Analysismoderateking

Half this graph's claims are unopposed; these are the ones that matter

also called unopposed claim load · counter-evidence work list

About half the individual statements in this project have nothing published arguing against them. Working through them in file order treats them all as equally important. This ranks them by how much else would move if one turned out to be wrong.

The same thing, for a scientist

1394 of 2917 evidence items carry no counter-evidence. Each is scored as log1p(edge degree of its entry) x log1p(entries citing the same PMID) x confidence weight x a risk multiplier for claim shapes this project has been wrong about before (absences, counted absences, universals, uniqueness, necessity/sufficiency). The top is dominated by hub entries resting on one referral series: **`fusion-kmt2a-aff1#E1` scores 97.9 at degree 131, citing a paper that 50 other entries also cite.**

Go deeper

The work list

#	score	entry	item	deg	reuse	risk
1	51.145	`resistance-men1-interface-mutations`	E2	27	12	universal, uniqueness
2	50.789	`cell-type-fetal-liver-hsc`	E1	19	16	universal, uniqueness
3	48.083	`fusion-kmt2a-mllt3`	E1	10	50	universal
4	42.295	`concept-phase2-go-no-go-register`	E3	23	15	uniqueness
5	41.468	`epi-kmt2a-rearranged-acute-leukemia-all-ages`	E5	8	50	uniqueness
6	40.677	`fusion-kmt2a-aff1`	E3	132	15	-
7	40.003	`mech-breakpoint-position-consequences`	E_corr	17	50	uniqueness
8	39.127	`risk-revumenib-intact-pth-elevation`	E2	12	23	uniqueness
9	39.02	`fusion-kmt2a-mllt1`	E1	6	50	universal
10	37.81	`comut-low-mutation-burden-infant-kmt2a-r`	E7	12	17	universal
11	37.483	`ana-kmt2a-lost-module-ledger`	E2	20	12	uniqueness
12	36.541	`fusion-kmt2a-afdn`	E1	11	50	universal

The scoring is multiplicative on purpose. A high-degree entry whose claim is a universal, resting on a paper the graph reuses in fifty places, is the expensive case: if it moves, everything downstream moves with it. A leaf entry making a hedged claim from a paper cited once is cheap to leave alone.

Reuse = 50 recurs at the top and is the single largest dependency in the graph. One diagnostic referral series underpins fifty entries. That is not a criticism of the paper, which is the field's reference catalogue, but a concentration worth knowing about.

Where the unopposed claims are

folder	unopposed items
`comparative`	162
`atlas2`	124
`mechanism-core`	97
`cells`	89
`interactome2`	75
`interactome`	72
`subtraction-kmt2a`	71
`impact`	60
`localization`	56
`analogy`	53

This is deliberately NOT the same ordering as the coverage ledger's percentage column. A beat can sit at 30% and matter less than a beat at 60% carrying more load, which is why comparative and atlas2 head this table while the ledger's worst percentages sit elsewhere.

A fifth of the remaining gap is not opposable in the usual sense

Measured on 2026-09-20 across 3,144 evidence items: 950 are unopposed, and 197 of those (20.7%) are definitional, statutory or registry-shaped — what a statute says, what a fee is, what a UniProt feature table places, what a PDB entry contains, what a registry record lists.

By citation type the pattern is sharp:

citation type	items	unopposed
paper	2,294	28%
database	523	36%
review	91	48%
trial_registry	70	53%
news	19	74%

This is not a licence to skip them. A statutory or registry claim can carry real counter-evidence, but it has to address what the claim CANNOT support rather than contradict its content — for example, that a registry outcome list is the protocol's prespecified measures rather than everything collected, so absence of an endocrine endpoint means it was not powered, not that the data do not exist. That note changes what a reader may infer and is worth writing.

What it does mean is that a blanket 100% item-level target is the wrong instrument for beats dominated by this class. Q8 impact is the clearest case: roughly half its remaining items are statutory, and driving it to 100% by the usual method would manufacture filler. The beat was taken to 56% by opposing the empirical subset and stopped there deliberately.

