The prey-only interactome is 36.3% written, not 0% - the census, by identity, of all 212 human prey-only partners of the six core proteins

Per-bait record counts: KMT2A 315, AFF4 171, MEN1 136, PSIP1 114, AFF1 81, DOT1L 62.

The corpus is being written concurrently and the denominator of node files moves: 2,927 at the moment of the join, 2,938 excluding this beat a short time later. The coverage FRACTION is stable under that (a new entry would have to be one of these 135 partners to move it) but the node count should be read with its timestamp. The presence test is deliberately generous, the same three-way join the directed-evidence recheck uses, so 135 is a ceiling on the gap and 77 a floor on the coverage. It is also too generous in one direction and that is corrected in a sibling entry: a no match from a symbol-keyed join is not a gap, and 30 of the 135 turn out to be the subject of an existing entry under the interface-<bait>-<partner>-<reason> naming convention.

The 0% of the prey-only half of the claim comes from cycle 34, in the same sentence as the 100% of the directed-evidence half. The directed half has been re-measured four times since (100% → 68.5% → 94.3% → 51/51). The prey-only half has never been re-measured at all, because the sentence's rhetorical work was done by the first half: it justified not dispatching an interactome beat, and once no beat runs, nothing goes looking.

There is a second reason and it is structural. Between cycle 34 and now, at least five beats wrote prey-only partners without calling them that - q126-the-partners-named-but-never-written, q128-the-menin-interactome-that-is-not-chromatin, q129-the-never-discussed-partners-of-kmt2a-and-psip1, q132-interactome-tail-promotion, q136-interactome-tail-size-and-abundance. The 77 covered partners are largely their output. The coverage figure did not go stale because the work was not done; it went stale because the work was done under a different name. That is a different failure from rule 90's and needs a different fix: a coverage line has to be regenerated from the id set, not from a memory of which beats ran.

Both of this project's interactome analyses query PSICQUIC as id:<accession>. This one queries the IntAct REST findInteractions endpoint and additionally pulls Q03164-PRO_0000390949 and Q03164-PRO_0000390950, the taspase1 N320 and C180 chains.

The coverage percentage FALLS when the better instrument is used, while the graph got strictly better - exactly the direction rule 93 warns about. Both figures are reported. The 36.3% is the one to quote, because 43.1% is a percentage of a population the query could not fully see.

One grade actually flips. Q04206 (RELA/p65) is prey-only in the PSICQUIC grading and has a bait record in REST. A protein graded prey-only on a partial record set is not conservatively graded - it is over-graded, in the direction that makes the prey-only population look larger than it is.

This census is bounded by IntAct curation of the six named human proteins. It therefore cannot see:

1. Any interactor of the FUSION protein as such. Every bait here is a wild-type component. analyses/directed-interactome-coverage-recheck had to exclude Q8TD98 ('MLL/AF4 fusion protein', 624 aa) and Q2Q440 ('MLL/ENL fusion protein', 20 aa) from its denominator as constructs rather than partners. Those two accessions are the only trace of the fusion itself in this record set, and one of them is a twenty-residue junction fragment.
2. Anything in a paper nobody curated. IntAct coverage of a 1990s co-IP is patchy.
3. Anything where our protein is prey to a bait outside the six and the depositor filed it under the bait only.

So the honest framing of the PI's first priority is not the prey-only tail is unwritten. It is: the tail is two-thirds unwritten, one flipped grade shows the grading itself is instrument-dependent, and no bait in the entire record set is the oncoprotein.