claude-task.md

Implement real LONG/SHORT trading on Binance Testnet with full visibility in the Testnet tab.

Context

• The DRL trading bot already makes LONG/SHORT decisions (PPO agent + composite scorer + 3-tier decision system)
• Currently the bot's trades are DRY RUN only (logged but not executed on the exchange)
• Binance Testnet API keys are in .env: BINANCE_TESTNET_API_KEY and BINANCE_TESTNET_API_SECRET
• The testnet connects to https://testnet.binance.vision/
• The server runs locally and can reach Binance testnet directly (no proxy needed)
• The BinanceConnector class in src/api/binance.py handles exchange connectivity

Requirements

1. Testnet Trade Execution

• Mirror the EXACT same trading logic the bot uses for dry-run trades (from live_trading_multi.py)
• When the bot decides to OPEN_LONG, OPEN_SHORT, CLOSE_LONG, CLOSE_SHORT — execute the same trade on Binance testnet
• Use the same position sizing, SL/TP logic, split entry (50% market + 50% limit)
• Store testnet trades separately from the bot's dry-run trades (different MongoDB collection: testnet_trades)
• Each testnet trade should record: symbol, action, side, price, amount, order_id, timestamp, pnl, sl, tp, confidence

2. Testnet Trade API Endpoints (add to api_server.py)

• GET /api/testnet/trades — all testnet trade history
• GET /api/testnet/positions — current open positions on testnet
• GET /api/testnet/pnl — realized + unrealized PNL from testnet trades
• POST /api/testnet/execute — manually trigger a testnet trade (for testing)

3. Testnet Tab UI (update app.py testnet section)

Show full visibility:

• Open Positions table: symbol, side (LONG/SHORT), entry price, current price, unrealized PNL, SL, TP
• Trade History table: all executed testnet trades with timestamp, symbol, action, price, amount, PNL, order_id
• PNL Summary: total realized PNL, total unrealized PNL, win rate, total trades
• Equity Curve chart: plot cumulative PNL over time from testnet trades
• Live Order Book: show any open/pending orders on testnet
• Make sure ALL data comes from the API endpoints (client-mode compatible)

4. Auto-Execution Hook

• Add a hook in the trading loop (live_trading_multi.py or api_server.py) that:
  • Listens for bot trade decisions
  • Mirrors each decision to testnet in real-time
  • Logs both the bot's dry-run result and the testnet execution result
• This should be toggleable via env var: TESTNET_MIRROR=true/false

5. Integration Tests

• Test testnet trade execution (place a small market buy on testnet)
• Test /api/testnet/trades returns trade history
• Test /api/testnet/positions returns open positions
• Test PNL calculation from testnet trades

Important Rules

• NO mock/fake data — all trades must be real testnet executions
• NO hardcoded values — all prices, amounts from real API responses
• Guard all None values in f-strings (ui-type-safety skill)
• Push to HF dev space after changes
• Check container logs — auto-fix loop until zero errors

Git/Deploy

• Work on hf-clean branch
• Push to chen470/drl-trading-bot-dev2
• HF_TOKEN from .env
• Factory restart after push
• Verify container logs clean

Write plan to TESTNET_TRADING_PLAN.md first, then implement.