Facts and figures

script	analyses/unopposed-claim-load/run.py
status	executed
run date	2026-09-19
n evidence items	2917
n unopposed	1394
n scored	1326
n skipped by type	68
headline	1394 of 2917 items unopposed; the top of the list is hub entries resting on one referral series cited by 50 entries.
caveat	A triage heuristic, not a measure of truth. A high score means the claim is EXPENSIVE if wrong, not that it is likely wrong. Nothing here reads a paper, and the ranking cannot distinguish a claim nobody has opposed from one nobody could oppose - a definition, a registry fact, an arithmetic result. Those are noise in the list and a reader has to skip them. Edge degree also rewards entries that happen to be well connected, which correlates with being old rather than with being load-bearing.

unopposed by folder

top

unopposed by folder — the numbers behind the chart

{
  "comparative": 162,
  "atlas2": 124,
  "mechanism-core": 97,
  "cells": 89,
  "interactome2": 75,
  "interactome": 72,
  "subtraction-kmt2a": 71,
  "impact": 60,
  "localization": 56,
  "analogy": 53,
  "genome": 52,
  "endocrine": 51
}

top — the numbers behind the chart

[
  {
    "score": 51.145,
    "entry": "resistance-men1-interface-mutations",
    "evidence_id": "E2",
    "edge_degree": 27,
    "entries_sharing_pmid": 12,
    "risk_tags": [
      "universal",
      "uniqueness"
    ]
  },
  {
    "score": 50.789,
    "entry": "cell-type-fetal-liver-hsc",
    "evidence_id": "E1",
    "edge_degree": 19,
    "entries_sharing_pmid": 16,
    "risk_tags": [
      "universal",
      "uniqueness"
    ]
  },
  {
    "score": 48.083,
    "entry": "fusion-kmt2a-mllt3",
    "evidence_id": "E1",
    "edge_degree": 10,
    "entries_sharing_pmid": 50,
    "risk_tags": [
      "universal"
    ]
  },
  {
    "score": 42.295,
    "entry": "concept-phase2-go-no-go-register",
    "evidence_id": "E3",
    "edge_degree": 23,
    "entries_sharing_pmid": 15,
    "risk_tags": [
      "uniqueness"
    ]
  },
  {
    "score": 41.468,
    "entry": "epi-kmt2a-rearranged-acute-leukemia-all-ages",
    "evidence_id": "E5",
    "edge_degree": 8,
    "entries_sharing_pmid": 50,
    "risk_tags": [
      "uniqueness"
    ]
  },
  {
    "score": 40.677,
    "entry": "fusion-kmt2a-aff1",
    "evidence_id": "E3",
    "edge_degree": 132,
    "entries_sharing_pmid": 15,
    "risk_tags": []
  },
  {
    "score": 40.003,
    "entry": "mech-breakpoint-position-consequences",
    "evidence_id": "E_corr",
    "edge_degree": 17,
    "entries_sharing_pmid": 50,
    "risk_tags": [
      "uniqueness"
    ]
  },
  {
    "score": 39.127,
    "entry": "risk-revumenib-intact-pth-elevation",
    "evidence_id": "E2",
    "edge_degree": 12,
    "entries_sharing_pmid": 23,
    "risk_tags": [
      "uniqueness"
    ]
  },
  {
    "score": 39.02,
    "entry": "fusion-kmt2a-mllt1",
    "evidence_id": "E1",
    "edge_degree": 6,
    "entries_sharing_pmid": 50,
    "risk_tags": [
      "universal"
    ]
  },
  {
    "score": 37.81,
    "entry": "comut-low-mutation-burden-infant-kmt2a-r",
    "evidence_id": "E7",
    "edge_degree": 12,
    "entries_sharing_pmid": 17,
    "risk_tags": [
      "universal"
    ]
  },
  {
    "score": 37.483,
    "entry": "ana-kmt2a-lost-module-ledger",
    "evidence_id": "E2",
    "edge_degree": 20,
    "entries_sharing_pmid": 12,
    "risk_tags": [
      "uniqueness"
    ]
  },
  {
    "score": 36.541,
    "entry": "fusion-kmt2a-afdn",
    "evidence_id": "E1",
    "edge_degree": 11,
    "entries_sharing_pmid": 50,
    "risk_tags": [
      "universal"
    ]
  },
  {
    "score": 35.929,
    "entry": "breakpoint-kmt2a-spliced-fusion-partner-exon-2",
    "evidence_id": "E2",
    "edge_degree": 5,
    "entries_sharing_pmid": 50,
    "risk_tags": [
      "universal"
    ]
  },
  {
    "score": 35.912,
    "entry": "ana-kmt2a-lost-module-ledger",
    "evidence_id": "E4",
    "edge_degree": 20,
    "entries_sharing_pmid": 50,
    "risk_tags": []
  },
  {
    "score": 35.572,
    "entry": "mechanism-kmt2a-subnuclear-localization",
    "evidence_id": "E2",
    "edge_degree": 16,
    "entries_sharing_pmid": 10,
    "risk_tags": [
      "universal",
      "necessity/sufficiency"
    ]
  },
  {
    "score": 35.261,
    "entry": "hyp-aff1-chd-dimerisation-in-kmt2a-fusion",
    "evidence_id": "E3",
    "edge_degree": 10,
    "entries_sharing_pmid": 50,
    "risk_tags": [
      "universal"
    ]
  },
  {
    "score": 34.811,
    "entry": "mechanism-sec-elongation-hijack",
    "evidence_id": "E1",
    "edge_degree": 10,
    "entries_sharing_pmid": 15,
    "risk_tags": [
      "universal",
      "necessity/sufficiency"
    ]
  },
  {
    "score": 34.731,
    "entry": "gene-wdr5",
    "evidence_id": "E3",
    "edge_degree": 18,
    "entries_sharing_pmid": 50,
    "risk_tags": []
  },
  {
    "score": 32.513,
    "entry": "hyp-oligomerisation-interfaces-as-mutation-resistant-targets",
    "evidence_id": "E4",
    "edge_degree": 22,
    "entries_sharing_pmid": 15,
    "risk_tags": [
      "universal"
    ]
  },
  {
    "score": 32.513,
    "entry": "mechanism-chromatin-elongation-hijack",
    "evidence_id": "E1",
    "edge_degree": 22,
    "entries_sharing_pmid": 15,
    "risk_tags": [
      "universal"
    ]
  },
  {
    "score": 31.706,
    "entry": "gene-men1",
    "evidence_id": "E5",
    "edge_degree": 87,
    "entries_sharing_pmid": 24,
    "risk_tags": []
  },
  {
    "score": 31.579,
    "entry": "subtraction-kmt2a-win-wrad-interface",
    "evidence_id": "E3",
    "edge_degree": 12,
    "entries_sharing_pmid": 12,
    "risk_tags": [
      "uniqueness"
    ]
  },
  {
    "score": 31.461,
    "entry": "drug-revumenib",
    "evidence_id": "E4",
    "edge_degree": 89,
    "entries_sharing_pmid": 23,
    "risk_tags": []
  },
  {
    "score": 30.921,
    "entry": "cell-type-fetal-preprob-cd10-negative",
    "evidence_id": "E1",
    "edge_degree": 11,
    "entries_sharing_pmid": 7,
    "risk_tags": [
      "universal",
      "uniqueness"
    ]
  },
  {
    "score": 30.255,
    "entry": "fusion-kmt2a-mllt10",
    "evidence_id": "E1",
    "edge_degree": 12,
    "entries_sharing_pmid": 50,
    "risk_tags": []
  }
]

