Since Sept 26, almost every request to typesafe-ai/jev via /v1/evaluate returns 429 "The upstream provider is currently experiencing high demand". The error metadata shows providerAttemptCount: 0, so the request is never dispatched. On Sept 25 the same code ran 1,500+ requests successfully. Now even 1 request per minute fails (1 success out of 16), a new API key behaves the same, and other models work fine with the same key. The endpoints API shows provider uptime around 99.97%.

Is anyone else seeing this with Jev on AI Gateway? And is there per-team load shedding after heavier usage, and if so, how long does it last?