Menin inhibition removes the fusion from its strongest sites without relocating it to a new programme, and PRC1.1 fills in only where the eviction is complete - so the drug's chromatin action is subtraction plus local substitution, not redistribution

[
  "the 52 gained bins are not CpG-island-enriched against expression- and input-matched bins",
  "a matched-input peak call on re-aligned reads finds no reproducible treated-only peak set",
  "PARTLY CONFIRMED, PARTLY REFUTED, cycle 76. Original: \"the same within-sample analysis on human GSE127508 gives large loss at the strongest sites and no coherent gained set\" LARGE LOSS AT THE STRONGEST SITES: CONFIRMED and quantitatively near-identical, in the menin channel - 14.909 -> 1.957 fold over own median against this beat's murine 15.9 -> 3.0, share ratio 0.1999 against 0.179, 91.51% losing rank against 92.3%, per-bin median 0.1333 against 0.174. NO COHERENT GAINED SET: REFUTED as stated. 53 gained bins in the MLL1-N channel reproduce in a second line against a matched null at p = 5.0e-4, which is more coherence than 'no coherent gained set' allows - though they are not at TSS (17.0% versus 76.1% for held), not islands, not blacklist-adjacent, and carry BELOW-average H3K79me2 (0.64x), so they are not newly transcribed either. The honest summary is that the loss half of this hypothesis replicates in the human index fusion and the no-redistribution half replicates only in the channel where redistribution was never measurable.",
  "H3K79me2 loss under VTP50469 is quartile-graded and tracks the RING1B gain, while under SGC0946 both are flat",
  "in a PCGF1 or BCOR knockout the fusion is evicted normally but the vacated sites are not silenced and the programme returns on drug withdrawal  ||  CASHED (Q125, cycle 88; recorded cycle 90): PARTLY REFUTED AND PARTLY CONFIRMED, RUN IN CYCLE 88 (Q125). Original: \"In a PCGF1 or BCOR knockout, the fusion is evicted normally but the vacated sites are not silenced, and the programme returns on drug withdrawal.\" SCORED: only the SILENCING clause, only for PCGF1, only in murine MLL-AF9 (GSE260742, cell line MLLAF9, myeloid, blood, taxid 10090), only at 2 kb resolution on the FLAG-defined top-0.1% set of U_m (1,228 bins), 500 nM VTP-50469 48 h against DMSO 72 h. (1) THE PREMISE ABOUT THE CONTROL IS CONFIRMED AND HAD NEVER BEEN MEASURED: in a normal cell the vacated sites ARE silenced, and the effect is graded by eviction depth - across Q117's four eviction quartiles the H2AK119ub share ratio runs 1.4450/1.1483/1.1238/1.0630 (paired) and 1.0151/0.9063/0.8878/0.8584 (rep2), H3K27me3 2.0007/1.5120/1.1304/0.9780, all Spearman +1.000 against eviction depth, while H3K4me3 runs 0.9491/1.1298/1.1761/1.2025, Spearman -1.000, against DMSO-decile-matched nulls flat to within 0.007. (2) THE CLAUSE AS WRITTEN IS REFUTED FOR THE PCGF1 KNOCKOUT: the vacated sites are STILL silenced there, just LESS - H2AK119ub at the most-evicted quartile 1.4450 -> 1.1325 and 1.0151 -> 0.8842, H3K27me3 2.0007 -> 1.0134 - and the knockout's residual gain is still eviction-graded (Spearman +1.000, paired). The correct statement is ATTENUATED, NOT ABOLISHED, which is what a PCGF1 knockout should look like when the antibody (NEB 8240S) reads H2AK119ub from any PRC1 and PCGF1 is one of six PCGF paralogs. (3) AGGREGATED over all 1,228 bins the same comparison is UNDECIDED (|R_KO - R_NT| = 0.0813 against thresholds of 0.0584 and 0.2089) and the control's own aggregate effect is 1.0383, i.e. essentially nothing - the gradient