opposed this cycle

[
  "fusion-kmt2a-aff1#E1",
  "disease-infant-all#E1",
  "concept-cytogenetics-karyotype-kmt2a#E1",
  "mechanism-kmt2a-subnuclear-localization#E5",
  "mechanism-kmt2a-r-tissue-localization#E5",
  "gene-kmt2a#E1"
]

How this could bite us back

This analysis makes the counter-evidence percentage easier to raise, which is not the same as making the graph more reliable. Opposing six high-degree claims moved the item-level number barely at all while leaving 1,400 others untouched.
It also ranks by age dressed up as importance: edge degree accumulates over cycles, so the newest and least scrutinised beats sit at the bottom of a list that claims to surface the riskiest claims.
Measuring that a fifth of the gap is hard to oppose is exactly the kind of finding that can be used to stop working on it. The Q15 sweep this cycle closed thirteen registry and label items with substantive notes, so the class is tractable and the measurement is a caution about method, not a permission to stop.

Still open

Does a claim's score predict anything? The test is prospective: record the scores now, and when a claim in this graph is next found to be wrong, check where it sat in this ranking.
Fifty entries cite one referral series. Is there a second, independent source for the partner frequencies many of those entries depend on, or is the concentration unavoidable?
Should entries be REQUIRED to carry item-level counter-evidence before they can be linked, rather than opposed afterwards? That would stop the tail forming but slows every beat.
Roughly 20% of unopposed items are definitional or statutory. Should the coverage ledger report item-level coverage over the OPPOSABLE denominator, or keep the raw one and footnote the class? The raw number is honest and the ceiling it implies is not 100%.

