The menin-dependence test I proposed in the cycle-146 log cannot fire by construction - and what the one pair CAN answer is that 85% of gene-body occupancy is menin-dependent while histone clusters preferentially survive

GEO labels BOTH aliquots 'treated with the Menin inhibitor'; the assignment of 148752 as the at-diagnosis baseline and 152985 as on-treatment is this corpus's own reading of a case-level string, taken from ana-the-infant-menin-inhibitor-occupancy-experiment-exists-and-it-is-kmt2a-enl-not-aff1, and was not re-derived from the publication this session. If that reading were reversed the direction of E2 would invert, though E1's by-construction argument would be unaffected.

• The unfireability is a property of THIS class definition, not of the question. A differently defined class - for example one requiring occupancy in >=3 AFF1 samples without requiring 0 in comparators - would be testable in the pair. The honest statement is that exclusivity and menin-dependence cannot be tested on the same gene set here, not that menin-dependence is untestable.

n = 1 pair. The two aliquots are recorded in this corpus as patient-matched B-ALL at diagnosis and AML respectively, so lineage and clonal evolution are confounded with the drug and this cannot be attributed to menin inhibition alone. The 16 'gained' calls are unexplained and are most likely noise at a collapsed denominator. The KMT2A-C and IgG arms were not scored for this pair. The promoter density collapse is itself a global-signal-loss measurement and is reported because it sets the direction of the bias, not as an independent result.

• A collapsing normalisation denominator makes the whole comparison fragile in both directions, not just the conservative one. If promoter density falls because the drug removes promoter signal, then the denominator is part of the effect being measured, and a ratio whose denominator is itself an outcome is not a clean measure of anything - this corpus has recorded that failure mode for a different ratio built to cancel a confound.
• Nothing here separates 'menin inhibition evicts the fusion' from 'the on-treatment aliquot is a different, largely non-leukaemic cell population'. A drug that kills blasts changes what is in the tube, and a 85% drop in occupancy calls is equally consistent with a purity change.

Histone-cluster membership is assigned by symbol prefix (H1-, H2A, H2B, H3C, H4C, H3-), which is a regex over names rather than a curated locus list and will mis-assign any gene whose symbol happens to start that way. The permutation tests only whether the retained set is a random draw from the untreated occupied set; it does not test whether histone clusters are enriched among occupied loci in the first place, which is stated as a raw ratio and not as a p-value. n = 1 pair throughout.

• Histone gene clusters are short, extremely highly transcribed, replication-dependent and sit in dense repetitive arrays - all properties that inflate coverage-based occupancy scores independently of any oncoprotein. Their over-representation may be entirely a mappability and transcription artefact, in which case this says nothing about menin and everything about the rule, which is arguably the more useful reading.
• Conversely, MLL1 has genuine biology at histone loci and the NSL/MOF complex acts there, so preferential retention could be real wild-type MLL1 activity that the menin inhibitor does not touch - the KMT2A-N antibody sees fusion and wild-type together. Distinguishing artefact from wild-type MLL1 needs the KMT2A-C arm, which was not scored here.

Histone-cluster membership is assigned by symbol prefix (H1-, H2A, H2B, H3C, H4C, H3-), a regex over names rather than a curated locus list. The per-sample spread is very wide - TB000694 has 0 of 54 calls and A80956 7 of 103, against 48.3% at the top - and that spread is unexplained here; depth, chromatin quality and blacklist overlap were not tested as explanations. The pooled 31.3% weights samples by their call count, so it is dominated by the high-call samples.

• A 68-fold enrichment is not automatically an artefact. Histone gene clusters are genuine MLL1 and NSL/MOF territory, and the KMT2A-N antibody sees wild-type MLL1 alongside the fusion, so a real wild-type signal would look exactly like this. Separating the two needs the KMT2A-C arm, which is not scored here.
• The variance across samples argues against a pure mappability artefact, which should be roughly constant per library. A range from 0.0% to 48.3% suggests something sample-specific - proliferation rate would be the obvious candidate, since replication-dependent histone transcription tracks S-phase fraction - and that is a biological explanation this analysis did not test.