Quant Strategy Backtester

Problem to solve

The hard part of quant backtesting is not only running a strategy, but separating selection, rebalancing, and risk logic into auditable stages. wm-backtest treats this as two subsystems: an execution library for historical simulation and a management layer for creating, previewing, publishing, and reading custom units. It is not a paper-trading app or a live intraday loop; it targets single-market A-share historical backtesting.

How it works

The main workflow is custom unit + staged scripts + stage testing:
- Selector: filter→bool or list→[]symbol controls the candidate universe and ranking.
- Trading / risk: emits OrderProposal; the third element is a weight ratio, not cash. Rebalancing belongs in trading; selector_freq only refreshes selection.
- Verification: after changing a stage script, run the stage harness, then lint, preview_custom, and run_custom.
- Evidence: instrument runs with simulation.trace, factor, and risk_*, then pull paginated deep / trace logs and inspect metrics.rule_counts to confirm rule triggers.

Boundaries and cautions

Use it for historical validation of a stock pool plus holding rules, strategy iteration, and result review such as equity, fills, and status. Avoid using it for live intraday trading, paper trading, screening-only workflows, or watchlist-only cleanup; it is not a substitute for Python or third-party backtest platforms. Business parameters must go under args, unconfirmed runs should not set confirm=true, and an existing unit_id will return UNIT_EXISTS.