策略坟场 〔内部文件〕 说明 / README
用途 / Purpose
本表记录 ETH 团队测试过的每一个策略（包括被淘汰的、只做纸面对照的、只做执行检验的），每个策略一行，网站专页直接展示。目的是让大家知道哪些想法已经试过、为什么没用，避免重复踩坑。首批 143 行覆盖 2026-10-21 至 10-22 两天的全部研究报告。
This file lists every strategy the ETH team has tested, one row per strategy, including those eliminated, kept only as paper benchmarks, or run only as execution checks. The website shows it on a dedicated page. Its purpose is to record which ideas were already tried and why they failed, so no one repeats them. The first 143 rows cover all research reports from 2026-10-21 to 10-22.
文件格式 / File format
UTF-8 带 BOM（Excel 可直接打开），逗号分隔；字段内含逗号、引号或换行时用双引号括起，字段内的双引号写成两个双引号（标准 CSV）。
UTF-8 with BOM (opens directly in Excel), comma-separated. Fields containing commas, quotes or line breaks are wrapped in double quotes, and a quote inside a field is written as two quotes (standard CSV).
各列含义 / Columns
| 列 | 含义 |
|---|---|
| id | 唯一编号 GY-0001 起，按追加顺序递增，永不复用 |
| 记录日期 | 结果所在报告的日期（UTC+8） |
| 策略名称 | 策略名与账户/版本标识 |
| 规则 | 入场、离场、止损、仓位的简述（详情见报告） |
| 品种/周期 | 交易品种与K线周期 |
| 价格与数据口径 | 用的是现货还是永续价格、数据来源 |
| 样本区间 | 回测区间；注明是全样本、剔除某段还是样本外 |
| 扣费后收益(主口径) | 该报告主口径下扣除成本后的收益 |
| 收益口径 | 主口径的定义：满仓/名义10%/单笔风险0.5%/$100账户、全样本还是样本外、复利还是逐笔相加 |
| 最大回撤 | 同一口径下的最大回撤 |
| 交易笔数 | 已平仓笔数；不足30笔标「样本不足」 |
| 成本口径 | 该行用的成本假设 |
| 是否含资金费 | 永续是否计入资金费 |
| 淘汰原因 | 为什么淘汰；未淘汰的写明原因 |
| 状态 | 见下 |
| 报告文件路径 | 数字出处，相对 〔内部路径〕，多个用 ; 分隔 |
| 备注 | 其他口径数字、更正说明 |
- id: unique ID starting at GY-0001, incremented in the order rows are appended, never reused.
- 记录日期 (record date): date of the report the result comes from (UTC+8).
- 策略名称 (strategy name): name plus account or version label.
- 规则 (rules): short summary of entry, exit, stop and sizing; the report has the details.
- 品种/周期 (instrument/timeframe): the traded instrument and candle period.
- 价格与数据口径 (price and data basis): spot or perpetual prices, and the data source.
- 样本区间 (sample period): the backtest window, saying whether it is the full sample, a window with a period removed, or out-of-sample.
- 扣费后收益(主口径) (return after costs, main basis): return after costs under the report's main basis.
- 收益口径 (return basis): definition of the main basis: full size, 10% notional, 0.5% risk per trade or $100 account; full sample or out-of-sample; compounded or summed per trade.
- 最大回撤 (max drawdown): max drawdown on the same basis.
- 交易笔数 (trades): closed trades; fewer than 30 is marked 「样本不足」 (insufficient sample).
- 成本口径 (cost basis): the cost assumption used in that row.
- 是否含资金费 (funding included?): whether perpetual funding is counted.
- 淘汰原因 (reason): why it was eliminated, or why it was not.
- 状态 (status): see below.
- 报告文件路径 (report path): where the numbers come from, relative to 〔内部路径〕 separate several paths with
;. - 备注 (notes): numbers on other bases, and correction notes.
状态取值：
- 淘汰：不再研究、不上线。
- 纸面对照：只作模拟盘或纸面对照，不上线（如 B0、D1_DONCHIAN_ATR 55/27/2）。
- 执行检验：主管决定上线纸面账户，只检验执行、不设盈利预期（EMA2050 只做多/只做空）。
- 保留观察：报告建议保留或继续观察，但没有确认的上线决定（如方案_v1 中「留」的 4h 账户）。
Status values:
- 淘汰 = eliminated: no further research, not going live.
- 纸面对照 = paper benchmark only (e.g. B0, D1_DONCHIAN_ATR 55/27/2).
- 执行检验 = execution check: paper account approved by the lead only to test execution, with no profit expectation (EMA2050 long-only and short-only).
