One arbiter couples six engines.
One model misses the crowd that moves the price. Six engines each do one job, and the arbiter decides which ones answer.
ArbiterIt answers on the fast path and calls the slow engines only when their answer could change the trade, cross-checking every claim.
Answers fast, slows down when it matters.
- distilled calls and live moods answer fast
- < 1 s
- sealed research models and crowds answer slow
- Minutes
- signals call the slow path
- 4
Every decision opens to its reasons.
- Decision
- Trim 40% over five days
- Path
- Slow: the crowds disagreed
- Flips if
- forced sellers pass 6% of float
- Quarantined
- 18% of simulated profit
- Crowd disagreement41%
- Other investors' fear33%
- Research26%
10parts in every decision record
We keep every crowd the prices allow.
Tap to add a price, and crowds that miss it drop out.
Profit the simulator invents never counts.
A result counts only on later real prices after trading costs.
- simulated participants
- 15,000
- decisions per run
- 5.85B
Six crowds agree on four prices, then split.
Covers four of five trade styles.
The first book spans single-name equity, macro, arbitrage and derivatives, sitting out high-frequency trading.
Estimated trade coverage by firm across five dimensions: single-name equity (EQ), macro (MAC), arbitrage (RV), high-frequency (HFT) and derivatives (DER). Core means a live book, partial means claimed or inferred, none means no public evidence, unknown means undisclosed. From public data, unverified.
- Bridgewater AIA Labs: EQ partial, MAC core, RV partial, HFT none, DER partial
- Abundance: EQ core, MAC partial, RV none, HFT none, DER none
- Poesis: EQ partial, MAC partial, RV partial, HFT none, DER partial
- ViserAI: EQ unknown, MAC unknown, RV unknown, HFT unknown, DER unknown
- Numerai: EQ core, MAC none, RV none, HFT none, DER none
- Fiscal.ai: EQ none, MAC none, RV none, HFT none, DER none
- Incumbent quants*: EQ core, MAC core, RV core, HFT core, DER core
- Cotangent: EQ core, MAC core, RV core, HFT none, DER core
From public data, unverified.
Three filings cover one instrument.
Patent pending under three U.S. provisional applications (US 64/171,895, 64/171,899, 64/171,901).
The system spends fine detail only where plausible markets disagree about the trade.
One policy holds across every crowd the data allows.
Simulated profit counts only where real data supports it.