Papers and working notes on quantitative research, investment agents and the infrastructure that keeps them working over long horizons.
When can a trained system meaningfully be called the same model? Behavioral equivalence is underdetermined by finite data, with consequences for reproducibility, traceability and governance.
Read the paper →From Markowitz on, portfolio management optimizes expected return penalized for risk, one step forward. This paper separates optimality from survivability and studies the difference.
Read the paper →A separate Guardian Agent reviews the Research Agent's plan before work begins and checks its evidence before the answer reaches the reader.
Natural language clarifies intent. A workflow DSL records the exact constraints, provenance, and evidence the research must carry.
Two unedited, public sessions: a stress-tested tactical strategy and a fossil-free index tracker, with the reasoning, checks and repairs left visible.
We asked it to find a tradable signal. After 29 rigorous steps it said there wasn't one. That refusal is the point.
Fundamental analysis and quantitative optimization each answer half the question. Here is how the agent runs both at once.
The process and mathematics behind Itoflow's in-house risk model for global stocks, ETFs, style tilts, mixed-asset covariance and single-stock shocks.
Asynchronous critique agents use spaced repetition to course-correct long-running investment strategies without stalling execution.
How we queue agent decisions and wait for each exchange to open, instead of executing against stale prices.
The patterns that made coding agents successful do not transfer cleanly to investing. The differences are structural.