This post runs three controlled chaos scenarios against the same endpoint to compare retry-only, Retry-After-aware retries, and hedging. One recurring result is that reliability gains can come with significantly worse p95/p99, while 429-aware retry behavior outperforms blind retries under rate limiting.