About

MoonWork Lab is my record of building with AI and measuring what runs on hardware I already own.

I publish local LLM benchmarks, configuration experiments, and the failures behind my tools. The useful part is the condition attached to a result: which model, which setting, what I tried, and what I still haven’t checked.

Who writes this

I publish under the name MoonWork. I’m a solo builder with a day job. I use AI assistants for planning, code generation, and English drafting. This isn’t a claim that I write every line of code by hand.

The projects and hardware are mine. I choose what to build and test, and I take responsibility for what this site publishes. AI helps turn the work into an article; generated prose is not evidence that a test happened.

How I handle evidence

  • Measurements come from recorded runs on my machines. When I cite somebody else’s result, I identify the source.
  • I include settings, commands, tables, or a test fixture where I have retained them. If a version, raw log, or timing wasn’t kept, I say so.
  • A result from one setup stays a result from that setup. I separate an observed failure from my explanation for it.
  • I keep unfavorable results. Buying the hardware doesn’t entitle it to win.
  • When I find an error, I correct the article and leave a dated note describing the material change.

Start with the work

Money and corrections

Advertising and affiliate links are intended ways to support this site. A link doesn’t change a benchmark result. See the affiliate disclosure for details.

Found a mistake, missing condition, or a different result on your machine? Send it through the contact page. Include the article, the relevant command or passage, and the environment you tested. Please remove credentials and private data from logs.

Editorial policy updated October 6, 2026.