Applied to contested science.

1 COVID-19 origins · navigate to the evidence

Five readings. One picture. See for yourself.

The Rootclaim COVID-origins debate turned on one term — where the earliest known cases clustered — and five independent-looking analyses weigh in. They are five readings of one shared picture, not five measurements of the world; because each is indexed to that same artifact by its observationBase, the schema lets the readings coexist over one base — and lets you walk down to it and judge for yourself. Do that, and the second surprise lands: the picture stands in for a dataset that was never released.

Walk down to that shared base, and here is what you find. Judge Eric Stansifer, building his own case inventory for the debate, wrote down the problem himself:

The 164 December-2019 Wuhan cases were never published as a data table. They exist publicly only as dots on a map in the WHO annex. To get coordinates at all, Worobey et al. had to digitize the picture — recovering 155 of the 164 points, and stating plainly that No line list of early COVID-19 cases is available, with the extraction introducing up to about 50 m of noise. Every later analysis — a spatial-statistics rebuttal, both judges, the losing side — reads from that same digitized picture.

The derivation, made visible

WHO Fig. 23 — the only public depiction of the 164 cases. Schematized here as a stylized dot-scatter; the original scan is not reproduced.

→

Worobey Fig. 1A — a digitized re-rendering of 155 of the 164 points. Adapted from Worobey et al., Science 377:951 (2022), CC BY 4.0.

So the anchor of the entire debate is one non-reproducible picture. The five analyses were never five independent measurements of the world; they are five models over one thin input. A spread that large across shared inputs is a signal about the analyses, not about the world — and you can only see it once the evidence is indexed by the artifact it stands on, not by the paper that cites it.

It accumulates — the graph accepts work that didn’t exist when it was built

The Rootclaim debate closed in 2024. In March 2026, Cell published a new class of evidence — selection dynamics on the branch preceding emergence, with the 1977 H1N1 outbreak (an accepted lab/vaccine-origin event) as a validated positive control. In a sealed-document world that paper is a new PDF nobody wires up. In the graph it is a Claim and two Evidence nodes that snap onto the existing structure.

The COVID Request writes itself

Both judges recorded what would have changed their minds — and had nowhere to put it. Judge Will Van Treuren, on the debate format:

He even named his own unexamined crux: A cursory search suggested some (more credible) reports of early seropositivity in retrospective studies of US blood banks that would have been hard to explain under the ZO theory. That sentence has sat inert in a PDF ever since. As a record it becomes a Request: request_target → the contested market-centrality Claim; request_for → an adversarial, pre-registered re-analysis of the same 155 coordinates. It needs no new data and no one’s cooperation — it hangs off the exact claim above.

This isn’t only a proposal. In one lab’s own issue tracker — Roam, with the Discourse Graphs plugin — the mechanism already runs. Over 40 months: 445 issues; 29% claimed; 15% claimed by someone other than the creator; 38% of claimed issues produced a result; median 12 days from claim to first result. One issue sat dormant for fourteen months until a new undergraduate found it, claimed it, and completed it before the PI knew it existed. The vocabulary maps without translation: Issue = Request, Experiment = Study, Result = Evidence.

Scope, honestly: that is one lab, in Roam, on the DG plugin. The handoff mechanism is validated; MIRA itself is not “deployed at scale.” The field data is drawn from Matt Akamatsu’s published lab updates.