The Demo Economy
The West keeps score in demo views and funding rounds. The buyer’s paper trail keeps score in inspections passed, money paid and repeat orders, and that side of the scoreboard is nearly blank.
Capital is being allocated against a record of announcements, while the record of transactions says something different and is public in one jurisdiction.
Published 20 pages
Abstract
Coverage of frontier robotics runs on announcements: demos, funding rounds and pilot press releases, none of which carries a delivery obligation. This report follows the other trail, the one a buyer leaves behind, in tenders, transacted prices, signed acceptances, payments and repeat orders. The announcement side is thinner than it looks: the most-cited Western deployment’s headline metrics appear twelve times in the reviewed record and trace to one origin, the vendor’s own pilot dataset, and eleven of twelve flagship claim lineages collapse the same way. The cure the report argues for is method rather than cynicism, which is why the model here hands the reader the trust allocation and keeps the evidence fixed.
Key findings
- 292 → 1 → 0 → 0Disclosed 2025 Chinese humanoid award projects, then one documented acceptance chain, then zero verified payments, then zero verified repeat purchases. The zeros are ceilings on what can be known, not floors on what happened.
- 12 : 1Appearances to independent origins for the most-cited Western deployment’s headline metrics: twelve citations, one source, the vendor’s own pilot dataset.
- 0Verified post-acceptance humanoid reorders anywhere in the reviewed public record, against company-wide revenue growth of 335 per cent at the record’s largest filer.
- 11 of 12Flagship claim lineages that collapse to a single origin under counting. No flagship programme publishes its autonomy share.
The report rests on 100 separately sourced claims resolving to 70 numbered references, every link verified. Each section header states how many claims it stands on, and every figure carries the claim ids behind the values it prints.
PDF, 20 pages, 3.2 MB. No form, and no mailing list to join, because we do not run one.
Argue with it.
Decide how much each class of evidence is worth to you, from announcements to signed acceptances and payments, and watch what the record can then support.
Everything loads at the report’s own case, and one click puts it back. Move anything beyond the published evidence and the affected numbers say so. The model is the one the report was written against, not a demonstration built beside it.
Open the world modelThis report is a very limited showcase of Quorum.
About one word in thirty-six of the finished chapters, and about one in seven thousand of the corpus. This report runs to roughly 7,000 words. It summarises one part of a study whose written chapters run to about 256,000 words, over a working corpus of approximately 50.1 million words in 29,111 files. That study asked 263 atomic research questions, collected 3,027 separate model reports against them, and ran 4,250 audits of those reports before a single chapter was written.
Every factual statement in it carries a reference to a verified source, and every figure carries the claim ids behind each value it prints. A commissioned engagement is not a longer version of this document. It is the research underneath it, run on your question, with your sign-off on the brief and an audit trail you can hand to a board.
- Adjudication and de-biasing
- All ten themes were pre-registered as a hypothesis and its explicit negation, and adjudicated evidence decided which side survived. Eight models framed the themes and seven adversarial models attacked them before collection. Every finding carries a grade, and a component inherits the worst grade it rests on.