Data is fetched from the lowest-priority-number enabled exchange; if it fails, the next enabled one is tried. Never mixed within one request.

If the primary provider fails or isn't configured, the fallback is tried automatically before giving up - same pattern as the exchange data source.

Endpoint applies to Ollama only.

Qwen only explains the prediction that Phase 1 already computed - it never sets or changes win probability, entry, stop loss, or take profit.

While on, the Telegram bot replies to every user with a maintenance notice and does nothing else.