메인 봇 PF 0.89, ML/멀티심볼/공개API 피처 전수 테스트 실패 이력을 정리하고,
피처 추가가 아닌 접근 방식 전환(MTF 풀백)으로의 의사결정 맥락을 기술.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
전략 핵심 아이디어, Module 1~4 작동 원리 상세, 메인 루프 흐름도,
메인 봇과의 차이점 비교표를 추가. 테스트 수치도 최신화(183→191).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add _remove_incomplete_candle() for timestamp-based conditional
slicing on both 15m and 1h data (replaces hardcoded [:-1])
- Add MetaFilter indicator caching to eliminate 3x duplicate calc
- Fix notifier encapsulation (_send → notify_info public API)
- Remove DataFetcher.poll_update() dead code
- Fix evaluate_oos.py symbol typo (xrpusdtusdt → xrpusdt)
- Add 20 pytest unit tests for MetaFilter, TriggerStrategy,
ExecutionManager, and _remove_incomplete_candle
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Fetches MTF trade JSONL from prod server via scp, calculates
win rate / PF / cumulative PnL / avg duration by Total/LONG/SHORT,
and outputs LIVE deploy go/no-go verdict (trades >= 5 AND PF >= 1.0).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- mtf_bot: ExecutionManager saves entry/exit records to
data/trade_history/mtf_{symbol}.jsonl on every close
- Jenkinsfile: split MTF_CHANGED from BOT_CHANGED so mtf_bot.py-only
changes restart mtf-bot service without touching cointrader
- docker-compose: mount ./data:/app/data on mtf-bot for persistence
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- README: add MTF bot to feature list and project structure
- ARCHITECTURE: add section 5-1 with MTF bot architecture, data flow,
and design principles (4-module structure, 250 candle buffer, etc.)
- CLAUDE.md: add mtf-pullback-bot to plan history table
- mtf_bot.py: fix stale comment (maxlen=200 → 250)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
deque(maxlen=200) + [:-1] slice left only 199 completed candles,
causing EMA 200 to return NaN and the bot to stay in WAIT forever.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Loop sleep 30s → 1s to never miss the 4-second TimeframeSync window
- Data polling remains at 30s intervals via monotonic timer
- Force poll before signal check to ensure fresh data
- Add [Heartbeat] log every 15m with Meta/ATR/Close/Position
- HOLD reasons now logged at INFO level (was DEBUG)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Sends alerts for: bot start, virtual entry (LONG/SHORT with SL/TP),
and SL/TP exits with PnL. Uses existing DiscordNotifier via
DISCORD_WEBHOOK_URL from .env. Also added env_file to mtf-bot
docker-compose service.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
ccxt requires aiohttp>=3.10.11, was pinned to ==3.9.3.
python-binance has no upper bound on aiohttp, so this is safe.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- exchange.py: cancel_all_orders() now cancels both standard and algo orders
- exchange.py: get_open_orders() merges standard + algo orders
- exchange.py: cancel_order() falls back to algo cancel on failure
- bot.py: store SL/TP prices for price-based close_reason re-determination
- bot.py: add _cancel_remaining_orders() for orphan SL/TP cleanup
- bot.py: re-classify MANUAL close_reason as SL/TP via price comparison
- bot.py: cancel orphan orders on startup when no position exists
- tests: fix env setup for testnet config and ML filter mocking
- docs: add backtest market context and algo order fix design specs
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Based on production API verification showing STOP_MARKET/TAKE_PROFIT_MARKET
are handled as algo orders with different behavior than testnet.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add BINANCE_TESTNET env var to switch between live/demo API keys
- Add KLINE_INTERVAL env var (default 15m) for configurable candle interval
- Pass testnet flag through to Exchange, DataStream, UDS, Notifier
- Add demo mode in bot: forced LONG entry with fixed 0.5% SL / 2% TP
- Fix UDS close_reason: use ot (original order type) field to correctly
classify STOP_MARKET/TAKE_PROFIT_MARKET triggers (was MANUAL)
- Add UDS raw event logging with ot field for debugging
- Add backtest market context (BTC/ETH regime, L/S ratio per fold)
- Separate testnet trade history to data/trade_history/testnet/
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Document why ML filter was disabled (ML OFF PF 1.16 > ML ON PF 0.71),
why SOL/DOGE/TRX were removed (all PF < 1.0 in walk-forward), and
what conditions are needed to re-enable ML in the future.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Collect top trader account L/S ratio and global L/S ratio every 15 minutes
for XRP, BTC, ETH (6 API calls/cycle) and persist to per-symbol parquet files.