is what carries the result and the aggregate averages it away. (4) NOT SCORED: 'the fusion is evicted normally' - no FLAG, MLL1 or MENIN track exists in the NTvKO arm by title, although GSM8123407-412 carry ChIP_FLAG filenames AND a ChIP_FLAG string in their own TDF headers while being titled H2AK119ub, so this clause MAY be reachable from this deposit and the channel could not be settled. (5) NOT SCORED AND UNREACHABLE FROM THE WHOLE SERIES: 'the programme returns on drug withdrawal' - no washout arm exists in any of the 158 records. (6) NOT SCORED: 'or BCOR' - the series' only genetic perturbation is the PCGF1 knockout. RESIDUAL CAVEATS: no spike-in, so every number is a within-sample share blind to uniform change; VTP is 48 h against a 72 h vehicle so every ratio is a lower bound; the knockout is one guide, one clone, no rescue and no deposited PCGF1 measurement; and no transcriptional readout exists for these cells, so 'silenced' means two repressive marks up and one active mark down. Evidence: ana-q125-the-menin-inhibitor-does-silence-the-sites-it-evicts-the-fusion-from-and-the-effect-is-graded-by-how-much-fusion-left, ana-q125-losing-pcgf1-reduces-the-silencing-of-the-vacated-sites-without-abolishing-it-in-both-repressive-marks."
]

[
  "exp-q117-spike-in-normalised-tagged-fusion-chip-under-a-menin-inhibitor-in-a-human-kmt2a-aff1-line"
]

[
  "ana-q117-the-pre-registered-gain-criteria-fire-on-52-low-amplitude-bins-with-no-coherent-identity",
  "ana-q117-the-residual-fusion-occupancy-fraction-under-a-menin-inhibitor-is-about-18-percent",
  "ana-q117-prc1-fills-in-uniformly-under-the-dot1l-inhibitor-and-only-where-the-fusion-leaves-under-the-menin-inhibitor",
  "ana-q117-the-cpg-island-test-settles-it-66-percent-of-the-fusions-real-sites-are-islands-and-2-percent-of-the-gained-set-is"
]

[
  "REDISTRIBUTION, CpG ARM. If the 52 VTP50469-specific gained bins in murine MLLAF9 (GSE260742, 2 kb bins, mm10) overlap annotated CpG islands at more than 2-fold the rate of bins matched on input coverage decile and DMSO FLAG percentile decile, with at least 20 of 52 overlapping a CGI, the fusion is redistributing onto free CXXC substrate and the no-redistribution half of this hypothesis is dead. RUN IN CYCLE 68 - DOES NOT FIRE. Against the UCSC mm10 cpgIslandExt track (15,997 islands on the analysed chromosomes), 1 of the 52 gained bins overlaps an island (1.92%) against a class-matched null rate of 1.42% over 2,600 draws - fold 1.35, Fisher exact two-sided p = 0.53 - while 806 of the 1,228 fusion-held bins (65.6%) overlap one, against a universe background of 1.62%. The gained set does not carry the CXXC signature and the redistribution model is refuted on its own predicted substrate. See ana-q117-the-cpg-island-test-settles-it-66-percent-of-the-fusions-real-sites-are-islands-and-2-percent-of-the-gained-set-is.",
  "REDISTRIBUTION, PEAK ARM. If a MACS2 narrow-peak call on the re-aligned GSE260742 FLAG arm (SRX23818722 DMSO versus SRX23818724 VTP50469, matched input GSM8123413/8123414, mm10) returns more than 200 treated-only peaks at q < 0.05 that are absent in DMSO and enriched at least 4-fold over input, the low-amplitude reading in this hypothesis is wrong. NOT YET RUN. **AMENDED 2026-09-23 (cycle 77), BEFORE RUNNING, TO REMOVE A THRESHOLD AMBIGUITY - the criterion is NOT relaxed or tightened, it is made reading-independent.