Merit AC™
Library / Building with Claude Code

Your first eval set for an AI feature

Build a 30-case test set from real inputs, grade each output pass or fail against written properties, and run it on every prompt change. Worked example: a support-reply drafter.

Playbook · 25 min · Members

What's inside
  • The worked example
  • Step 1: Collect 30 real inputs
  • Step 2: Write expected properties, not expected text
  • Step 3: Grade pass or fail
  • Step 4: Track it in a spreadsheet
  • Step 5: Run it on every change
  • Step 6: Grow it from real failures
  • Common mistakes

All 36 pieces with AI Skills Academy · $49/mo

  • Every playbook, template, checklist and worksheet in the Library, and each new one as it's added
  • The full 30-lesson course, the Agent Authority Scanner and your evidence package
  • Six governed agents to download and run in Claude Code
Join · $49/moWhat's in the Academy

Cancel any time from the billing link in your receipt.

More in Building with Claude Code

A CLAUDE.md that worksTemplate · 12 min · FreeClaude Code permissions before you let it runChecklist · 20 min · MembersCode review with Claude CodePlaybook · 15 min · MembersShould this be an agent, a script or a personWorksheet · 20 min · Members