Expand description
sb-sim — point-in-time market simulator + reference agents (Phase 1).
The engine feeds an Agent a sharpebench_protocol::MarketObservation that only
ever contains data at or before the decision date (look-ahead is structurally
impossible — Dataset never hands out a future bar), applies the resulting
orders with transaction costs and seeded execution slippage, and emits an
sharpebench_core::Run (per-period returns + decision trace) ready for scoring.
Re-exports§
pub use agent::Agent;pub use agent::BuyAndHold;pub use agent::HoldAgent;pub use agent::Momentum;pub use agent::RandomAgent;pub use agent::TeamAgent;pub use costs::CostModel;pub use costs::CostProfile;pub use data::Dataset;pub use engine::run_backtest;pub use engine::Window;pub use env::EnvState;pub use env::Scenario;pub use env::StepInfo;pub use env::StepResult;pub use env::TradingEnv;pub use external::ExternalAgent;pub use external::HttpAgent;pub use trajectory::replay_run;pub use trajectory::replay_submission;pub use trajectory::run_backtest_capture;pub use windows::tag_regime;pub use windows::walk_forward;pub use windows::Regime;
Modules§
- agent
- In-process trading agents.
- costs
- Transaction costs + a tiny seeded PRNG for execution noise.
- data
- Point-in-time price data.
- engine
- The point-in-time backtest engine.
- env
- The Gym-style open-loop environment: the caller drives
reset()/step(). - external
- External-process agent — speak the JSON protocol to an any-language agent.
- trajectory
- Trajectory capture + replay-recompute — the Tier 1 verification boundary.
- windows
- Window generation: walk-forward OOS splits + coarse regime tagging.