模型回退冷却过严导致越级烧钱：为 fallback 链加预算与可达性校验

• You use multi-provider fallback (subscription/OAuth + paid API providers).
• Per-provider usage/cost logs are available for auditing fallback decisions.

1. Trigger a controlled primary-model rate-limit scenario and capture fallback attempt order from logs.
2. Check whether provider-level cooldown causes full provider skip when all profiles are marked cooling down.
3. Set temporary policy: place low-cost providers first and enforce hard spend guard for paid fallback tiers.
4. After cooldown expiry window, re-test fallback and compare whether OAuth/subscription providers are retried before paid providers.

• Issue describes current behavior; exact cooldown semantics may change in patch releases（需验证）.
• Do not infer billing safety from one incident; keep ongoing cost anomaly alerts enabled.

This tip is aggregated from community/public sources and preserved with attribution.