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

{
Provider/model-specific throttling spans configuration, request scheduling, shared rate-limit state, concurrency, retries, and multiple provider/model routing paths. Correcting failed-request accounting is comparatively narrower, mainly requiring consistent quota updates across proxy success and error flows.
}
https://github.com/berriai/litellm/issues/32620 3:1 https://github.com/berriai/litellm/issues/21312