Wickra GymO(1) steps. Deterministic rollouts.

A Gymnasium-compatible, microstructure-aware backtest environment: precompute the dataset to a fixed feature tensor once, and every step() becomes a pure array index.

View on GitHub

How it works

⚡

Every step is an index

The dataset is precomputed once into a fixed feature tensor. After that `step()` does no indicator work at all — it reads a row. Rollout speed stops depending on how many features you asked for.

🔬

Microstructure-aware

The observation is built from the same market structure the rest of the stack reads, not from a bare price series.

🎯

Deterministic rollouts

The same seed and the same spec replay the identical episode, so a training run is reproducible rather than approximately repeatable.

🧩

514 indicators as features

The feature tensor is drawn from the Wickra core's 514 streaming indicators, so a feature set is a list of names rather than code you maintain.