[devam2001]

Why

The last go-live gate needs 4+ weeks of continuous paper trading. Until now, a reboot, crash or sleep ended the test: the panel could only start a new session, and the paper-vs-replay check couldn't see that the bot had been offline.

What

• Resume (panel button, POST /api/paper/resume):
  • Restarts the last session's bot on the same database, so the wallet, open trades and history carry on.
  • Uses the same dry-run guard as Start.
  • Is enabled only when a stopped session exists.
• Pinned pair lists:
  • New sessions save their pair list in their own config, so the monthly config.json update can't change a running test.
  • Older sessions get their list back from the bot log when resumed.
• Downtime detection: the paper-vs-replay check reports any stretch of 10+ minutes without log output (the bot logs a heartbeat every minute). The replay trades straight through those gaps.
• Docs: README panel section and the runbook's paper-vs-replay notes.

Testing

• ruff check . clean; pytest 73 passed. New tests cover:
  • resume, including that pairs are pinned to the session's own list
  • the dry-run guard on resume
  • that the risk-control resume (/api/bot/resume) stays separate from session resume
  • gap detection
• Checked live in the panel:
  • Resumed a real October 11 session. It came back as the same session, in dry-run mode, with its 100 USDT wallet and its 25 original pairs recovered from the log. Then stopped it.
  • Downtime detection on that session found the one 10-day offline period.