method

the lanes

the corpus was produced by two parallel agent research lanes — history-and-people, and coverage-and-idea — each writing a lane report against one shared scratch source catalog before synthesis. the pattern is the company-research formula proven on roam-research: lane-prefixed source ids frozen upfront, per-claim confidence, and a declared deduplication owner.

the evidence rules

1. every event, quote, and claim carries at least one source id.
2. dates keep their precision — a year, a month, or a day — and approximate dates are labeled rather than sharpened.
3. confidence stays honest: confirmed, reported, and inferred are distinct claims, not stages of the same claim.
4. sources are typed by role — primary, interview, reporting, archive, community — so a claim's foundation is visible before it is read.
5. sentiment labels describe the coverage, not the truth of it.

known gaps

1. total funding is unreconciled — pitchbook logs a sep-2025 round, fröberg's linkedin claims $2.2m [self-reported], and tracxn lists the company as unfunded. the discrepancy is preserved, not resolved.
2. headcount is imprecise — directories read 11 to 18, and there is no public team page.
3. customer status and commercial terms with anthropic, openai, google deepmind, and xai are undocumented — partner or paying customer is not public record.
4. the mechanics of early-access to frontier models — what andon gets, when, and under what terms — are undisclosed.
5. public trace and methodology availability is limited; most audits stay internal, and the single-run human baseline ($844.05) is asterisked.
6. no benchmark paper has peer review; everything is an arxiv preprint or an andonlabs.com pdf.
7. pion's traction, pricing, and revenue-share terms are unreported beyond the launch post and trade coverage.
8. andon market's finances are self-reported — the ~$40k loss figure comes from andon itself via press interviews.
9. no bbc coverage was located; the corpus skews to english-language us and sf reporting.
10. no standalone “sleep-deprivation study” exists — the likely referent is the bengt experiment, which removed the agent's ability to sleep, plus sleep_until_tomorrow behavior in vending-bench arena.

corrections

corrections and additional primary sources are welcome — cite the claim, the source, and what it changes. the catalog is designed to be audited, not just read.