** As written, 'more than 200 treated-only peaks' is a bare count with no denominator: MACS2's total yield depends on read depth, the q cutoff and the input, none of which this clause fixes, so 200 could be 0.2% or 20% of the call and the verdict would differ. A corpus sweep this cycle (`analyses/count-threshold-transferability/`) found this is the ONLY live instance of that defect in any scored criterion in the graph. **SCORE IT ON BOTH CLAUSES AND REPORT BOTH: (a) more than 200 treated-only peaks at q < 0.05, AND (b) treated-only peaks exceeding 5% of the DMSO peak count in the same call.** The 5% figure is chosen to match the intent of the original number rather than to preserve its arithmetic: at Q117's own observed scale the gained set was 52 bins against 1,228 held (4.2%), so a redistribution worth the name should clear 5% of the DMSO call. **If (a) and (b) disagree, the peak arm is UNDECIDED and says so - it does not get to pick the clause that fires.** Report the DMSO peak count with the result either way, because without it neither clause can be re-scored by anyone else.",
  "HUMAN ARM. If the identical within-sample analysis on GSE127508 RS4;11 (human KMT2A::AFF1, MLL1n ChIP, VTP-50469 versus DMSO at day 3) yields a gained set whose median enrichment exceeds 8-fold over that track's genome median, then the index fusion does redistribute even though murine MLLAF9 does not, and this hypothesis does not generalise to the disease the lab studies. **RUN IN CYCLE 76 - THE NUMERIC CONDITION FIRES AND THE INFERENCE DOES NOT FOLLOW.** Q118 ran the identical within-sample analysis on GSE127507 - which is not a substitution: GSE127507's own record carries `!Series_relation = SubSeries of: GSE127508`, so these are the samples this falsifier named. RS4;11, MLL1n ChIP, VTP-50469 versus DMSO at day 3, 2 kb bins, hg19, universe 1,389,514. **THE GAINED SET'S MEDIAN ENRICHMENT IS 15.276-FOLD OVER THAT TRACK'S GENOME MEDIAN, AGAINST THE 8-FOLD THRESHOLD THIS FALSIFIER SET - SO AS WRITTEN, IT FIRES**, on 53 bins that also reproduce in MOLM13 at the same coordinates (1.7716, input-corrected 1.7516, against a class-matched null of 1.0134 over 2,000 draws, one-sided p = 5.0e-4, MOLM13 input flat at 1.0115). **BUT THE CONSEQUENT - 'the index fusion does redistribute' - IS NOT LICENSED BY THE REAGENT THIS FALSIFIER CHOSE, AND THE SAME BEAT IS WHAT ESTABLISHED THAT.** MLL1n is Bethyl A300-086A, an MLL1 N-TERMINUS epitope retained in the fusion, so it reads fusion and wild-type MLL1 together; Q118 measured the masking at D = 3.54x and 4.00x (MENIN 0.1999-0.2560 versus MLL1-N 0.7996-0.9069 at the same bins). Worse for the inference: **the MLL1-N channel is barely evicted at all** - held share ratio 0.9069, per-bin median 0.8196, only 18.7% of held bins below 0.5 - so 'gained' bins in that channel are not the vacancy-and-relocation this hypothesis is about. **IN THE CHANNEL THAT IS ACTUALLY EVICTED, THE FALSIFIER WOULD NOT FIRE: THE MENIN GAINED SET'S MEDIAN ENRICHMENT IS 5.058-FOLD, BELOW THE 8-FOLD THRESHOLD** (menin held share ratio 0.1999, 91.51% losing rank). **THEREFORE THE PRIOR IS NOT MOVED.** Firing a falsifier whose consequent its own reagent cannot support is not evidence about the fusion, and lowering the prior on it would import a wild-type MLL1 signal as a fusion result. What this outcome does establish is that THIS FALSIFIER WAS MIS-SPECIFIED AT WRITING: it named a non-fusion-specific antibody for a claim about the fusion, and no number from that antibody could ever have settled it. The replacement is `exp-q118-a-tagged-kmt2a-aff1-chip-in-rs4-11-under-vtp-50469`, whose decisive form is a share ratio below 0.35 at the tag's own top 0.1%. See ana-q118-the-mll1-n-antibody-masks-the-occupancy-collapse-four-fold-in-rs4-11 and concept-q118-an-mll1-n-terminus-chip-cannot-attribute-occupancy-to-the-fusion.",
