How it works
Four stages between an idea and a fill.
Each stage is instrumented; each output is reviewable. Quantitative signals at the core, a news-based decision layer on top, the firm's own treasury at the end of every fill.
Research
We start with a hypothesis about a market behavior — a microstructure quirk, a regime relationship, a flow imbalance, a recurring news pattern. Nothing leaves the bench until it has a thesis we can defend in plain language.
- Instrument selection by liquidity, fee structure, and execution venues.
- Data audit — gaps, survivorship, corporate actions, splits.
- Statistical priors and news-feature design before any code is written.
Modeling
Strategies are specified as code, backtested across regimes, and stress-tested against the failure modes their authors most want to forget. Quantitative signals are fused with a news-based decision layer; reinforcement learning is used where it earns its keep.
- Walk-forward backtests with realistic costs and slippage.
- Regime tagging — bull, bear, vol blow-out, low-liquidity, weekend.
- News-aware features audited for leakage and timing realism.
- A live paper-trading bake-off before any real treasury touches the strategy.
Execution
Live trading runs on a co-located, deterministic stack. Order routing, kill-switches, and reconcile-required halts are part of the system itself — not bolted on.
- Microsecond-grade order placement on supported venues.
- Hard caps on per-trade and daily exposure (the Plimsoll line).
- Audit log of every decision and every fill, retained immutably.
Reporting
Daily reconciliation against the firm's treasury. Plain-language commentary on what worked, what didn't, and what we adjusted. No curated distance.
- Position-level P&L, exposures, and attribution.
- Drift alerts when a strategy's behavior changes meaningfully.
- A monthly long-form letter on regime, posture, and changes in view.
Want to go deeper?
Existing investors find strategy specs, performance, and audit in the portal. Otherwise, get in touch.