An unstated system is not a coin flip: four out of five resolve away from the index disease, so silence in this graph is systematically optimistic
also called provenance optimism · unstated system prior · silence is not neutral
When a paper does not say what cells it used, a reader quietly assumes the cells were relevant. Across sixteen cases we checked, thirteen turned out to be a different fusion, species, lineage or tissue from the disease this project studies. Not knowing is not the same as it being fine - the unknown leans one way, and it leans against us.
The same thing, for a scientist
This is a finding about the graph rather than about the biology. Of sixteen unstated systems resolved this cycle, **thirteen (81%) were imports** - the deposited or recovered system differed from human KMT2A::AFF1 B-lymphoid leukaemia on at least one of fusion, species, lineage or tissue - and typically on two or three at once [E1]. Only three landed on the index fusion, and all three did so through a single cell line, SEM [E1]. The consequence is not that thirteen entries are wrong. It is that **an unstated system has a strongly non-uniform prior, and the graph has been reading it as uniform.** A caveat reading *the fusion is not named in the abstract* is treated by a downstream reader as an ordinary hedge on an otherwise on-target result. The measured base rate says the right default reading is *this is probably mouse MLL-AF9 in a myeloid cell*, because that is what it was in eight of sixteen cases [E1]. The asymmetry has a plain mechanical cause. **An author who is working in the index system usually says so**, because it is the specific and interesting fact about their paper; an author working in a convenient model has less occasion to foreground it, and abstract word limits do the rest [E2]. So the papers that leave the system unstated are enriched for the papers whose system is the generic one. Silence is not missing information distributed at random; it is information that was **omitted because it was unremarkable to its author**, and what is unremarkable to a mouse-AML laboratory is exactly what is decisive to an infant-B-ALL project.
What the measured base rate licenses
| Reading of an unstated system | Status |
|---|---|
| probably fine, the paper is on-topic | not supported - 81% were imports |
| unknown, treat as uninformative | not supported - the distribution is far from uniform |
| probably mouse MLL-AF9 myeloid until shown otherwise | supported by 8 of 16 |
| if it were the index fusion the author would have said so | supported by 3 of 16 and by the mechanism in E2 |
The practical rule: an entry citing a source whose system is unstated should carry the confidence appropriate to an import, not the confidence appropriate to an unknown. That is a one-line change in how such caveats are written, and it is the whole operational content of this concept.
A worked demonstration sits in ana-crispr-chip-fusion-identity-resolved: the keystone H3K79-partitioning evidence, sitting under this project's central hypothesis, read as index-relevant for a full cycle and resolved to murine MLL-AF9. It was not an outlier. It was the modal case.
The two directions this cuts, because it cuts both
Against the graph. Thirteen entries were carrying evidence at a confidence their systems do not support, and in several the import stacks three deep: partner, species and lineage. In two - ana-unstated-system-kdm4c-prmt1-resolved and ana-unstated-system-h2bk34ub-msl2-resolved - the resolution went further than any caveat anticipated, to a non-KMT2A fusion and to a non-haematopoietic tissue respectively.
For the graph. Two of the three index-fusion resolutions were entries whose caveats had been understating evidence this project already had. ana-index-fusion-h3k79-evidence-that-exists exists specifically to answer how much H3K79 evidence do we have in our own fusion?, and it was hedging its best citation - 30 of 59 SEM samples - on the grounds that the abstract did not name the lines 3. A blanket pessimism about unstated systems would have been wrong there.
So the correct posture is neither optimism nor pessimism but resolution: the question is cheap to ask and the answer moves entries in both directions. What is not defensible is leaving it unasked, because leaving it unasked is not neutrality - it is a bet, at 4:1 against.
Where the optimism is invisible: the entries that never wrote a caveat
The 26 sources audited are the ones where somebody noticed and wrote it down. The sweep that found them reads evidence[].caveats and cannot see an entry whose author never registered the gap 1.
That unmeasured population is almost certainly larger, and there is a reason to think it is more import-heavy rather than less: an author who noticed the system was unstated was often an author who had already looked and failed, i.e. one already primed to suspect a problem. An author who did not notice had no such prompt.
So 81% is a measurement on the flagged subset and plausibly a lower bound on the whole. Quantifying the unflagged population would require checking systems for citations that carry no caveat at all - a much larger sweep, and the obvious next beat.
| operational rule | an entry citing a source whose experimental system is unstated should carry the confidence appropriate to an import, not to an unknown, until the system is resolved |
|---|---|
| known blind spot | the sweep reads evidence[].caveats and cannot see an entry whose author never registered the gap; that unflagged population is unmeasured and plausibly more import-heavy |
| worked example | ana-crispr-chip-fusion-identity-resolved - the keystone H3K79 partition read as index-relevant for a full cycle and resolved to murine MLL-AF9 |
| next beat | check systems for citations that carry NO caveat, to estimate the unflagged population |
base rate — the numbers behind the chart
{
"resolutions": 16,
"imports": 13,
"import_fraction_pct": 81.3,
"murine_mll_af9_myeloid": 8,
"index_fusion": 3,
"distinct_index_fusion_systems": 1
}import axes definition
[ "fusion", "species", "lineage", "tissue" ]
- Applying this concept uniformly would lower the confidence of a large number of entries at once, which would make the graph look less certain while making it more accurate - and the institutional pressure runs the other way.
