty-six records in one query were microbiology and wastewater papers - and the fix it prescribes, always query KMT2A, or MLL with a disease term, would not have helped here: KMT2A[tiab] AND CXXC[tiab] returns 8 and does not include Liang either, because the paper is from 2021 and still says MLL1. The existing row addresses precision and this is a recall failure on the same token.

How it was actually found: by reading the counter-evidence array of exp-cxxc-reader-depletion-versus-kmt2a-fusion-occupancy, one of the six entries this beat was sent to amend. The graph already held the paper that corrected the beat's own analysis, in the entry the beat was correcting. That is the second time in two cycles - cycle 45's Kerry finding had the same shape.

This graph cited PMID 20010842 in six entries and PMID 11842107 in five, and the two sets are disjoint - no entry cited both. That is exactly what you would expect if the two papers were found by different queries on different beats and never co-retrieved. The structural beats searched CXXC and found Cierpicki, Allen, Risner and Xu; the methylome beat searched methylation and CpG and found Birke. The graph's own citation pattern is a shadow of the retrieval split.

The brief's rule 70 says a counted absence about a noncoding element is bounded by its name list. This is the coding-domain twin, and it is worse in one respect: a gene has HGNC, so SEPT9/SEPTIN9 is resolvable by lookup. A protein domain has no naming authority at all. Pfam calls this one zf-CXXC (PF02008), UniProt calls it a CXXC-type zinc finger with boundaries 1147-1195, the 2002 literature calls it MT or BCB with boundaries 1147-1244, and the structures call it CXXC with boundaries 1147-1203. Four names, four boundaries, no authority.

What to do, generalised: before recording any domain-level absence or comparing any two domain-level measurements, enumerate the domain's names from the full text of the founding paper - not from the current structural literature, which has usually renamed it - and run each name as a separate query. Then check whether the name sets intersect. If they do not, you have two literatures and you have read one.

Grepped data/symbol-collisions.json (87 rows) for MLL, CXXC and CpG before any query, as rule 88 requires. Three relevant rows exist - MLL as a free-text token, CXXCn as gene symbol versus ordinal domain reference, and MLL2/MLL4 - and none covers this. Proposed new row:

{
  "token": "MT domain / BCB domain (KMT2A 1147-1244)",
  "class": "one_protein_domain_two_names_two_disjoint_literatures",
  "this_graph_means": "the CpG-binding region of KMT2A; the MT/BCB name covers 1147-1244, the CXXC name usually covers 1147-1203",
  "measured": {
    "MLL[tiab] AND CXXC[tiab]": 21,
    "\"MT domain\"[tiab] AND MLL[tiab]": 1,
    "intersection": 0,
    "CXXC[tiab] AND 11842107[uid]": 0,
    "\"BCB domain\"[tiab]": 0,
    "date": "2026-09-21",
    "tool": "PubMed E-utilities esearch, one tag per term"
  },
  "why_it_matters": "The only SPR affinity for this domain (PMID 11842107, 33 nM) and the only multi-oligonucleotide ITC ladder plus the published reconciliation of the 1300-fold discrepancy (PMID 18483194) are BOTH unreachable from any CXXC query. This graph cited the two literatures in disjoint entry sets for 45 cycles.",
  "block_autogeneration": false,
  "found_by": "Q82-cxxc-affinity"
}