Earnings Trading Platform

An independent research platform for trading around company earnings prints. A radar tracks upcoming prints across a watchlist; each ticker gets a prep board assembled before the event — consensus EPS and revenue estimates, the options-implied expected move, past print reactions, and an analyst picture — so every decision starts from the same structured context.

The simulator re-runs any historical print through the full pipeline — filing detected, parsed, features extracted, signals scored, outcome reviewed. Engines compete champion vs challenger: a new configuration (including LLM-driven variants) must beat the incumbent across graded replays before it earns promotion, and calibration gates decide which engines are eligible to act at all. Backtests run walk-forward over the same universe, with per-trade replay results to inspect.

Like its sibling research platform, the system measures before it trades: shadow modes, paper-only runs, and a reviewable trail for every decision. The research is still in progress — this page is an early preview, and the full case study lands when the methodology closes. Code available on request.

This system does not trade real money yet — on purpose. Before a single live order, the edge has to survive validation end-to-end: replayed prints, walk-forward backtests, and champion/challenger runs that hold up out-of-sample, net of fees, slippage, and the adverse selection that quietly kills paper edges. Until that bar is cleared, everything runs in shadow and paper modes. The project is in active development; if the edge proves real, it goes live small and measured — and if it does not, the postmortem will say so honestly. Either outcome is a result.