公开研究
20,790,667 tries.
20,790,667 tries.
一个赢家。
We don't sell a magic bot — we sell the discipline of testing everything and keeping almost nothing. Here is every strategy configuration we evaluated between 2026-06-21 and 2026-09-12, and the single one that survived out-of-sample. Nothing here is live-trading performance.
漏斗
每一行都比上一行更难。来自我们策略实验室的真实数据——幸存者在每一步都在减少。
20,790,667
配置评估
每个参数组合都尝试过——网格扫描、优化器试验、退出机制运行
28,135
完整回测
真实K线、下一开盘价成交、费用+滑点、样本内/样本外分割
11,927
样本外为正
在从未调参的数据上盈利
961
通过门槛
样本外为正 + 夏普比率≥1 + 至少30笔交易
1
滚动窗口稳健
在滚动窗口前进验证+多重检验校正中幸存
幸存者
未入选者
公开而非埋没。这些看起来合理，但未能通过相同的样本外测试——这正是关键所在。
动量∧质量
+6.43%
vs +3.07% hold — refuted
ROE
-2.04%
vs +3.07% hold — refuted
如何阅读此表。 Every figure binds to our live strategy-lab aggregates (2026-06-21 → 2026-09-12). Backtests use real candles, next-open fills, fees + slippage, and an in-sample / out-of-sample split. "Passed the gate" is a basic filter; only the walk-forward-robust winner is one we stand behind — and even that is research, not investment advice. No strategy is cleared for live trading.