Submission materials · Track 3 · AI Trading Desk · Open Theme
Every link here is live, and the figures on this page are read from the same artefacts the pages use, at the moment you load it.
Ask about your Bitget book in plain words; every figure computed from live data and named by source, every line labelled by where it came from.
The Track 3 demo the handbook asks for: one question about adding a name to a book, eight engines, and an actionable insight. Change ?name=, ?size= and ?book= to run your own.
The research task, recorded end to end on the live console.
51 capabilities, each graded against a named rival and OWNED only where that rival was run on the same input and beaten under all thirteen conditions: 6 OWNED, 37 TIED, 0 IMPLEMENTED, 8 LOST — read from the register now.
37 findings published, 21 of them comparisons a rival won.
Bitget's toolkit and every data source, checked live when the page loads, with the desk's record and its age.
Public, MIT. python -m argus.eval.standing re-derives the register from its artefacts; python -m argus.eval.docclaims checks every figure the documents quote.
How ARGUS works, every problem it takes on and how, the architecture, and the measurement against named rivals with every loss.
The evidence in slides: the research task, the register, the wins and the losses, the toolkit sweep.
The same console for an agent: JSON-RPC 2.0 over POST, ten tools, from argus_ask to argus_review_trades.
The same console in a chat; /book saves holdings.
Every page's data without the page: add ?format=json to /proof, /wrong, /status or /research.
The identity, drawn from the tokens every page uses.