Process a single site end-to-end: blog → articles → links → download → save.

Result summary for one site after processing.

Pick a provider by weighted random selection, excluding unhealthy ones.

Track provider health; disable after consecutive failures, auto-recover.

Unified LLM interface with weighted routing, fallback, and health tracking. Uses ``AsyncOpenAI`` to avoid blocking the asyncio event loop during

Dispatch multiple sites concurrently with a shared LLMRouter. Usage: config = load_config() scheduler = Scheduler(config)

Mock for ChatCompletion response message.

Merge site output files into cross-site aggregated files.