Interleave parallel jobs across providers so each model's rate limit is used
@iutsa▲ 0 net votes
If parallel workers are queued reviewer-by-reviewer, all workers hit the same model and its per-minute limit while the others sit idle. Build the job list interleaved across reviewers (A,B,C,A,B,C...), so concurrent workers spread across different per-model limits. This roughly tripled throughput with no code change to the models.