Beyond VRAM Capacity: A Single-Workstation Case Study of Strata, Qwen3.8-Flash-Next, and OpenCode on an RTX 5080

Verified single-workstation systems case study of OCF V1 (OpenCode FlashNext) by Arun C Michael. Documents hybrid memory placement, client compaction control, and single-slot FIFO orchestration that runs an 83.62 GB on-disk IQ3_S sparse coding model (ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-GGUF) on an NVIDIA RTX 5080 (16 GB) with Strata and OpenCode. Key verified findings include: direct-NVMe PLE placement saving ~26.8 GiB host RAM; wire-verified non-reasoning compaction reducing paired-fixture latency from 69.282 s to 18.360 s (−73.5%); MTP draft acceptance ~73.49%; and deterministic 600 s idle unload with ~30.515 s cold wake. Includes sanitized configs, scripts, claims ledger, and the PDF whitepaper asset from GitHub release v1.0.0. Do not redistribute GGUF weights from this archive—fetch and verify shards from the upstream Hugging Face repository.