arbitrage-engine/docs/AB_TEST_CHECKLIST.md
root 7ebdb98643 feat: store and serve indicators per strategy
- signal_indicators table: added strategy column
- Each strategy gets its own row per cycle
- API /api/signals/latest?strategy=v51_baseline|v52_8signals
- API /api/signals/signal-history?strategy=...
- V5.1 page reads v51_baseline data, V5.2 reads v52_8signals
- Now V5.1 and V5.2 show truly independent scores
2026-03-02 02:52:17 +00:00

37 lines
1.2 KiB
Markdown
Raw Permalink Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

# AB测试观测清单2026-03-02 ~ 03-16
## 冻结期规则
- 不改权重、不改阈值、不改评分逻辑
- 如需改动必须打新版本号并分段统计
- 单写入源(小周生产环境)
## 两周后评审项目
### 1. 确认层重复计分审计
- **问题**方向层和确认层都用CVD_fast/CVD_mid同源重复
- **审计方法**:统计确认层=15 vs 确认层=0时的胜率差异
- **如果差异不显著**V5.3降权或重构为"CVD斜率加速+趋势强度"
### 2. 拥挤层 vs FR相关性
- **审计**`corr(FR_score, crowd_score)`
- **如果>0.7**:说明重复表达,降一层权重
### 3. OI持续性审计
- **字段**`oi_persist_n`(连续同向窗口数)— 目前未记录需V5.3加
- **审计**:高分单里`oi_persist_n=1`的胜率是否显著差于`>=2`
- **如果差异明显**:升为正式门槛
### 4. 清算触发率审计(按币种)
- 各币种清算信号触发率
- 触发后净R分布
- 避免某币种几乎不触发/过度触发
### 5. config_hash落库V5.3
- 每笔强制落库:`strategy`, `strategy_version`, `config_hash`, `engine_instance`
- 报表按config_hash分组
## 数据目标
- V5.1500+笔当前282
- V5.2200+笔当前12
- 每策略每币种50+笔