rank201 · decomposition（审计版）

生成时间：2026-04-11 12:36:14 北京时间。目标：把 baseline / increment / 决策口径拆到可复述和可执行。

← 返回该 rank 说明页 ｜ 研究复盘页（时间线） ｜ P3/P2 总表

⚠ 当前状态：观察线 / 非主研发线

1) 冻结版策略 Spec（固定，不外推）

项冻结定义
UniverseBTCUSDT, ETHUSDT, SOLUSDT, BNBUSDT, XRPUSDT, DOGEUSDT, ADAUSDT, LINKUSDT
Bar15m
Long UTC 时段20:00~21:59 UTC
Short UTC 时段22:00~23:59 UTC
Entry/Exit20:00 开多；22:00 平多并翻空；00:00 平空
Cost8 bps round-trip

2) baseline / increment / ablation 拆解

模块定义执行约束
baseline事件驱动主baseline（时钟/季节性）作为对照母体，不与其它改动混测。
incrementUTC 时钟季节性主线化只允许“UTC 时段切换”这一处增量。
ablation Abaseline only不加 increment，不加额外过滤。
ablation Bbaseline + increment仅新增 increment，验证增量贡献方向。
ablation Cbaseline + increment + cost stress固定 8 bps，并记录费率上调后的存活性。

3) 当前 verdict 与观察线解释

项当前口径
VerdictP3 / observe_only
5Y 结论冻结 schedule 在 5Y 重叠窗口整体为负，不满足“长期稳定规律”。
为何仍保留 runnerobserve_only 仅用于维持在线观测与监控链路；不表示通过长期有效性验证，也不代表扩投入结论。
决策边界在无新增验证前，不把该 schedule 描述为“可长期部署 alpha”。

4) 当前唯一下一步动作（SOP）

唯一动作：保持 runner live（因此仍归类 P3）；降级为观察线/非主研发线：不作为主赚钱候选，不分配主研究时间，不扩展参数与变体

  1. 确认 runner 与 timer 持续在线。
  2. 只监控：是否断更、是否异常跳变、是否告警触发。
  3. 如无异常，不改参数、不新增研究分支、不改 verdict。
  4. 默认不分配主研究时间，不再扩展参数与变体。
  5. 若决定下线 runner，则同步迁出 P3 运行态（转入非 live 队列）。

5) 现有证据摘录

证据摘录：reports/artifacts/rank201_5y_validation/README.md

# Rank 201 / 5Y validation (fixed UTC sleeves)

## Strategy frozen for validation
- Universe: `BTCUSDT, ETHUSDT, SOLUSDT, BNBUSDT, XRPUSDT, DOGEUSDT, ADAUSDT, LINKUSDT`
- Bar: `15m`
- Long sleeve: `20:00~21:59 UTC`
- Short sleeve: `22:00~23:59 UTC`
- Round-trip cost: `8 bps`
- Entry/exit rule matches the current paper runner logic:
  - `20:00` open long
  - `22:00` close long and flip short
  - `00:00` close short

## Data span used
Because the 8-symbol universe does not all start at the same timestamp, the validation uses the **common overlapping window**:
- Start: `2021-04-01 08:00:00 UTC`
- End: `2026-03-19 03:15:00 UTC`

## Headline result
This exact frozen Rank 201 schedule is **not** stable over the full 5-year overlap.

- Lifetime total return: `-91.61%`
- Max drawdown: `-93.67%`
- Positive months: `17 / 60` (`28.33%`)
- Mean monthly return: `-3.66%`
- Median monthly return: `-4.86%`
- Best month: `2021-05` = `+26.34%`
- Worst month: `2023-08` = `-21.67%`

## Yearly returns
- `2021`: `-3.92%`
- `2022`: `-25.14%`
- `2023`: `-67.23%`
- `2024`: `-28.85%`
- `2025`: `-49.37%`
- `2026 YTD`: `-1.20%`

## Monthly consistency by year
- `2021`: `3 / 9` positive months
- `2022`: `5 / 12` positive months
- `2023`: `0 / 12` positive months
- `2024`: `4 / 12` positive months
- `2025`: `3 / 12` positive months
- `2026 YTD`: `2 / 3` positive months

## By-symbol totals
All 8 symbols are negative over the 5-year frozen test.
Best to worst:
1. `XRPUSDT`: `-77.97%`
2. `SOLUSDT`: `-87.60%`
3. `ETHUSDT`: `-91.92%`
4. `ADAUSDT`: `-92.52%`
5. `BTCUSDT`: `-92.84%`
6. `DOGEUSDT`: `-94.67%`
7. `LINKUSDT`: `-94.88%`
8. `BNBUSDT`: `-96.36%`

## By-sleeve totals
Both sleeves lose money over the full 5-year overlap:
- `20_22_long`: `-99.88%`
- `22_00_short`: `-100.00%`

## Interpretation
This strongly suggests that the current Rank 201 paper candidate is **not** a long-horizon universal UTC clock law.

A more honest reading is:
- the broader **time-of-day / clock** family may be real,
- but the current frozen pocket (`20~21 long / 22~23 short`) looks like a **recent pocket**, not a 5-year always-on schedule.

That is consistent with the earlier desk conclusion that the paper runner wa