Field guides, newest first
Working notes from the bench - what we learned getting agents, models and silicon to behave.
2026-10-28 · Runtimes
RidgeFill: The Short Cut Through the Long Wait
Why long prompts stall the first answer, and how a measured translation sheet per upper layer files the paperwork without carrying the parcels - no math, diagrams included.
2026-10-10 · AI useJev and the Decision Interface
Typed decisions with zero output tokens: what Jev-style models are, the benchmark argument quoted fairly, and how we run Jev-Omni behind a vLLM pooling proxy on one R9700.
2026-09-30 · AI useEffective Context Engineering for AI Agents
Attention budgets, right-altitude prompts, just-in-time retrieval, compaction and sub-agent architectures - Anthropic's field post, rendered.
No guides filed under this topic yet.