2026-04-12 provider ordering criteria discussion

Source: Telegram conversation with Buu on 2026-04-12.

Key durable points extracted:

• Buu does not rank upstream providers or groups by success rate alone.
• The ordering logic is a combined judgment across at least:
  • real cost
  • free quota / subsidy / promotional room
  • stability
  • practical fit for the actual workload
• In fallback chains, cheaper or free groups may intentionally be placed earlier even if they are less stable.
• More stable but more expensive groups may intentionally be placed later as fallback capacity.
• The point of this ordering is not abstract purity but matching Buu’s actual cost-performance preference under real provider instability.
• This rule is intended to apply across text models, image models, and video models, not just one specific model.
• Future analysis should not assume that the highest observed success rate should automatically move a provider to the front of the chain.