量化台用 Opus 5.5 生成策略、Jev 做快速筛选
一个量化交易台用 Opus 5.5 批量生成策略配置，再由 Jev 跑快速校验，87% 的策略被 5 条淘汰规则筛掉，只有 13% 存活。
This f*cking quant desk KILLS 87% of its own strategies. Watch where the 13% that survive go. Opus 5.5 breeds the configs, Jev runs the fast checks. Every strategy has to pass 5 kill rules - steal them for your own backtests: 1. Name who pays you. Funding longs, forced sellers, dealers hedging gamma, the slower venue. No payer, no trade. 2. Trade the plateau, never the spike. If the edge dies when you nudge a setting by 10%, it was luck. 3. Count your tries. Test enough ideas and one will look brilliant by accident: SR_max ≈ σ · √(2 ln N) N = how many versions you tested, σ = how much Sharpe wobbles by pure chance (about 0.5 with 4 years of data). 200 tries gives a "best" Sharpe near 1.6 from noise alone. 4. Check markouts at 1s, 10s, 60s. If price keeps running against your fills, someone is trading on you. 5. Survive every regime - trend, chop, crash, squeeze. One blowup and it stays on paper. Survivors get sized small and fire into the ladder. Simulated, paper only for now. Bookmark this before your next backtest.
这条是引用着另一条写的：x.com/maxdeg0/status/2110042746622750787