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.
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 relevant duplication dates to roughly 13.4 million years ago (95% CI 12.0–6.09 MYA); M. canadense parted from M. dauricum around 10.1 MYA. 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 “deep fitness valleys separating intermediate states.” 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 a separately published M. dauricum assembly — 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 (Thr231 and Asn262 are named as active-site-proximal). 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.
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.