Run the current strategy on the reserved holdout data window and compare training performance against unseen data.

The holdout window is reserved automatically when the first backtest runs for a project.

Returns — raw in-sample and out-of-sample metrics, the per-strategy validation_count, holdout_trade_count, and a degradation_pct plus degradation_bucket (improved | low | medium | high). improved fires when holdout outperformed training (negative degradation). The caller decides what to do with the numbers.

Args — strategy_id to validate.

validate_on_holdout(strategy_id)

Errors if the strategy is not found, has no project or holdout window, or has no completed development backtests.