Rate limits

Limits apply per token and are generous for real workloads (directional, final numbers at GA): reads 600/min, writes 120/min, order creation 60/min, file-manifest registration 20/min. Every response carries X-RateLimit-Limit, X-RateLimit-Remaining, and X-RateLimit-Reset; a 429 carries Retry-After.

• Agents: honor Retry-After, back off with jitter, and never poll faster than the event stream can move. The snapshot endpoint is cheap; hammering it is still pointless because webhooks and replay exist.
• Bursts: order creation is idempotent on external_order_id, so a retried burst cannot double-book. Identical duplicates replay the original response; 409 fires only when the same ID arrives with a different payload.