The room between the Sledgehammer Wing and the Caliper Room. A case waits here — read, scored, held open — until there's enough evidence to know which room it belongs in. Some never resolve. That's not a failure of the room. That's the room doing its job.
The Sledgehammer Wing holds the swings that connected — pillars that came out, floors that genuinely shifted. The Caliper Room holds the swings that missed — anomalies that looked like revolutions and turned out to be measurement problems. Both rooms only make sense in hindsight. The Docket is what the room looks like before hindsight is available: mechanism argued, prediction staked, and nobody — including the authors — yet knows which way it resolves.
It is the exact inverse of the Teachers' Lounge, which holds the settled cases. The Docket holds the unsettled ones. A case graduates out of here one of two ways: independent confirmation moves it to the Sledgehammer Wing, or a sharper instrument dissolves it into the Caliper Room. Until then, it's read here, on its own terms, without borrowing the certainty of either destination.
The Caliper Room's four tells are diagnostic in hindsight. Applied to a Docket case, they become a checklist of what hasn't happened yet:
Does the gap shrink or hold as precision grows?
no 2nd measurement yetIndependent methods, or one un-cross-checked approach?
one group, one paperDoes an unrelated instrument land on the same answer?
the instrument hasn't lookedDoes the claim fall out of the math? Is there a falsifiable prediction on record?
a case can pass this aloneCases wait here together, each read on its own terms:
Knotted solitons in the early universe explain why matter beat antimatter — and stake a gravitational-wave prediction future detectors could test.
Where is the threshold, who set it, and what falls through? The Legibility Problem read across domains — and the catalysis paper it holds open is a Docket case in its own right.
A plant evolved a halogenase from a flavonol synthase across a rugged fitness landscape — and the paper stakes a bet that its own dataset can teach an AI to design new ones.
Catfish melanoma in a Vermont–Quebec lake turns out to be one clonal lineage — the fourth transmissible cancer ever documented, and the first in a fish. But the outbreak began after a flood, and the clonality result never touched that question.
A mechanism claim measured against the evidence offered for it. May turn out to be the “did-it-right” contrast rather than a miss.
A plant evolved a halogenase from a flavonol synthase across a rugged fitness landscape — and the paper stakes a bet that its own dataset can teach an AI to design new ones.
The mechanism half is thorough and largely self-contained. A chromosomal-level genome assembly of Menispermum canadense, structural modeling, molecular dynamics, QM/MM optimization, and site-directed mutagenesis together trace how DAH — dechloroacutumine halogenase, the only characterized halogenase in all land plants — arose from its progenitor flavonol synthase (FLS) by tandem duplication, neofunctionalization, and gene loss. The path runs through two dead-end intermediate pseudogenes still sitting in the genome — DAH-like and DAHy-like — both of which express as insoluble protein, exactly as pseudogenization predicts. The authors describe adaptive peaks “separated by deep fitness valleys.” The landscape really is rugged, and they say so.
What's unsettled is the claim the authors make on their way out the door. The discussion closes on a bet:
That is the exhibit. Not the evolution — the wager riding on it: that this trajectory is a ground-truth resource an AI can learn from. Eleven months on (13 August 2025 → July 2026), nobody — including this lab — has published that follow-through.
No second measurement. No independent lab has reconstructed or re-tested the FLS→DAH trajectory. The team compared against separately published related genomes — but that's a comparison, not a replication.
no 2nd measurement yetOne group, one lineage of papers. The Weng lab characterized DAH itself (2020, Nature Communications) and produced this trajectory paper (2025). Same lab, same enzyme, both papers.
one group, one lineageThe instrument hasn't looked yet. The data are deposited and public — genome assembly and annotation, QM/MM structural models, LC-MS raw data. Nobody outside has picked them up and attempted the LLM-guided design task.
the instrument hasn't lookedThe one tell this case passes cleanly. Modeling, MD, QM/MM, and mutagenesis all land on the same residue: Lys205 — the K205A mutant abolishes halogenation outright, and GBSA energy decomposition gives it −9.7 kcal/mol (the nearby Thr231 and Asn262 were tested too — T231A and N262A changed nothing, so Lys205 carries it alone). And the closing claim is genuinely falsifiable: either someone builds a working novel halogenase using this dataset as guidance, or they don't.
passes — mechanism argued, bet falsifiableLink out — Paper: DOI 10.1126/sciadv.adv6898. Data: Zenodo 10.5281/zenodo.15802776 (large — genome ~1 GB, structural models ~5 GB; not a phone download) · NCBI BioProject PRJNA1285983.
The measurement half is clean, and it is calibrated. Since 2012, brown bullhead in Lake Memphremagog — which straddles Vermont and Quebec — have carried raised black melanistic lesions, found in 23–37% of fish sampled 2014–2017. The working assumption was contamination: tropical storm Irene flooded the watershed in 2011, and runoff pollution is the kind of thing that causes cancer clusters in an indicator species. The genomics wouldn't cohere with it — tumour mitochondria didn't match their own host fish — so the team chased the contradiction instead of forcing the pollution story, and landed somewhere else entirely: the tumours are a single clonal lineage, and the cancer cells themselves are the infectious agent. Fourth documented transmissible cancer in animals, after dogs, Tasmanian devils, and the bivalves. First in any fish. First in freshwater.
