rank229 · decomposition(baseline / components / ablation)

生成时间:2026-04-08 13:50:59 北京时间。本页服务于“把 P2/P3 策略拆到可执行组件”的需求。

← 返回该 rank 主报告P3/P2 总表

一、baseline / components / ablation 拆分

模块定义为什么需要
baseline事件驱动主baseline(时钟/季节性)提供主 alpha 假设与对照面。
incrementabnormal day veto(异常日过滤)该 rank 相比 baseline 的独立改动点。
roleveto否决层,在特定坏环境下阻止入场,关注误杀率与漏杀率平衡。
stage/statusP3 / keep决定当前是继续推进、观测,还是归档/降级。

二、信号定义(具体口径)

策略类型:风险过滤

  1. 输入层(Input):价格/成交/跨资产数据,要求明确频率与对齐时点。
  2. 候选触发层(Candidate Trigger):先给出“触发候选”的条件,不直接等同于下单。
  3. 过滤/否决层(Filter/Veto):说明哪些条件会阻止候选触发变成真实交易。
  4. 执行层(Execution Signal):给出最终下单定义(方向、仓位、并发约束、冷却时间)。
  5. 退出层(Exit):至少覆盖止损、止盈、超时退出与异常风控保护。
  6. 成本层(Cost):固定费率/滑点假设 + 敏感性结论(费率上升后是否仍成立)。

简化记忆法:候选信号 ≠ 真实执行信号;只有通过过滤与执行约束后,才是可归因交易。

三、ablation 建议

当前下一步:维持 live,按周复核 veto 命中与误杀比

四、现有证据摘录

证据摘要来源:reports/artifacts/paper_rank229_eth_abnormal_day/rank229_frozen_launch_spec.md

# Rank 229 Frozen Launch Spec

- Candidate: `Rank 229 / ETH-led abnormal-day continuation (session-defined)`
- Status: `paper_launch_queue_frozen_spec`
- Primary frozen spec:
  - Symbol: `ETHUSDT` perpetual
  - Venue: `Binance Futures`
  - Bar: `5m`
  - Session offset: `20h`
  - Trigger threshold: `k = 1.25 * sigma_session`
  - Minimum remaining bars: `12`
  - Entry: `next-bar open`
  - Exit: `session close`
  - Cost assumption: `12 bps round-trip`

## Why this is the launch spec
This is the thickest ETH ridge that still survived the honest `next-bar open` admission path and the later time/parameter segmentation:

- trades: `90`
- gross: `+98.69 bps`
- net-12: `+86.69 bps`
- halves: both positive
- thirds: `3/3` positive
- weakest third: `+6.65 bps`

## Conservative backup
- `offset 0h / k=1.75 / M>=12`
- net-12: `+44.22 bps`
- trades: `59`
- kept as backup because it is more even but thinner than the primary ridge.

## Scope boundary
This document freezes the queue-side paper spec so scheduler wiring can target one explicit runner. It does **not** claim that raw-bar live recomputation already exists.