The same case, written for agents
The proposal asks for records a machine can check. So the site is published that way too.
/llms.txt
A short index for agents: what the site proposes, where the full text and data live.
/reproducible-cmp.md
The whole case as markdown, in reading order. Each paragraph ends with the ids of the claims it rests on; the claims are listed at the end with sources and labels.
/data/workflow.json
The run log, the next step after the assistant: its five steps and the 10 systems underneath them, as structured data: what each step does and proves, what AI does, who can see what, and which pieces exist today.
/data/cmp.json
Everything on the site: 6 parts, 5 problems, 8 decision makers, 59 claims. Stable ids, a source URL, a confidence and a checked flag on every claim.
How to read a claim
{
"id": "c-report-prb",
"text": "The Pittsburgh report was published in Physical Review B (113, 119601) on 20 March 2026 and gives recommendations for researchers, publishers, universities and research laboratories, and government and funding organizations, drawn from a conference at the University of Pittsburgh on 9 to 11 May 2024.",
"source": "Akrap et al. 2026, Physical Review B 113, 119601",
"url": "https://doi.org/10.1103/27h6-yghn",
"checked": "ai",
"confidence": "confident"
}checked is human when David has worked through the claim, else ai. Right now 0 of 59 claims are human-checked. A scope field says when a number comes from one field or one study. A null url means the claim rests on David's notes and is labeled as such.