Deployed as a separate Docker service reusing the bot image.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add set_margin_type("ISOLATED") call before each position entry.
With cross margin, a single bad trade's loss draws from the entire
account balance. Isolated margin caps loss to the position's allocated
margin only.
Binance returns -4046 if already ISOLATED, which is silently ignored.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
After keepalive ping timeout, the existing BinanceSocketManager's
listenKey becomes invalid. Reusing it causes silent failure — the
WebSocket appears connected but receives no ORDER_TRADE_UPDATE events.
This caused SL/TP fills to be detected only by the 5-minute position
monitor polling (SYNC) instead of real-time User Data Stream (TP/SL),
affecting 11 of 12 production trades.
Fix: create fresh AsyncClient + BinanceSocketManager on every reconnect
iteration, ensuring a valid listenKey is obtained each time.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- SOL/DOGE/TRX removed: all showed PF < 1.0 across all parameter combinations
in strategy sweep (2026-03-17) and live trading (2026-03-21)
- ML filter disabled: Walk-Forward showed ML ON PF < ML OFF PF;
ablation confirmed signal_strength/side dependency (A→C drop 0.08-0.09)
- XRP ADX threshold: 30 → 25 (ADX=30 blocked all trades in current market)
- Current production: XRPUSDT only, SL=1.5x TP=4.0x ADX≥25, NO_ML_FILTER=true
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Adds Fast Kill (8 consecutive losses) and Slow Kill (PF < 0.75 over 15 trades)
to the backtester, matching bot.py behavior. Without this, ML OFF overtrades
and self-destructs, making ML ON look artificially better.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Runs WalkForwardBacktester twice (use_ml=True/False), prints side-by-side
comparison of PF, win rate, MDD, Sharpe, and auto-judges ML filter value.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add LOOKAHEAD embargo between train/val splits in all 3 WF functions
to prevent label leakage from 6h lookahead window
- Add --ablation flag to train_model.py for signal_strength/side
dependency diagnosis (A/B/C experiment with drop analysis)
- Criteria: A→C drop ≤0.05=good, 0.05-0.10=conditional, ≥0.10=redesign
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Module-level ATR_SL_MULT was 1.5, now 2.0 to match config.py and CLI defaults.
This ensures generate_dataset_vectorized() produces correct labels even without
explicit parameters.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- MLFilter.from_model() classmethod eliminates brittle __new__() private-attribute
manipulation in backtester walk-forward model injection
- backtest_validator._check_invariants() now accepts cfg and uses cfg.initial_balance
instead of a hardcoded 1000.0 for the negative-balance invariant check
- backtester.py walk-forward injection block simplified to use the new factory method
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add normalize=False parameter to MLXFilter.fit() so external callers
can skip internal normalization. Remove the external normalization +
manual _mean/_std reset hack from walk_forward_auc() in train_mlx_model.py.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add atr_sl_mult and atr_tp_mult parameters to _calc_labels_vectorized
and generate_dataset_vectorized, defaulting to existing constants (1.5,
2.0) for full backward compatibility. Callers (train scripts, backtester)
can now pass symbol-specific multipliers without modifying module-level
constants.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add pagination controls to Trades tab (prev/next, offset support)
- Reset page on symbol change
- Use package-lock.json + npm ci for reproducible UI builds
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
C1: /api/reset에 API key 인증 추가 (DASHBOARD_RESET_KEY 환경변수)
C2: /proc 스캐닝 제거, PID file + SIGHUP 기반 파서 재파싱으로 교체
C3: daily_pnl 업데이트를 trades 테이블에서 재계산하여 idempotent하게 변경
I1: CORS origins를 CORS_ORIGINS 환경변수로 설정 가능하게 변경
I2: offset 파라미터에 ge=0 검증 추가
I3: 매 줄 commit → 파일 단위 배치 commit으로 성능 개선
I4: _pending_candles 크기 제한으로 메모리 누적 방지
I5: bot.log glob 중복 파싱 제거 (sorted(set(...)))
