RavenGraphWatch a shock propagate.
A shock enters the graph at its origin and cascades through calibrated edges — weights and lags learned from history. Pick a scenario, set the severity, scrub the timeline.
A +8.0% crude shock prices a -2.7% move into DAL within ~26 minutes — down through the airline cluster, before the tape reflects it.
Everyone trades the same isolated time series.
Price, volume, a handful of factors — one ticker at a time. The relationships between assets — who leads, who follows, how a shock spreads — never enter the model. That structure is where the edge is.
Each ticker modeled alone — moving on its own clock.
One move, propagating through the structure.
One graph. Agents research over it. Models trade on it.
Today’s AI trades by reasoning over price series and headlines. It can read that oil spiked — then it infers the consequences from scratch, every time. The graph makes propagation explicit: who leads, who follows, at what lag. It is the world model the whole system reasons over, not something re-derived from text on every call.
“A fund that scales like a software company.”Agents run the research and every strategy reads the same graph — AUM grows without growing the team.
Early, live, and honest about it.
Up ~38% year-to-date, roughly 26 points ahead of the S&P 500. Live trading, on a small own book.
Live on Hyperliquid in a pilot with Avant Protocol, our DeFi design partner. Micro-size, with performance shared weekly.
Still in build — the core bet, and the hard part. Today’s results come from the baseline model.