{"generatedAt":"2026-10-27T08:57:48.921Z","production":{"day":{"requests":3,"ok":3,"failed":0,"throttled":0,"uptime":1,"p50":285,"p95":285,"tps":47.736625514403286,"cache":0},"week":{"requests":3,"ok":3,"failed":0,"throttled":0,"uptime":1,"p50":285,"p95":285,"tps":47.736625514403286,"cache":0}},"benchmark":{"day":{"requests":60,"ok":60,"failed":0,"throttled":0,"uptime":1,"p50":4416,"p95":25043,"tps":44.46054502931492,"cache":0.6138319381088442},"week":{"requests":60,"ok":60,"failed":0,"throttled":0,"uptime":1,"p50":4416,"p95":25043,"tps":44.46054502931492,"cache":0.6138319381088442}},"deployment":{"Region":"Japan","Hardware":"NVIDIA RTX 3090 24 GB per engine","Serving":"vLLM 0.32, W4A16 weights, FP8 KV cache, prefix caching","Concurrency":"2 streams per engine; excess requests receive HTTP 429"}}