  "SUBSTITUTION ARM. If RING1B's within-sample share ratio at fusion-held bins under VTP50469 is within 0.1 of its value in the least-evicted quartile - that is, if the Q1-versus-Q4 difference of 1.38 against 0.88 does not reproduce - in a second replicate or a second dataset, the displacement-coupled half of this hypothesis is refuted and the depositing paper's 'independent of MLL-FP displacement' stands unqualified. NOT YET RUN; no replicate exists in GSE260742.",
  "DEPLETION ARM. If an immunoblot of FLAG-MLLAF9 in the same cells at 48 h of VTP50469 500 nM shows a reduction in total fusion protein of more than 50%, then the occupancy loss measured here is at least partly depletion rather than eviction and the word 'evicts' in this hypothesis is wrong. NOT YET RUN; the depositing paper may already contain this blot and it was not examined."
]

{
  "verdict": "NUMERIC CONDITION FIRES; CONSEQUENT NOT LICENSED; PRIOR UNMOVED",
  "measured_mll1n_gained_median_enrichment": 15.276,
  "threshold_this_falsifier_set": 8,
  "measured_menin_gained_median_enrichment": 5.058,
  "why_the_two_differ": "MLL1n is an N-terminus antibody retained in the fusion and its held set is barely evicted (share ratio 0.9069); MENIN is the channel the drug acts on and is evicted to 0.1999. The falsifier read the channel that does not move.",
  "series_check_I_ran_myself": "the falsifier names GSE127508 and Q118 measured GSE127507. NOT a substitution: GSE127507 carries !Series_relation = SubSeries of: GSE127508. I checked this before raising it, having first mis-read it as a mismatch.",
  "reagent_check_I_ran_myself": "the falsifier explicitly names `MLL1n ChIP`, so Q118's use of the MLL1-N arm is what was asked for. I had suspected an undeclared antibody switch and was wrong; reading the falsifier's own text settled it.",
  "source": "analyses/q118-menin-at-the-human-index-fusion/results-instrumentB.json and replication-matched-null.json, both re-read by the king",
  "the_general_lesson": "A FALSIFIER IS ONLY AS GOOD AS THE REAGENT IT NAMES. This one was scoreable, cheap and pre-registered - it satisfied every process rule this project has - and it still could not settle its own question, because the antibody in its own text cannot attribute occupancy to the fusion. Pre-registration disciplines the ANALYSIS; it does nothing about whether the instrument can see the thing. Name the reagent AND state what it cannot distinguish, in the falsifier, at writing time."
}

{
  "what_changed": "a proportional clause (b) was added beside the bare count (a); neither replaces the other and disagreement between them is a reportable UNDECIDED",
  "why_it_is_legitimate": "the falsifier is uncashed. An uncashed falsifier can be sharpened; a cashed one cannot, and the HUMAN ARM cashed in cycle 76 was not touched",
  "the_defect": "'more than 200 treated-only peaks' fixes no denominator, and MACS2's total yield varies with depth, q cutoff and input - so the same biology could pass or fail depending on the call's size",
  "how_it_was_found": "analyses/count-threshold-transferability/run.py, which walked 7,647 falsifier/prediction/criteria/section-body instances across 2,301 entries and found 5 scored criteria carrying a count threshold, of which this is the only one that is both fragile and live",
  "source": "cycle 77"
}