WITHDRAWN - "thirteen genuinely absent" was wrong: 14 of the 15 are adjudicated in corpus prose and the fifteenth is not an interactor. Superseded by ana-zero-of-fifteen-again-the-same-worklist-the-same-thirteen-and-the-third-recurrence
also called the unwritten dispatch targets are 13 not 15 · H3F3A was decided, not missed · ledger-staleness-audit's worklist, corrected · 13 named interactors with no entry
An old audit listed fifteen genes that were promised as entries and never written. Thirteen really are missing. The other two were deliberately handled a different way, and the entry that handled them explains the choice in its own title - the audit just could not see it.
The same thing, for a scientist
`analyses/ledger-staleness-audit` holds a field `unwritten_dispatch_targets` listing **15 gene ids** past dispatch plans named and no beat delivered. That directory is one of the **5 of 234** analysis directories cited by no entry, so the list has been invisible since it was written. Checked against the live graph, **none of the 15 ids exists** among 2,894 nodes, and each symbol is named in **2 to 12 existing node files** - a naive reading says fifteen gaps. **The naive reading is wrong for two of them.** 🔴 **`H3F3A` and `H3F3B` are not gaps - they were adjudicated.** Both are aliases of two existing entries, `protein-histone-h3-3` and `variant-h3-3`, and the first is titled *"Histone H3.3 - one protein, two genes, and the reason this graph should not file it as `gene-h3f3a`"*. **A past cycle considered the exact id the audit lists as missing, rejected it, and wrote the reason into the title.** Both entries also already carry the current symbols `H3-3A` and `H3-3B` as aliases, so the corpus was never holding a retired name. Writing these two would have produced duplicate entries for one protein and reversed a decision that was correct [E1]. **The thirteen that are genuinely absent**, with no entry under any id, alias or title, each already named in existing entries: | id to write | symbol | GeneID | role | named in N entries | |---|---|---|---|---:| | `gene-il1r1` | IL1R1 | 3554 | IL-1 receptor | 4 | | `gene-tnfrsf1a` | TNFRSF1A | 7132 | TNF receptor 1 | 4 | | `gene-myd88` | MYD88 | 4615 | adaptor | 4 | | `gene-irak1` | IRAK1 | 3654 | kinase | 2 | | `gene-irak4` | IRAK4 | 51135 | kinase | 2 | | `gene-traf6` | TRAF6 | 7189 | E3 ligase | 10 | | `gene-map3k7` | MAP3K7 | 6885 | MAP3K (TAK1) | 5 | | `gene-nfkbib` | NFKBIB | 4793 | IkB-beta | 2 | | `gene-nfkbie` | NFKBIE | 4794 | IkB-epsilon | 2 | | `gene-hdac6` | HDAC6 | 10013 | deacetylase | 4 | | `gene-kat6b` | KAT6B | 23522 | acetyltransferase | 12 | | `gene-nap1l1` | NAP1L1 | 4673 | histone chaperone | 3 | | `gene-sirt6` | SIRT6 | 51548 | sirtuin deacetylase | 5 | 🔵 **And the thirteen are two modules, not thirteen scattered genes.** Nine are one connected chain - the **IL1R/TNFR to NF-kB axis**, from receptor (`IL1R1`, `TNFRSF1A`) through adaptor (`MYD88`), kinases (`IRAK1`, `IRAK4`), E3 (`TRAF6`) and MAP3K (`MAP3K7`/TAK1) to the inhibitors (`NFKBIB`, `NFKBIE`). The other four are **chromatin-modifying enzymes and a chaperone**. **A whole signalling axis was queued and dropped**, which is a different and more tractable gap than thirteen unrelated omissions. 🟠 **This entry is a worklist, not biology.** It establishes which entries are missing and which were already decided. It makes **no claim that any of the thirteen is a KMT2A::AFF1 interactor** - that is what writing them would have to establish from fetched primary sources.
Why the audit could not see the decision, and why agreement did not catch it
The audit tested whether the id gene-h3f3a exists. It does not. But the graph covers that biology under protein-histone-h3-3 and variant-h3-3, with H3F3A as an alias - so an id-existence test cannot distinguish never considered from considered and filed elsewhere. The refusal was not hidden: it is in a title. It was simply not in the namespace the check walked.
🔴 The part worth keeping is how confidently the wrong answer reproduced. My first pass re-derived the mention counts from scratch and got 2, 2, 2, 3, 3, 4, 4, 4, 4, 5, 5, 5, 10, 12 - every one identical to the stored values - and confirmed all 15 ids absent. Perfect agreement with the audit, and still wrong about two of them, because both checks asked the same question. Reproducibility measured the method's consistency, not its validity.
🟠 My own resolution probe was also wrong, in the other direction. A first version reported TNFRSF1A as covered by gene-erg - because I had put the generic alias p55 in the lookup and ERG carries it. Had I trusted it, I would have dropped a genuine gap. Both error directions appeared in one cycle from one loose matcher: too-narrow matching invented two gaps, too-broad matching erased one.
