None of the five selective unwritten interactors survives a null, and FANCD2 - the one that carried a story - is the second weakest of them at p = 0.12

The triage-set difference distribution over 254 genes: min -0.527, Q1 -0.046, median -0.006, Q3 +0.039, max +0.262.

The shortlist was defined as difference at or below -0.15 with pan-essentiality above -0.5. Twenty-two of the 254 clear the difference bar; the pan-essentiality filter cut that to five. So the headline number - five of 254 - is mostly the second filter doing work, and the first filter selects a populated tail rather than an outlier.

This is the failure the project keeps meeting from a new angle. A similarity or consistency score finds candidates and reading separates defect from design; here a selectivity score found candidates and only a null separates signal from tail. Four instruments in four cycles flagged more than was wrong. This is the fifth, and it flagged five things of which at best one is real.

QKI has the best p-value and the smallest claim. Its index mean is -0.122 against +0.035 in other leukaemia lines - it is close to not required at all in the comparator - so the difference is clean and the effect is negligible. A low p on a negligible effect is exactly what a large enough comparator arm buys.

gene-daxx survives as an entry and its evidence item is downgraded. DAXX is still rank 4 of 254 with p = 0.0128, the strongest of the five - and it still does not survive correction, and the entry already said the honest prior is artefact and that it was written so a future cycle could kill it cheaply. It has now been partly killed by the cycle that wrote it, which is the outcome that entry was shaped for. The literature absence it records - zero papers connecting DAXX to this disease - is unaffected, because that was never a dependency claim.

gene-hdgfl2 is unaffected in substance. Its DepMap evidence was already graded moderate and already contradicted by the paper that measured it directly: HRP-2 knockdown reduces colony formation in MLL-germline lines as much as in MLL-rearranged ones. The null agrees with the colony assay. Two independent instruments now say the same thing, and the entry's reason for existing - menin selecting between two IBD paralogues - never rested on selectivity at all.

The FANCD2 convergence loses a leg and should be stated as two legs, not three. The 2003 association paper and the unexplained Fanconi complex signal remain; the dependency number does not. I had sent that number to the menin-interactome agent as context, flagged as weak - it is weaker than flagged, and the correction has been sent.

Every number here rests on eleven KMT2A-rearranged cell lines. This project has already established what those eleven are: nine individuals, of whom the index-fusion arm is three people - a 32-year-old, a 10-year-old and one 5-year-old counted twice - and none is under one year. The lymphoid half of the arm is four lines. And at least ten infant-derived KMT2A::AFF1 lines exist, none of them screened, including one from a nine-day-old.

So a permutation test here answers 'could this allocation of eleven adult and paediatric lines produce this split by chance' and nothing more. It does not make the panel the disease. The right conclusion from this analysis is not that these five genes are uninteresting - it is that a screen with no infant index-fusion line cannot settle a question about infant index-fusion biology, and the correct next move is the screening gap rather than a better statistic on the same eleven lines.

ALL-PO (CVCL_1069, age 4 months) is already a DepMap collection line with methylation, exome, proteomics and RNA-seq and no CRISPR gene-effect value. One screen in a line already in hand.