jackyu66gitandCursor
e2fbe1c2b3
Keep frozen research conclusions and code; raw ledgers and secrets stay out of git. Co-authored-by: Cursor <cursoragent@cursor.com>
159 lines
5.3 KiB
Markdown
159 lines
5.3 KiB
Markdown
# nautilus_mm — Research Freeze / Data Collection Phase
|
||
|
||
Freqtrade 保留缠论 / 中低频；**Maker / L2 / Fill 事件**迁到 NautilusTrader。
|
||
|
||
**研究对象：可验证的市场现象（不是策略）。**
|
||
|
||
```
|
||
Trading OS
|
||
├─ Market Intelligence (Market Pulse) ← Stage5 才接（Quote Adjustment）
|
||
├─ Execution Reality Layer (本仓库) ← Fill Alpha Dataset
|
||
├─ Freqtrade ← Chan / 中低频
|
||
└─ NautilusTrader ← 事件驱动执行
|
||
```
|
||
|
||
## 冻结研究路径（禁止跳级）
|
||
|
||
```
|
||
Stage 0 Data Integrity
|
||
↓
|
||
Stage 1 Fill Alpha ← 当前
|
||
↓
|
||
Stage 2 Maker Edge Report ← 当前（决策门）
|
||
↓
|
||
Stage 3 Economic Simulation ← LOCKED until Edge PASS
|
||
↓
|
||
Stage 4 Quote Engine
|
||
↓
|
||
Stage 5 Market Regime Adaptation (Market Pulse → Quote Adjustment)
|
||
```
|
||
|
||
| Stage | 目标 | 状态 |
|
||
|-------|------|------|
|
||
| **0** | WS/L2 可信：seq gap / latency / book_age | 探针已记 |
|
||
| **1** | Fill Alpha Dataset：真实成交 + 路径 | **进行中** |
|
||
| **2** | Maker Edge 决策门：PASS / FAIL / COLLECTING | **进行中** |
|
||
| **3** | 经济仿真：quote→fill→inventory→exit（partial/cancel/funding/fee） | **未解锁** |
|
||
| **4** | Quote Engine | 未开始 |
|
||
| **5** | Regime Adaptation | 未开始 |
|
||
|
||
原则：**先证明成交有优势，再谈账户收益，再设计报价。**
|
||
禁止现在加：Quote Engine / AI / Market Pulse 交易信号 / 参数优化。
|
||
|
||
### 核心问题（交给 ~2000 真实 fills）
|
||
|
||
> 个人开发者在 BTC 永续上，通过被动流动性提供，是否能获得统计优势？
|
||
> 若有：优势来自哪里？
|
||
|
||
可能结果：
|
||
|
||
| 情况 | 含义 | 下一步 |
|
||
|------|------|--------|
|
||
| **A** 全市场 PASS | 稳定被动流动性优势 | Symmetric MM |
|
||
| **B** 仅特定状态 PASS | 高波动 / 吸收 / 震荡等 | Event-driven LP（更可能） |
|
||
| **C** 全部 FAIL | 普通 Maker edge 不存在 | Cash Carry / Basis / Funding / 跨所 |
|
||
|
||
**最值得等待的不是 PASS，而是 Edge 来自哪里。** 第一份 FAIL 也是高价值结果。
|
||
|
||
### 防自我欺骗（已内建）
|
||
|
||
1. **Cluster-weighted** — 暴跌 50 笔 Bid ≠ 50 独立样本；同时看 fill-w 与 cluster-w，方向一致才可信
|
||
2. **Matched Mid / Maker Alpha** — `MakerAlpha = Fill Outcome − Market Move`（剥离方向收益）
|
||
3. **Adverse Selection** — 成交是否天然站在错误一侧；spread capture 挡不住毒流
|
||
|
||
Stage 3（Economic Simulation）只在 Edge PASS 后做：partial fill、cancel latency、inventory limit、position aging、funding、fee tier → 真实账户收益分布。
|
||
|
||
## Maker Edge Report v0.1
|
||
|
||
```bash
|
||
./scripts/analyze.sh # 默认 min-fills=2000
|
||
./scripts/analyze.sh 500 # 早期预览（仍为 COLLECTING）