What makes it a measurement rather than an impression is the baseline. Conventional cancers share some mutations by chance, mutational bias and selection — so “these tumours look similar” is worthless without knowing how similar independent tumours normally look. They went to TCGA and got the number: across 463 human melanomas, 94% of SNVs are unique to a single cancer, and 40 out of 498,648 — eight thousandths of a percent — are shared by ten or more. The bullhead numbers sit on the other side of the world from that: 245,189 tumour-specific SNV sites against 61,699 brain-specific; 59% of the tumour-specific sites shared by at least 14 of 16 fish, while 83.4% of the brain-specific sites appear in exactly one fish. Copy-number profiles correlate at r ≈ 0.83 between tumours in different fish, and r ≈ 0.22 between a tumour and its own host. (Every figure in this paragraph was checked against the paper and its supplementary tables before filing — house standard.)
It isn't, necessarily. Clonality tells you what the tumour is. It says nothing about why prevalence went from unremarkable to roughly 30% starting in 2012. If the lineage is old — and it may be very old — then nothing arrived in that lake; what changed is that it started taking. Immunosuppression from post-flood water quality would explain that completely, and the paper names the mechanism before setting it down and moving on. No water chemistry appears in the paper at all. The contaminant hypothesis was not falsified. It was displaced by a more interesting one.
No independent group has sequenced affected bullhead from this or any other population. The internal cross-checks are real — two variant callers (GATK HaplotypeCaller and Mutect2) agree on the sharing distribution, two phasing approaches return the same mitochondrial haplotypes — but that is one dataset checked twice, not convergence.
no 2nd measurement yetSingle team: University of Vermont plus Vermont Fish & Wildlife. Second author came out of the Metzger lab, which produced much of the bivalve transmissible-cancer literature cited here. Named reviewers — Ujvari, Hamede, Storfer — are devils and bivalves. Not a criticism of conduct; a structural fact: in a field this small, the people qualified to check the claim expect it to be true. In the first days after publication it showed: the independent experts who praised the paper (via the Science Media Centre) are drawn from that same small field — and one of them, Ujvari, is a named reviewer here. The one caveat any of them raised — that the route of transmission is still unclarified — is Tell 3, below.
one group, one lineageThe tell that makes the exhibit worth filing. An unrelated instrument did look — once. In 1925 Osburn described black tumours in Massachusetts bullheads and induced tumours in naive fish by contact with tumour material. No genomics, completely different method, same answer. Then a century passed and nobody ran it again. The experiment that converts inference to demonstration is a fish in a tank, and this paper does not contain it.
the instrument looked in 1925, not sincePasses. The clonality claim falls out of a quantified comparison against an external baseline (TCGA) — the strongest form this tell takes. And it leaves falsifiable predictions on record: run the 1925 challenge experiment; produce a molecular-clock age for the lineage; apply eDNA surveillance to other water bodies — a method in the paper's own reference 35 (Weinandt et al., PNAS 2026) and not applied here.
passes — mechanism argued, bets falsifiableRead alongside the Supplementary Information and Supplementary Tables 1–11. Two of these cut against the obvious objections; one does not.
The claim actually left exposed is not clonality. It is the assertion that the lineage originated outside Lake Memphremagog — which rests on tumours grouping nearer to reference fish sequenced at 74× while lake fish sat at 11–18×, with n = 2 from New Hampshire and n = 1 from Maine. That is the sentence to make them defend.
This exhibit resolves as two verdicts, not one — the same structure as Exhibit III.
Measured, externally calibrated against TCGA, and confirmed rather than weakened by the supplementary data. Independent sequencing of a second affected population graduates this to the Sledgehammer Wing.
Claim B — if a molecular clock predates 2012Open. A clock showing the lineage substantially predates the outbreak would send this toward the Caliper Room — not because the measurement was wrong, but because it answered a different question than the lake was asking. Claim A would be untouched.
Lake Memphremagog is the drinking-water source for more than 175,000 people. That fact appears once in the paper, in the second paragraph, as justification for why the melanomas were being studied at all. It does not come back. The headline — contagious cancer, not pollution — functionally retires a public-health question using evidence that never addressed it. Nobody wrote that the water is fine. But that is what the label now implies to every reader who doesn't get past the abstract, and labels outlive the findings that produced them. That is the whole reason this room exists, and it is why this case is held open rather than filed.
Link out — Paper: DOI 10.1038/s41586-026-10828-6 (open access, CC BY-NC-ND 4.0). Peer review file on the article page (Ujvari, co-reviewed with Hamede; Storfer; one anonymous). Data: NCBI BioProjects PRJNA1284832 and PRJNA1284812; genomes PV874414.1 (mitochondrial) and GCA_051624135.1 (nuclear). Code: github.com/limey-bean/Bullhead_2025. Prior art: Osburn, “Black tumor of the catfish,” Bull. Bur. Fish. 41, 9–13 (1925).
Independent confirmation lands. The swing connected; the floor really shifted.
If a sharper instrument dissolves itThe anomaly turns out to be a measurement problem. Ruler, not revolution.
The inverse roomWhere the settled cases rest — the ones the Docket already resolved.