Spec a client-side limit for a partner API quota

Your Go sync job calls a partner's API from many goroutines at once. The partner allows 100 requests per second per API key and answers 429 when you go over. After 3 days with too many 429 responses, they suspend the key. An agent will add rate limiting to the client.

Write the spec you would hand the agent.

What you know:

• one process runs the sync, and it is the only user of this API key
• about 200 goroutines make calls, and each call takes 50 to 500 ms
• the partner counts requests in any rolling 1-second window
• a 429 response carries a Retry-After header, in seconds

Then, in a few sentences: what does "throttle the API calls to 100 per second" get wrong?