- An 81% figure from n = 16 is exactly the kind of clean number this project has twice been wrong about; quoted without its denominator it will harden into a fact, and the denominator is the flagged subset rather than the literature.
- If the rule is applied as blanket pessimism rather than as a prompt to resolve, it will suppress two genuinely favourable index-fusion findings alongside the thirteen unfavourable ones, and the graph will end up less accurate in the direction nobody is watching.
- What is the import rate among citations that carry no provenance caveat at all? That is the number this concept most needs and the sweep cannot reach it.
- Does the 81% hold in other fusion families' beats, or is it specific to a project whose index fusion is under-modelled relative to MLL-AF9?
- Is there a detectable textual signature - abstract length, journal, presence of a data-availability statement - that predicts whether an unstated system will resolve to an import?
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
Of sixteen unstated experimental systems resolved from depositions or recovered full text this cycle, thirteen differed from human KMT2A::AFF1 B-lymphoid leukaemia on at least one of fusion, species, lineage or tissue, eight were murine MLL-AF9 myeloid systems, and the three that matched the index fusion all did so through the SEM cell line.
MLL-AF4 Spreading Identifies Binding Sites that Are Distinct from Super-Enhancers and that Govern Sensitivity to DOT1L Inhibition in Leukemia (GSE83671) · GEO submitters · NCBI Gene Expression Omnibus 2017What this evidence cannot show
A tally over sixteen resolutions each recorded in its own entry with its accession; the underlying per-source worklist is in ana-unstated-systems-enumeration. n = 16 is a small sample and the 81% figure carries wide uncertainty - a single reclassification moves it by six points. The import definition is this beat's four-axis one and a stricter or looser definition would give a different number. Three resolutions are partial. Most importantly, the denominator is caveats that were WRITTEN, not systems that are unstated.
Evidence that argues against this (1)
- The sixteen sources are not a random sample of the literature this graph cites - they are the sources an author found ambiguous enough to hedge - so the base rate applies to that population and should not be extended to the graph's citations as a whole without the larger sweep. Rogic S et al. 2026
- E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
The asymmetry has a mechanical cause in publication practice: a study in a distinctive or index system foregrounds it, whereas a study in a conventional model has less occasion to name it in an abstract, so sources that leave the system unstated are enriched for conventional models.
MLL-AF4 Spreading Identifies Binding Sites that Are Distinct from Super-Enhancers and that Govern Sensitivity to DOT1L Inhibition in Leukemia. · Kerry J, Godfrey L, Repapi E, Tapia M, Blackledge NP, Ma H, Ballabio E, O'Byrne S, Ponthan F, Heidenreich O, Roy A, Roberts I, Konopleva M, Klose RJ, Geng H, Milne TA · Cell Rep 2017What this evidence cannot show
This is an inference about authorial practice, not a measured effect, and is graded moderate accordingly. It is consistent with the data here - the three index-fusion resolutions all come from papers whose titles or depositions foreground t(4;11) - but consistency is not demonstration, and abstract word limits, journal style and field convention are uncontrolled alternatives that would produce the same pattern.
Evidence that argues against this (1)
- A direct counter-example sits in this beat: the MLL-AF4 spreading paper's title names MLL-AF4 prominently and its abstract still does not name the cell lines, so foregrounding the fusion and omitting the system are compatible and the mechanism is at best partial. Kerry J et al. 2017
- E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
Two of the three index-fusion resolutions corrected caveats that had been understating evidence the graph already possessed, including a hedge on a deposition in which 30 of 59 samples are SEM and another on a deposition whose genotype field reads KMT2A::AFF1 explicitly.
USP22 is a novel vulnerability regulating MEIS1 protein abundance and gene transcription in KMT2Ar acute leukemia (GSE337426) · GEO submitters · NCBI Gene Expression Omnibus 2026What this evidence cannot show
Established from the per-sample GEO records read on 2026-09-19 and detailed in ana-unstated-system-mll-af4-spreading-resolved and ana-unstated-system-usp22-meis1-resolved. Two of sixteen is a small favourable fraction and does not offset thirteen imports; it is recorded because a concept that only ever lowers confidence would be a biased instrument and would be resisted for good reason.
Evidence that argues against this (1)
- Both favourable resolutions run through SEM, a single long-cultured relapse-derived line, so the index-fusion evidence they restore is narrower than the count of samples suggests and should not be read as patient-level support. Greil J et al. 1994
supplies the base rate the concept is built on
gives the measured prior that makes the GEO route worth running by default
the keystone case is the modal case, not an outlier
quantifies the import risk this entry describes qualitatively: 13 of 16
- related to →Import class 2: AHD-dependent mechanism, which cannot transfer because KMT2A::AFF1 has no AHD
the import class now has a measured base rate
- related to →KMT2A::AFF1 (MLL-AF4), t(4;11)(q21;q23)
three of sixteen resolutions reach the index fusion, all through one cell line
the hypothesis whose keystone evidence this class of failure already moved