Working papers from the engine room: gate batteries, compiler bugs, serving forensics — every number links to a raw log.https://tiyuvta.ai/enhttps://tiyuvta.ai/blog/batch-changed-the-models-mind/https://tiyuvta.ai/blog/batch-changed-the-models-mind/Under batched prefill, co-arrivals changed a MoE request's expert selection in 16% of (layer,token) pairs. How the isolation gate caught it, and the fix.Wed, 05 Aug 2026 00:00:00 GMThttps://tiyuvta.ai/blog/compiler-ate-my-byte/https://tiyuvta.ai/blog/compiler-ate-my-byte/nvcc 13.0.88 at -O3 for sm_120a zeroed one byte per quantized block. A bit-parity gate caught it on the first run. The repro, the fix, the receipts.Wed, 05 Aug 2026 00:00:00 GMThttps://tiyuvta.ai/blog/dogfood-day/https://tiyuvta.ai/blog/dogfood-day/Dogfooding exposed three sampler bugs of one class that my entire gate battery could not see. Fixes, receipts, and the gates that now catch them.Wed, 05 Aug 2026 00:00:00 GMThttps://tiyuvta.ai/blog/spec-decode-economics/https://tiyuvta.ai/blog/spec-decode-economics/MTP speculative decoding takes a 27B to 170.5 tok/s single-stream on an RTX PRO 6000, token-identical to plain decode — and what that does to $/token.Wed, 05 Aug 2026 00:00:00 GMThttps://tiyuvta.ai/blog/session-affinity/https://tiyuvta.ai/blog/session-affinity/Agent clients rewrite history between turns, defeating prefix caching. Session affinity keeps TTFT flat across 25 turns — plus the gap it does not close.Wed, 05 Aug 2026 00:00:00 GMT