- 保留观察 = keep and watch: a report recommends keeping it, but no go-live decision has been confirmed (e.g. 4h accounts marked 「留」 in 方案_v1).
统一成本口径 / Standard cost basis
默认每边手续费 0.2% + 滑点 0.1%，即每边 0.3%、来回 0.6%；压力测试为每边 0.6%、来回 1.2%。资金费套利有两条腿：每条腿每边 0.3%，开仓加平仓合计 1.2% 名义。
资金费按 HTX 历史资金费率计入：持仓期间每到结算时点（ETH 每 8 小时，XAU 每 4 小时），按仓位名义 × 费率计算，多头付正费率、空头收正费率。现货口径、或注明「永续价格作现货代理、只做多」的行不计资金费。做空版写「资金费UNKNOWN」的，是没有计入，不等于 0。程序重放（10-21）用的是程序自带的 manifest 设置（同样是来回 0.6%，另有 5% 止损、单笔风险 0.5%、名义 ≤10%）。
Default costs are 0.2% fee + 0.1% slippage per side, i.e. 0.3% per side and 0.6% round trip. The stress test is 0.6% per side, 1.2% round trip. Funding carry has two legs, each charged 0.3% per side, for 1.2% of notional to open and close.
Funding uses HTX historical funding rates. At each settlement while a position is open (every 8h for ETH, every 4h for XAU), it is charged as notional × rate; longs pay positive rates and shorts receive them. Spot-basis rows, and rows marked "perpetual price as spot proxy, long only", do not include funding. A short row marked 「资金费UNKNOWN」 means funding was not counted, not that it was zero. The 10-21 program replay used the program's own manifest (also 0.6% round trip, plus a 5% stop, 0.5% risk per trade and notional ≤10%).
UNKNOWN 的含义 / Meaning of UNKNOWN
UNKNOWN 表示所列报告里没有这个数字，或者数据拿不到。我们不重新回测、不估算、不拿相近口径的数字来填。所以 UNKNOWN 不是 0，也不代表结果好或坏。
UNKNOWN means the cited report does not contain the figure, or the data is unavailable. We do not re-run backtests, estimate, or substitute a figure from a similar basis. UNKNOWN is not zero and says nothing about whether the result was good or bad.
追加新行的规则 / Rules for adding rows
- 只追加、不修改旧行。新行的 id 接着最大编号往下编。
- 每个数字都必须能在「报告文件路径」所列的报告里找到。报告没有的写 UNKNOWN。
- 发现旧行有误时，不改旧行。另起一行，在策略名称后加「（更正 GY-xxxx）」，在备注写明更正了什么、依据哪份报告。
- 口径必须写清：现货还是永续、是否含资金费、全样本还是样本外、满仓还是名义 10% 或单笔风险 0.5%。
-
追加后运行 〔内部文件〕，必须显示 PASS。〔内部文件〕 只用于首次生成，再次运行时会拒绝覆盖。
-
Append only; never edit existing rows. A new row's id continues from the highest existing number.
- Every number must appear in a report listed in 报告文件路径. Write UNKNOWN if the report lacks it.
- If an existing row is wrong, leave it as is. Add a new row, put 「（更正 GY-xxxx）」 after the strategy name, and say in 备注 what was corrected and which report it is based on.
- State the basis clearly: spot or perpetual, funding included or not, full sample or out-of-sample, full size or 10% notional or 0.5% risk per trade.
- After appending, run 〔内部文件〕; it must print PASS. 〔内部文件〕 is only for the first build and refuses to overwrite the file if run again.
免责声明 / Disclaimer
本表全部来自历史数据上的模拟研究，没有真实交易，不构成投资建议。历史回测不代表未来收益，样本不足的结果尤其不可靠。是否上线由主管审核、Codex 批准。
Everything here comes from simulated research on historical data. No real trades were made, and nothing here is investment advice. Past backtests do not predict future returns; results with insufficient samples are especially unreliable. Any go-live decision is reviewed by the lead and approved by Codex.
对外公开范围（主管 2026-10-22 20:42 裁定）
网站专页只公开状态为「淘汰」的条目；「保留观察」「纸面对照」「执行检验」暂不对外。没有回放过的策略不进坟场。以后每测完一个新策略，当天追加一行，并重新运行 〔内部文件〕。
Public scope (supervisor ruling, 2026-10-22 20:42)
The website page shows only rows with status 「淘汰」 (eliminated). 「保留观察」 (watch), 「纸面对照」 (paper benchmark) and 「执行检验」 (execution check) rows are not public for now. Strategies that were never replayed do not enter the graveyard. Each newly tested strategy is appended as one row on the same day, followed by a rerun of 〔内部文件〕.