Hedge Backtesting

Turns mechanical strategy rules into a real, executed backtest. No number in this skill's output may be estimated, guessed, or recalled from memory or training data — every metric must be traceable to the stdout of a script that actually ran against actually-downloaded data. If the data can't be pulled or the script can't run, the honest output is INSUFFICIENT, never a plausible-looking guess.

This mirrors the project's harder-edged law: strategy-discovery-backtest is the final gate before capital moves; this skill is the fast, iterative loop that gets a strategy into good enough shape to be worth sending there. This skill's author-agent must never also be the strategy's grader — if the strategy came out of hedge-strategy-generation in the same session, that's fine (generation ≠ grading), but the PASS/FAIL verdict on whether to trust the edge is strategy-discovery-backtest's call, not this skill's.

Input

Strategy rules — ideally the structured output of hedge-strategy-generation (indicators, entry/exit, stops/targets, sizing), or pasted rules from the user. If any rule is not mechanically codeable (requires discretion), stop and ask for a codeable rewrite before writing any script — do not silently interpret an ambiguous rule your own way, that's a hidden assumption the backtest result can't be trusted against.