A service runs at 75% of its TPM quota during peak. What is the recommended action?
Answer: Request a quota raise or reduce per-call tokens — 75% leaves insufficient headroom for normal spike variance
70% is the typical operating ceiling. At 75%, a normal traffic variability swing pushes you past the limit, triggering 429 errors. The remediation is one of: request a quota raise from Anthropic (they grant generously for established traffic patterns), or reduce per-call token consumption through caching, prompt compression, or output limits.