What this beat looked at and did not write: rejected candidates, with counts for both halves of every search

The other four intersection hits are a review of Dot1 crosstalk (33879367), a chemical-biology review (33847478), and two proteasome-inhibition papers about the KMT2A transcriptional complex (36781850, 32733069) that concern protein degradation rather than histone H2B ubiquitination. Correction (coverage repair, 2026-09-19). This entry originally read: 'The entire KMT2A-specific literature on the H2B ubiquitination axis is two primary papers, thirteen years and five years old.' That sentence is withdrawn. The three counts above reproduce exactly on re-running, but the term set itself undercounts. Intersecting USP22[Title/Abstract] against the same fixed KMT2A set returns PMID 42418681 (Broechtel et al., Blood 2026): a genome-wide CRISPR screen in a KMT2A-rearranged B-ALL line identifying the SAGA H2B deubiquitinase USP22 as a regulator of MEIS1 protein stability. That is primary, KMT2A-specific, on an enzyme of this axis - and this graph already cites it as gene-usp22 E1, so the beat held the paper while its own ledger said it did not exist. The correct statement is: the KMT2A-specific literature on the H2B ubiquitination axis is at least three primary papers (2013, 2018, 2026), and this ledger's term set cannot be relied on to have found all of them. Checked and found to add nothing: H2Bub1 (1 hit, already in the six), uH2B, BRE1, MSL2, RAD6, H2BK34 (0 each); WAC and ATXN7L3 each returned one incidental hit (PMIDs 29158550, 37120350) which were fetched and discarded. Recorded as counter-evidence on 1.

The thinness is still real and is still the justification for the beat existing. Three primary papers in thirteen years is not materially different from two. What is different is that a term-set count is now on record here as a floor, not a census.

Intersection counts against the same fixed KMT2A term set. Ratio = intersection / mark alone, a crude index of how much of a mark's literature is about this disease.

The pattern worth noting. H3K79me2 has 13.4% of its total literature intersecting KMT2A. Every other mark examined is between 0.15% and 1.4%. The graph's distribution of attention across marks is, on this measure, an accurate reflection of the field's — and the one place where it was badly wrong was H2B ubiquitination, which has 632 papers of its own, a mechanistically necessary relationship to the disease's defining mark, and until this cycle no entry at all.

A full-text sweep (rule 80) overturned the counts behind two rows of the table above, and the H3K27ac row's DECISION with them.

H3K27ac. This entry recorded "eleven records, all enhancer-landscape papers in which the mark is a readout rather than a subject" and declined to write the mark. Both halves are wrong:

• PMID 27856324 was inside that count and its title is MLL-AF4 binds directly to a BCL-2 specific enhancer and modulates H3K27 acetylation. The mark is the subject.
• The decisive KMT2A::AFF1 dataset was outside the count entirely. PMID 31243293 contains KEE-restricted H3K27ac ChIP-seq in SEM cells — 49 in-text hits — and the abstract-level intersection does not return it. This graph cites that paper in fourteen entries and still missed the measurement in it.

H4K16ac. Recorded as "a literature of one KMT2A-intersecting record"; the list returns three, and the only locus-resolved H4K16ac observation in a KMT2A-rearranged cell sits outside the count — PMID 25822366, Supplementary Figure 8, H4K16ac elevated at MLL-AF9 targets.

The mark is now written: see mark-h3k27ac and mark-h4k16ac. The failure was not the arithmetic but the instrument — an abstract-level intersection cannot see a measurement that lives in a supporting figure, and this entry used one to make a write/do-not-write decision. That is rule 80 applied to a curation decision rather than to an absence claim, and it is the more expensive version: an absence claim is provisional by nature, whereas a decision not to write closes the question.

The counter-evidence note added to this entry on 2026-09-20 cited PMID 25822366 with the title "AF9 YEATS domain links histone acetylation to DOT1L-mediated transformation" and first author Li Y. Both were written from recall and both are wrong. The paper is "DOT1L inhibits SIRT1-mediated epigenetic silencing to maintain leukemic gene expression in MLL-rearranged leukemia", Chen CW et al., Nat Med 2015. The DOI (10.1038/nm.3832) and the PMID were right, and nine other citations of the same PMID elsewhere in this graph carry the correct title — only the one written from recall was wrong.

The substantive claim is unaffected: Supplementary Figure 8 of that paper does show H4K16ac elevated at MLL-AF9 targets, which is what the note asserts.

It was caught by a sister beat's --strict run, not by mine, and only on the AUTHOR check. The title check passed it — weighted overlap 0.1979 against a threshold of 0.18, on two shared tokens (dot1l, mediated), the second of which is shared only because both titles contain an X-mediated construction. See the hardening note in scripts/verify-citations.py.