Approach
A model earns capital only by surviving every stage of a deliberate process. Most ideas are designed to fail one of them — and that is exactly what makes the survivors worth trusting.
State the idea precisely enough to be wrong.
Each research thread starts with an economic hypothesis and a mechanism: a reason an inefficiency should exist and persist. Framing comes before searching, which keeps us from mistaking coincidence for insight.
Make the past honest.
We assemble point-in-time datasets that reflect only what was knowable at each moment — corrected for survivorship, look-ahead and vendor revisions. The quality of every downstream conclusion is bounded here.
Try, deliberately, to break it.
Candidate models run a gauntlet: out-of-sample windows, cross-regime splits, transaction-cost and capacity modelling, and sensitivity to assumptions. We assume an effect is spurious until it refuses to disappear.
Size by risk, not by conviction.
Survivors are combined with explicit attention to correlation, drawdown, liquidity and capacity. No individual model is permitted to dominate the portfolio's outcome. Risk is designed in, not bolted on.
An immune system for alpha.
Live behaviour is continuously compared to its research baseline. When a model's evidence weakens — as markets adapt and edges decay — it is scaled down or retired automatically, before conviction can override data.
The standard
We would rather deploy nothing than deploy something we do not understand. Discipline is not a constraint on the research — it is the research.