Sorting LiteLLM GitHub issues by difficulty to implement. Higher = harder.

{
A full scheduling subsystem requires durable state, worker coordination, fairness and priority semantics, failure recovery, observability, and backward-compatible request handling. The other change is comparatively bounded to an opt-in response/streaming contract and upstream-status translation, despite protocol design and provider compatibility risks.
}
https://github.com/berriai/litellm/issues/31830 4:1 https://github.com/berriai/litellm/issues/26693