proposal.md 1.5 KB
Why
The initial ultra-short strategy search used too few sampled windows to justify a trading conclusion. The project needs a repeatable validation path that rejects small-sample winners and only promotes strategies whose long-range window statistics remain positive.
What Changes
- Add a robust ultra-short validation capability that evaluates candidate strategies across long historical ranges and all non-overlapping windows.
- Record statistical outputs required for candidate acceptance: sample count, average return, 95% confidence interval, median return, positive-window rate, tail returns, trade count, win rate, and max drawdown.
- Promote RSI2 as the current candidate family only when its long-sample confidence interval lower bound remains positive.
- Reject the earlier ETH 3m VWAP candidate because long-sample validation does not show a positive confidence interval.
- Keep validation as a research/backtest workflow; do not change live or paper trading behavior in this change.
Capabilities
New Capabilities
robust-strategy-validation: Defines the long-sample validation contract for ultra-short candidate strategies and the acceptance criteria for promoting a strategy candidate.
Modified Capabilities
None.
Impact
- Adds OpenSpec artifacts for strategy validation requirements.
- Affects research/backtest scripts and generated CSV outputs.
- Does not change OKX order placement, paper positions, authenticated APIs, or the existing CLI trading commands.