Every source behind this entry (2)

E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.analysis
Scoring every unopposed evidence item by the edge degree of its entry, the number of entries sharing its citation, its asserted confidence and a risk multiplier for absence, universal, uniqueness and necessity claims ranks 1394 unopposed items of 2917 total. The highest scores concentrate on hub entries citing one referral series: fusion-kmt2a-aff1#E1 at 97.9 (degree 131, 50 entries sharing its PMID), disease-infant-all#E1 at 73.5 and concept-cytogenetics-karyotype-kmt2a#E1 at 70.5. The six highest-ranked items were opposed on the strength of this ranking in the same cycle.
Load-ranking of unopposed claims in the Oncofusion knowledge graph · Oncofusion knowledge graph (internal analysis) 2026
What this evidence cannot show
Run by the coordinator on 2026-09-19 over 654 entries. A triage heuristic, not a measure of truth: a high score means a claim is EXPENSIVE if wrong, not that it is likely to be wrong. Nothing in it reads a paper. It cannot distinguish a claim nobody has opposed from one nobody could oppose - a definition, a registry fact, an arithmetic result - and those are noise a reader has to skip. 68 items in collaborator, trial and patient-story entries were excluded by type as normally unopposable.
Evidence that argues against this (3)
- Edge degree is a poor proxy for importance and a good proxy for age. Entries written in early cycles have had more chances to be linked, so this ranking systematically promotes old claims over recent ones, and a genuinely load-bearing claim written last cycle scores near the bottom. The newest beats are invisible at the top of this list for that reason alone. Oncofusion knowledge graph (internal analysis) 2026
- The risk multipliers were chosen by the coordinator from this project's own history of errors, not calibrated against anything. Weighting absence claims at 2.2 and universals at 1.7 encodes a belief about which mistakes this project makes; if that belief is wrong the ranking is confidently misordered, and no labelled set of claims that turned out to be wrong exists here to check it against. Oncofusion knowledge graph (internal analysis) 2026
- A ranked work list can make the problem worse. Opposing the top items raises the item-level percentage fastest while leaving the long tail permanently unexamined, and the tail is where a wrong claim sits undisturbed precisely because nothing links to it. Optimising the metric and fixing the graph are not the same task. Oncofusion knowledge graph (internal analysis) 2026
E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.analysis
Of 3,144 evidence items in the graph, 950 are unopposed, and 197 of those (20.7%) are definitional, statutory or registry-shaped claims for which counter-evidence must address scope rather than contradict content. Unopposed fractions differ sharply by citation type: 28% for papers, 36% for database citations, 48% for reviews, 53% for trial-registry records and 74% for news items.
Load-ranking of unopposed claims in the Oncofusion knowledge graph · Oncofusion knowledge graph (internal analysis) 2026
What this evidence cannot show
Counted by the coordinator on 2026-09-20 with a regular expression over claim text plus citation type, so the 20.7% is a heuristic classification and not a reviewed one. It will both over-count (an empirical claim that happens to quote a statute) and under-count (a definitional claim phrased without any of the matched patterns). The citation-type breakdown is exact, since type is a stored field.
Evidence that argues against this (2)
- This risks becoming an excuse. Every one of the registry and label claims opposed in the Q15 sweep this cycle was in the supposedly unopposable class, and each took a substantive note — that a label instruction is a precaution rather than an observed finding, that a FAERS null cannot generate reassurance because the regulator said so in advance. The class is harder to oppose, not impossible, and classifying it invites treating it as exempt. Oncofusion knowledge graph (internal analysis) 2026
- The 74% unopposed rate for news citations points at a different problem than opposability. A news item is a weak source that should usually be replaced by the primary record rather than argued with, so the right action for those 14 items is substitution, and counting them as an opposability problem misdiagnoses it. Oncofusion knowledge graph (internal analysis) 2026

Connected to

← related toWhich partner, in which patient: KMT2A partner frequencies across six age-lineage strata
This entry moved from 2/10 to 10/10 item-level counter-evidence in cycle 41. Every disagreeing-cohort item is now itself bounded, mostly by the detection-method asymmetry (karyotype and FISH versus transcriptome profiling) that generates the disagreements in the first place.
related to →18% of this graph's evidence is the same finding said again
Both measure how much independent support the graph actually has; this one ranks where the gaps cost most.
related to →KMT2A::AFF1 (MLL-AF4), t(4;11)(q21;q23)
Highest-load unopposed claim in the graph, at degree 131 on a source shared by 50 entries.
related to →KMT2A / MLL1 — the gene that breaks
Hub entry whose partner-count claim conflates recurrent drivers with single-patient observations.
related to →Auditing the graph against the PI's brief, not against its own ledger (run)
The coverage ledger reports counter-argument percentage; this says which of the gaps to close first.
← related toOne paper holds up 70 entries, and one dataset holds up 167