Vigil

Make the weak ideas visible

Finding a pattern in historical prices is easy when many patterns are tested. Vigil keeps the path from a candidate setup to a decision inspectable: what was proposed, what was tested, and why it was rejected or promoted. An empty result is a useful outcome when the checks do not support a hypothesis.

Test before promotion

The research engine scans candidate setups, records a hypothesis for each survivor, and freezes its parameters before validation. It records in-sample and out-of-sample results, walk-forward folds, a Monte Carlo null test, cost stress, and additional stress replay. The point is to expose fragile ideas before they become strategy specifications, not to claim a profitable result.

Proposed strategies are declarative rules checked against an indicator whitelist. The model may suggest an idea, but generated code does not run in the execution path. Each promotion retains its validation record and explicit criteria for later review or demotion.

Paper-first operation

Vigil has a native Mac app and a browser dashboard for reviewing each hypothesis, its validation record, and the current risk state. Paper trading is the default; live mode requires separate configuration and operator interlocks.