A Neolab for
Financial Superintelligence.
Global asset markets are the most demanding reasoning environment humanity runs — temporally sensitive, adversarial, reflexive, and scored in public every day. Finquill uses them as a crucible: forging systems for game-theoretic reasoning and resource allocation that hold up where reasoning is hardest, and publishing the graded record as we go.
Why markets.
Benchmarks get solved, gamed, and forgotten. Markets are not a benchmark — they are a complex adaptive system that prices your reasoning against every other participant's, on a deadline. Three properties make them the crucible:
Adversarial & reflexive.
Being right isn't enough — you must be right, non-consensus, and early, against participants who adapt to you.
Noisy, delayed feedback.
Good decisions lose; bad ones win. A system that learns here must grade reasoning independently of outcome — or it trains on noise.
Temporally unforgiving.
Every claim carries a deadline. There is no offline eval; the test set arrives every morning at 9:30.
A system that reasons well here — about odds, adversaries, and allocation under pressure — is the substrate of something much bigger than trading.
Falsification as an operating system.
The lab's discipline is not a whitepaper. Each mechanism below runs in production today.
Every belief is a dated, falsifiable claim.
Freeform conviction is compiled into structured claims with deadlines and kill criteria, then graded against what happened. Retrieval narrates; prediction forgets; we keep receipts.
Adversaries by design.
Theses face panels of refuters assigned opposing lenses, with independence manufactured across model families — unanimity is treated as an alarm, not a certification.
Two graphs, one mind.
A knowledge graph maps the world as it is. A reasoning graph records reads and plays — what we believed, what we did, and how the process graded.
No resulting.
A decision can grade well and lose; a coin-flip win can grade badly. Process is graded against what was knowable at the time, not against the outcome alone.
What runs today, and what's being forged.
Two instruments are live and gradeable right now. One is in development. The fourth is the direction the first three are pointed.
The Terminal
The same engine, hosted for your own book: capture your thesis in plain English, let adversarial agents red-team it, monitor its kill criteria against the world, and grade the record — calibration and process, not just P&L.
The Node
Democratizing decentralized AI — making distributed training runs, inference markets, and the broader DePIN economy legible and usable for anyone. Resource-allocation reasoning, pointed at the networks where compute, capital, and intelligence converge.
Where this goes
Near: graded research at machine scale — and the firm is preparing to trade its own research from a live account, in public, citing the same theses and terminal artifacts as everything else it publishes. Medium: reasoning systems any allocator can stress-test a decision against. Long: general resource-allocation reasoning that shows its record instead of asking for your trust.