I6: /api/health 에러 메시지에서 내부 경로 미노출
I7: RSI 차트(데이터 없음)를 OI 변화율 차트로 교체
M1: pnlColor 변수 shadowing 수정 (posPnlColor)
M2: 거래 목록에 API total 필드 사용
M3: dashboard/ui/.dockerignore 추가
M4: API Dockerfile Python 3.11→3.12
M5: 테스트 fixture에서 temp DB cleanup 추가
M6: 누락 테스트 9건 추가 (health, daily, reset 인증, offset, pagination)
M7: 파서 SIGTERM graceful shutdown + entrypoint.sh signal forwarding
DB: 양쪽 busy_timeout=5000 + WAL pragma 설정
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
C5: Remove duplicate entry_fee deduction in backtester (balance and net_pnl)
C1: Add SL/TP retry (3x) with emergency market close on final failure
C3: Add _close_lock to prevent PnL double recording between callback and monitor
C8: Add SIGTERM/SIGINT handler with per-symbol order cancellation before exit
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- fix(data_stream): add reconnect loop to MultiSymbolStream matching UserDataStream pattern
Prevents bot-wide crash on WebSocket disconnect (#3 Critical)
- fix(data_stream): increase buffer_size 200→300 and preload 200→300
Ensures z-score window (288) has sufficient data (#5 Important)
- fix(bot): sync risk manager when Binance has no position but local state does
Prevents ghost entries in open_positions blocking future trades (#1 Critical)
- fix(ml_filter): add np.nan_to_num for ONNX input to handle NaN features
Prevents all signals being blocked during initial ~2h warmup (#2 Critical)
- fix(bot): replace _close_handled_by_sync with current_trade_side==None guard
Eliminates race window in SYNC PnL double recording (#4 Important)
- feat(bot): add _ensure_sl_tp_orders in _recover_position
Detects and re-places missing SL/TP orders on bot restart (#6 Important)
- feat(exchange): add get_open_orders method for SL/TP verification
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- fix(bot): prevent PnL double recording in _close_and_reenter using asyncio.Event
- fix(bot): prevent SYNC detection PnL duplication with _close_handled_by_sync flag
- fix(notifier): move sync HTTP call to background thread via run_in_executor
- fix(risk_manager): make is_trading_allowed async with lock for thread safety
- fix(exchange): cache exchange info at class level (1 API call for all symbols)
- fix(exchange): use `is not None` instead of truthy check for price/stop_price
- refactor(backtester): extract _calc_trade_stats to eliminate code duplication
- fix(ml_features): apply rolling z-score to OI/funding rate in serving (train-serve skew)
- fix(bot): use config.correlation_symbols instead of hardcoded BTCUSDT/ETHUSDT
- fix(bot): expand OI/funding history deque to 96 for z-score window
- cleanup(config): remove unused stop_loss_pct, take_profit_pct, trailing_stop_pct fields
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Load trade history from data/trade_history/{symbol}.jsonl
- Show per-symbol: consecutive loss streak vs threshold, recent 15-trade PF
- 2-tier alert: clean numbers for normal, ⚠/🔴 KILLED for danger zone
- Inserted before ML retraining checklist in Discord report
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Replace full file scan with _tail_lines() that reads from EOF
- Only load max(FAST_KILL=8, SLOW_KILL=15) lines on boot
- Trim in-memory trade history to prevent unbounded growth
- No I/O bottleneck regardless of history file size
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Fast Kill: 8 consecutive net losses → block new entries for symbol
- Slow Kill: last 15 trades PF < 0.75 → block new entries for symbol
- Trade history persisted to data/trade_history/{symbol}.jsonl (survives restart)
- Boot-time retrospective check restores kill state from history
- Manual reset via RESET_KILL_SWITCH_{SYMBOL}=True in .env + restart
- Entry blocked, exits (SL/TP/manual) always work normally
- Discord alert on kill switch activation
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
