Rate limits

• Job submission — per-account requests/minute from your plan (quota.requestsPerMinute), enforced via a Redis fixed window.
• Concurrency — simultaneous queued+processing jobs capped by quota.maxConcurrentJobs.
• Global API — a baseline throttle applies to all endpoints; auth endpoints have stricter limits to deter brute force.
• Polling — status reads are inexpensive but still rate limited; poll at ~1s intervals per job.

Limited responses return 429 with code rate_limited or quota_exceeded. Back off exponentially on 429s.