🔵 The cheap general fix: before calling any id a gap, resolve the symbol against the graph's aliases and titles, not just its ids - and treat an alias hit as coverage only after reading the covering entry. Reading protein-histone-h3-3's title was the entire cost of catching this.
| source | analyses/ledger-staleness-audit/results.json, field unwritten_dispatch_targets |
|---|---|
| source was uncited | true |
| n targets listed | 15 |
| n IDS absent from graph | 15 |
| n already adjudicated | 2 |
| n genuinely absent | 13 |
| adjudicating title | Histone H3.3 - one protein, two genes, and the reason this graph should not file it as gene-h3f3a |
| graph node count at check | 2894 |
adjudicated
{
"gene-h3f3a": [
"protein-histone-h3-3",
"variant-h3-3"
],
"gene-h3f3b": [
"protein-histone-h3-3",
"variant-h3-3"
]
}module nfkb il1r tnfr
[ "IL1R1", "TNFRSF1A", "MYD88", "IRAK1", "IRAK4", "TRAF6", "MAP3K7", "NFKBIB", "NFKBIE" ]
module chromatin
[ "HDAC6", "KAT6B", "NAP1L1", "SIRT6" ]
geneids
{
"HDAC6": "10013",
"IL1R1": "3554",
"IRAK1": "3654",
"IRAK4": "51135",
"KAT6B": "23522",
"MAP3K7": "6885",
"MYD88": "4615",
"NAP1L1": "4673",
"NFKBIB": "4793",
"NFKBIE": "4794",
"SIRT6": "51548",
"TNFRSF1A": "7132",
"TRAF6": "7189"
}mentions per symbol — the numbers behind the chart
{
"H3F3A": 5,
"H3F3B": 3,
"HDAC6": 4,
"IL1R1": 4,
"IRAK1": 2,
"IRAK4": 2,
"KAT6B": 12,
"MAP3K7": 5,
"MYD88": 4,
"NAP1L1": 3,
"NFKBIB": 2,
"NFKBIE": 2,
"SIRT6": 5,
"TNFRSF1A": 4,
"TRAF6": 10
}discarded probes
[ "NCBI [All Fields] query returned GeneID 673 (BRAF) for H3F3A and HTTP 429 for 14 of 15", "alias matcher containing the generic alias p55 reported TNFRSF1A as covered by gene-erg" ]
measurements
[
{
"quantity": "named dispatch targets genuinely lacking any entry",
"value": 13,
"unit": "gene entries",
"subject": "the unwritten_dispatch_targets list, after resolving each symbol against node ids, aliases and titles",
"universe": "15 listed; 2,894 node ids in the graph at cycle 142"
},
{
"quantity": "of the listed targets, ones already covered by an existing entry",
"value": 2,
"unit": "gene entries",
"subject": "gene-h3f3a and gene-h3f3b, both aliases of protein-histone-h3-3 and variant-h3-3",
"universe": "15 listed targets; the covering entry's title names gene-h3f3a as the filing it rejects"
}
]- Write the nine IL1R/TNFR-to-NF-kB genes as one beat rather than piecemeal. They are a connected chain and whether the fusion touches that axis at all is a single question with one literature, not nine.
- Re-run the audit that produced this list with alias and title resolution, not id existence. If 2 of 15 of its gap claims were already decided, its other findings - absence_claims_stale and not_checkable - deserve the same re-check before anyone acts on them.
- Record refusals where a checker can read them. The H3.3 decision survived only because someone put it in a title; a machine-readable `decided_against` field would have made the audit correct at no cost.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.
Of the 15 gene ids in analyses/ledger-staleness-audit's unwritten_dispatch_targets, none exists among the graph's 2,894 node ids, but gene-h3f3a and gene-h3f3b are covered by two existing entries, protein-histone-h3-3 and variant-h3-3, which carry H3F3A, H3F3B, H3-3A and H3-3B as aliases and whose title states the graph should not file H3.3 as gene-h3f3a; the remaining 13 symbols match no node id, alias or title and are genuinely absent, each named in 2 to 12 existing node files.
What this evidence cannot show
Coverage was tested by exact match of symbol and known aliases against node ids, entry `aliases` arrays and title prefixes. An entry covering one of the thirteen under a name not in that alias list would still read as absent, so 13 is an upper bound on the true gap. An earlier version of this same probe returned a false positive (gene-erg for TNFRSF1A, matching on the generic alias `p55`), and an earlier NCBI query using `[All Fields]` instead of `[sym]` returned an unrelated gene and 14 HTTP 429s; both were discarded and re-run. The mention counts are word-boundary matches on raw entry text and would over-count a symbol that is also an English word. The audit file carries no `generated` timestamp, so how long these have been outstanding is not established.
Evidence that argues against this (3)
- A dispatch plan naming a gene is not evidence the gene belongs in this graph. The H3F3A case proves the point directly: two of the fifteen were queued and then deliberately not written, and that judgement was right. Some of the remaining thirteen may have been dropped for reasons equally sound and simply not recorded anywhere a checker can read, which is the same defect this entry documents.
- Being named in 2 to 12 entries is weak evidence of importance. KAT6B's 12 mentions may all descend from a single boilerplate list, as one analysis directory in this corpus is name-dropped in 263 entries without any of them depending on it. The counts rank the worklist; they do not justify writing any item on it.
- The two-module reading is post hoc. Nine NF-kB-pathway genes appearing together is unsurprising if a single past dispatch plan proposed that pathway as one beat, in which case the 'axis queued and dropped' is one dropped beat rather than a pattern. The audit file does not record which plan named which target, so this cannot be settled from the evidence here.
third of the five uncited directories written up; its worklist was invisible until that sweep
- refutes →Histone H3.3 - one protein, two genes, and the reason this graph should not file it as gene-h3f3a
this entry already adjudicated gene-h3f3a, which the audit listed as an unwritten gap
second entry carrying H3F3A/H3F3B as aliases, so the audit ids were covered twice over
supersedes the same-cycle claim of thirteen genuine gaps; the true answer is zero