Sequel to the 35B post: trying Qwen3.8-Flash-Next (177B MoE) on the same Unraid box via AtomicChat's SSD-offloading quants. Measured 1.1 t/s on 64 GiB of RAM — plus every trap found on the way (Ollama import bug, Xet downloads, docker tags, sharded GGUFs). Benchmark lands with the 96 GiB upgrade.