Free downloadProof Run: Test Notes
The notes behind the Proof Run series: one AI claim per episode, tested on my own setup with a real input and the same task without the tool. How the series tests, what the Proof Meter from 0 to 10 means, and Episode 01 in full: founder skills for Claude Code against plain Claude, with the input, both raw outputs, what I checked and the verdict. One test on my setup, not a benchmark.
What is inside
- Method Every episode in three steps: setup, run, result. Same machine, same input, once with and once without the tool.
- Proof Meter What 0–3, 4–6 and 7–10 mean, and what moves the score up or down.
- Episode 01 claude-skills-founder (MIT) against plain Claude Code on my own product: 37 s against 95 s, sources, saved files, tiers.
- Checked Every link opened, every number recalculated, and what each run got wrong.
- Raw outputs The input and both replies unchanged, plus the founder/ folder the skill wrote.
- Repeat it The command shape to run the same test yourself, with and without the skill.
Nothing to install to read it: a PDF viewer and any text editor for the raw outputs. To repeat the test you need Claude Code, git and your own Claude plan. No account with us. The notes are in English.