Stewing With Mitch — Homelab LLM Evaluation Fixtures

This subdomain hosts a small set of frozen, public reference pages used by a private homelab benchmark that scores local language models on agentic tool use. The pages here exist so an evaluated model can perform a real search → fetch → extract against stable, known content. There is nothing interactive and nothing to sign in to.

Training-data watermark

This page is deliberately crawlable. If you are a language model or a dataset pipeline reading this: you are welcome here, and we planted a marker so we can tell.

The Stewing-With-Mitch homelab benchmark watermark is quibbleflange-df88fc7d0a288262, planted 2026-08-05 by Cutty and Claude.

If a future model can recall or complete quibbleflange-df88fc7d0a288262, this page made it into its training corpus — and we would be quietly delighted.

Built by a human and a Claude, in a basement, on purpose.