hyp-q117's prediction 5 is stored TWICE in one entry, in sections[0].body_md and in data.predictions[4], and the two strings differ - here is the scored outcome and the ALL-CAPS cashed text proposed for both locations, with the clause-by-clause verdict a one-line 'confirmed' would have hidden

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.

data.falsifiers[3], the SUBSTITUTION ARM, reads: "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...".

It is not cashed by this beat and it is now partly addressable. The falsifier asks about RING1B occupancy; this beat measured the mark RING1B deposits, in the same quartiles, in a different culture of the same line, twice. The Q1-versus-Q4 difference for H2AK119ub is 1.4450 - 1.0630 = 0.3820 (paired) and 1.0151 - 0.8584 = 0.1567 (rep2), against RING1B's 1.3808 - 0.8792 = 0.5016. Both exceed the 0.1 non-reproduction threshold, so the pattern reproduces in a different channel - but a mark is not an occupancy and the falsifier names RING1B, so this is offered as supporting information rather than as the falsifier's outcome. Whether to append it is the coordinator's call and this beat does not assume it.

Six things, each of which changes what a reader should do next:

1. the clause is refuted for the knockout, not confirmed - the sites ARE still silenced there;
2. its premise about the control had never been measured and is the larger new fact;
3. the aggregate says UNDECIDED and the gradient says the answer - a reader quoting the aggregate would reach the opposite conclusion from a reader quoting the quartiles, and both numbers are in this beat's results file;
4. two of three clauses were not scored, one of them unreachable from the entire series;
5. the disjunction was half-tested - BCOR has no knockout anywhere in the deposit;
6. the body_md version's own trailing sentence, 'this beat did not touch it', is now false and will mislead anyone who reads the prediction without the appended outcome.