lulu/arbitrage-engine
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
136 Commits 8 Branches 5 Tags 3.9 MiB
318bcb99a3
Commit Graph

27 Commits

Author SHA1 Message Date
root
318bcb99a3 fix: clamp direction score to max weight (accel_bonus included) 2026-03-02 01:36:20 +00:00
root
d4f0bbbcb1 fix: read fundingRate field (not lastFundingRate) from market_indicators 
DB stores Binance API response with 'fundingRate' key, but code was
looking for 'lastFundingRate' - always got 0.
 2026-03-02 01:07:05 +00:00
root
1fa6f178b6 feat: linear FR scoring based on historical max 
- FR score = (|current_FR| / max_historical_FR) × 5, clamped to [-5, +5]
- Contrarian direction: positive score (going against crowded side)
- Same direction: negative score (going with crowded side)
- Historical max FR cached per symbol, refreshed every hour
- Score is float with 2 decimal places
- Total score supports 1 decimal place
 2026-03-02 00:56:22 +00:00
root
05673c0850 fix: V5.2 scoring uses strategy weights, total capped at 100 
Backend:
- Each layer score scaled by strategy config weights
- direction: 0~40, crowding: 0~18, FR: 0~5, environment: 0~12
- confirmation: 0~15, liquidation: 0~5, auxiliary: 0~5
- total_score clamped to 0~100
- factors include max field for frontend

Frontend:
- V5.2 signals page reads max from factors
 2026-03-02 00:34:22 +00:00
root
01b1992643 feat: save factors to signal_indicators + show FR/Liq on signals page 
- DB: added factors JSONB column to signal_indicators
- Backend: save_indicator now includes factors JSON
- API: /api/signals/latest returns factors field
- Frontend: signals page shows FR and 清算 score bars
 2026-03-01 23:57:55 +00:00
root
a571a221e5 feat: gradient liquidation scoring - no threshold, ratio-based 
Removed USD threshold gates. Now any liquidation data triggers scoring:
- ratio ≥ 2.0 → +5 (extreme imbalance)
- ratio ≥ 1.5 → +3
- ratio ≥ 1.2 → +1
- ratio 0.8~1.2 → 0 (balanced)
- Inverse for SHORT direction
 2026-03-01 23:35:45 +00:00
root
8279b16c68 fix: FR thresholds based on actual data 
Historical FR range: BTC ±0.005%, SOL/XRP max -0.022%
Old thresholds were unreachable (0.03% extreme / 0.005% moderate)

New thresholds:
- Extreme: ±0.01% (0.0001) — clearly biased market
- Moderate: ±0.003% (0.00003) — slight bias, reward contrarian
 2026-03-01 23:23:43 +00:00
root
7d2bb9f392 fix: FR/Liq scoring bugs + compact UI 
Backend:
- FR threshold: 0.001→0.0003 (extreme), 0.0003→0.00005 (moderate)
- Liquidation SQL: side='SELL'/'BUY' → 'LONG'/'SHORT' (was never matching!)

Frontend:
- FR/Liq scores now inline in score column (compact)
- Removed bulky green badge buttons from position cards
 2026-03-01 23:14:07 +00:00
root
a9c3523a24 feat: independent strategy paper trading controls 
- Each strategy has its own position count (max 4 each)
- paper_config.enabled_strategies: per-strategy toggle
- is_strategy_enabled() checks both master switch and strategy list
- API: /api/paper/config now supports enabled_strategies array
- Config auto-loads on API startup

Usage: POST /api/paper/config {enabled_strategies: ['v52_8signals']}
 2026-03-01 12:43:46 +00:00
root
732b01691b Implement V5.2 FR/liquidation scoring and strategy AB loop 2026-03-01 11:54:58 +00:00
root
2f9dce483c fix: simulate limit orders for TP/SL (match real trading) 
- TP/SL now exit at order price (limit order), not market price
- SL exits at sl_price, TP1 at tp1_price, TP2 at tp2_price
- Only timeout and signal_flip use market price (current price)
- Updated fix_historical_pnl.py to also correct exit_price
- This eliminates fake slippage in paper trading stats
 2026-03-01 09:40:00 +00:00
root
45bad25156 fix(P0): pnl_r calculation + cooldown bypass + partition month bug 
P0-1: Reverse signal now bypasses cooldown - evaluate_signal always
      outputs direction, main loop checks direction+score>=60 for
      closing positions against trend (not blocked by COOLDOWN_MS)

P0-2: pnl_r unified to (exit-entry)/risk_distance across all exit
      scenarios (tp, sl, sl_be, timeout) in both paper_monitor.py
      and signal_engine.py. Old hardcoded values (1.5R/2.25R) were
      ~2x too high vs actual risk_distance basis.

P1-1: ensure_partitions month calculation fixed from timedelta(30d)
      to proper month arithmetic. Also fixed UTC timezone for
      partition boundaries.

docs: V52-TODO.md with full audit backlog for V5.2
 2026-03-01 09:29:32 +00:00
root
9528d69a42 fix: deduplicate recent_large_trades to prevent memory bloat 
Each evaluate cycle was re-appending all qualifying trades from
win_fast, causing duplicates. Added seen-set dedup.
 2026-03-01 07:58:51 +00:00
root
022ead6d6c feat: paper_trades store score_factors JSONB (direction/crowding/environment/confirmation/auxiliary breakdown) 2026-02-28 13:45:26 +00:00
root
95b45d0f07 fix: skip first 3 cycles after cold start to prevent duplicate paper trades 2026-02-28 12:03:48 +00:00
root
7b901a2390 feat: paper_monitor.py - WebSocket realtime TP/SL + frontend WS price feed (1s update) 2026-02-28 11:55:40 +00:00
root
66810701fb feat: paper trading signal flip - reverse signal closes existing position then opens new 2026-02-28 11:45:48 +00:00
root
47004ece8c feat: paper trading deduct taker fee 0.05% per side (0.1% round trip) 2026-02-28 11:30:17 +00:00
root
282aed138a feat: paper trading switch + config API + max positions limit 2026-02-28 11:13:39 +00:00
root
e054db112d feat: paper trading - backend (table+signal_engine integration+5 APIs) + frontend page 2026-02-28 11:10:28 +00:00
root
317031ab57 fix: signal_engine now reads real market_indicators (JSONB parsing + OI change rate) 2026-02-28 11:03:44 +00:00
root
0ac2225979 feat: add XRP and SOL symbols (4 coins total) 2026-02-28 10:11:05 +00:00
root
424cb993f8 feat: signal history list + always compute scoring even without signal 2026-02-28 06:09:32 +00:00
root
5c38a2f9bf feat: upgrade signal engine to V5.1 layered scoring 2026-02-28 05:24:16 +00:00
root
fbf84f2be5 tune: signal-engine循环15秒(从5秒,降CPU60%) 2026-02-27 18:38:03 +00:00
root
4168c1dd88 refactor: SQLite→PostgreSQL migration - db.py连接层 + main/collector/signal-engine/backfill全部改PG 
Phase 1: 核心数据表(agg_trades/rate_snapshots/signal*)迁PG
auth.py暂保留SQLite(低频，不影响性能)
- db.py: psycopg2同步池 + asyncpg异步池 + PG schema + 分区管理
- main.py: 全部改asyncpg查询
- collector: psycopg2 + execute_values批量写入
- signal-engine: psycopg2同步读写
- backfill: psycopg2 + ON CONFLICT DO NOTHING
 2026-02-27 16:15:16 +00:00
root
547f093352 feat: V5 signal-engine + API endpoints for indicators/latest/trades 2026-02-27 15:31:12 +00:00