|
||
```
|
||
|
||
报告结构：Data Integrity → Sample Independence → Fill Quality + Benchmark → Adverse Selection（`POSITIVE_EDGE` / `EDGE_AFTER_COST` / `NO_EDGE`）→ Path Attribution → State Stability → Decision。
|
||
|
||
## 快速开始
|
||
|
||
```bash
|
||
cd nautilus_mm
|
||
python3 -m venv .venv && source .venv/bin/activate
|
||
pip install -r requirements.txt
|
||
cp .env.example .env # TESTNET key；代理默认 7897
|
||
|
||
./scripts/run_probe.sh # 攒真实 fills
|
||
./scripts/analyze.sh 2000 # 决策门报告
|
||
```
|
||
|
||
探针只记 quote / fill / outcome。`market_state_snapshot` 预留为 null。
|
||
|
||
## 目录
|
||
|
||
```
|
||
nautilus_mm/
|
||
├── configs/
|
||
├── logs/maker_edge/ # Fill Alpha Dataset (jsonl)
|
||
├── scripts/
|
||
│ ├── run_probe.sh
|
||
│ ├── analyze.sh
|
||
│ ├── analyze_maker_edge.py
|
||
│ └── record_l2_ccxt.py
|
||
└── src/nautilus_mm/
|
||
├── recorder.py
|
||
├── health.py
|
||
├── book_utils.py
|
||
├── run_live.py
|
||
└── strategies/maker_edge_probe.py
|
||
```
|
||
|
||
## 与 Freqtrade
|
||
|
||
| | Freqtrade | nautilus_mm |
|
||
|--|-----------|-------------|
|
||
| 用途 | 缠论 / 中低频 | Maker / L2 / Edge 验证 |
|
||
| 驱动 | K 线 | order book / fill 事件 |
|
||
|
||
## 安全
|
||
|
||
- 默认 `BINANCE_ENVIRONMENT=TESTNET`
|
||
- `ENABLE_TRADING=false` 可只订数据
|
||
- **独立 `.venv`**，勿与 freqtrade 混装
|
||
|
||
## 实验冻结（见 [FREEZE.md](FREEZE.md)）
|
||
|
||
**三不动：** Quote Logic / 成本模型 / PASS·COLLECTING·FAIL
|
||
|
||
| ✅ 冻结期允许 | ❌ Gate 解锁前禁止 |
|
||
|-------------|-------------------|
|
||
| 数据字段 | 新交易规则 |
|
||
| 数据质量检查 | 新过滤条件 |
|
||
| 报告解释能力 | 新收益优化参数 |
|
||
|
||
Decision：`PASS` / `PARTIAL_PASS` / `COLLECTING` / `FAIL`
|
||
观察窗：500 异常 · 2000 初步 · 10000 稳定性；**clusters > fills**。
|
||
第一份报告先看分布（Fill / Cluster / Toxicity），再看 Decision。
|
||
|
||
## 服务器 Data Collection（zun_hk）
|
||
|
||
默认：
|
||
|
||
| | |
|
||
|--|--|
|
||
| Host | `jack@jackyu66.com` |
|
||
| Key | `~/Project/deploy/zun_hk/id_ed25519_hk` |
|
||
| Dir | `/www/Project/nautilus_mm` |
|
||
| Experiment | `MM_EDGE_EXP_001` |
|
||
|
||
```bash
|
||
./scripts/deploy_server.sh
|
||
|
||
ssh -i ~/Project/deploy/zun_hk/id_ed25519_hk jack@jackyu66.com
|
||
nano /www/Project/nautilus_mm/.env # BINANCE_API_KEY / SECRET
|
||
systemctl --user start mm-edge-probe # 用户级 systemd（无需 sudo）
|
||
journalctl --user -u mm-edge-probe -f
|
||
|
||
./scripts/probe_status.sh
|
||
./scripts/pull_report.sh 2000
|
||
```
|
||
|
||
本地短测：`USE_PROXY=true ./scripts/run_probe.sh`
|