Crate yuzu_core 

Pure, I/O-free backtest engine core for US equity strategies.

Pipeline

1. Build panel::Panels of market / factor data (dates × symbols, f64).
2. Put them in an EvalContext (optionally with a sector map).
3. Evaluate a lemon JSON spec::Expr with run_strategy → position panel.
4. Or run end-to-end with run_backtest → report::Report (equity, trades, metrics).

No network and no platform deps: matrices in, report out. Compiles to native and WASM. Conceptual overview: docs/backtest-engine.md. Strategy language: docs/lemon.md.

Example

See the crate example basic_backtest:

cargo run -p yuzu-core --example basic_backtest

Re-exports

pub use error::EngineError;

pub use eval::run_backtest;

pub use eval::run_strategy;

pub use eval::EvalContext;

Modules

align

align reshapes two panels onto a shared index before an elementwise op: rows = union of dates (sorted), columns = intersection of symbols (left’s order). Absent cells become NaN (standard reshape/align semantics).

backtest

Daily-equity NAV loop: turns a position-weight matrix + price panel into an equity curve and a trade list. See docs/backtest-engine.md for the model.

bootstrap

Circular block bootstrap over daily returns: resample the return series in blocks (preserving short-range autocorrelation), rebuild synthetic equity curves, and report percentile confidence intervals for the headline metrics. Deterministic — a built-in SplitMix64 PRNG with a fixed seed, no external dependencies (WASM-safe).

error

EngineError: shape violations from panel construction and evaluation failures.

eval

Evaluates an Expr tree against a data context (HashMap<String, Panel>).

metrics

Standard performance metrics (CAGR, Sharpe, Sortino, max drawdown, etc.). All functions operate on a daily equity curve (&[f64], base 1.0) or a daily returns slice. Conventions: annualization 252, rf = 0, std ddof = 1.

ops

DSL operations, grouped by family. Each submodule adds impl Panel methods.

panel

The Panel type: a dense f64 matrix indexed by dates (rows, YYYYMMDD) and symbols (columns). NaN = missing; booleans are 1.0/0.0 with NaN falsy (see is_true).

report

Serializable backtest report — the JSON contract the web app renders (equity/drawdown series + trade list + metrics). Engine computes; frontend draws.

research

Factor diagnostics and event studies — research workflows that take panels and return JSON, deliberately kept out of the lemon AST (see issue #45). These are library entry points, not strategy ops: they don’t produce positions or change backtest NAV semantics.

spec

Expr: the serializable strategy AST. A JSON tree (tagged by "op") that the website and the batch runner both emit; the engine’s evaluator walks it against a data context to produce a position matrix.