Our method
How we test
Most Claude tutorials show something working once. We try to find out when it stops working. Here is what happens behind every guide.
1. Plant a trap
We build a small, fake project and hide something in it: a bug, a wrong number, a secret, a hostile instruction. Because we put it there, we know the right answer before Claude does.
2. Run it again and again
One lucky run proves nothing. We repeat the task several times with and without the thing being tested (a rules file, a hook, a skill, a spec) and count what happens with a script, not by eye.
3. Check the numbers ourselves
When Claude gives us a figure, we recompute it independently. Mistakes we find, in Claude or in our own test, go in the guide.
4. Report what broke
Every guide ends with what it does not tell you: how few runs we did, what we skipped, and where a real project would differ.
What to know about our guides
- Screenshots come from our own runs. Personal details are covered or cropped out.
- Cover illustrations are AI-generated. Everything else is written and checked by us.
- Costs we quote are the figures Claude Code reported. We use a Pro plan, so we were not billed per run.
- Our tests are small on purpose. They show what can happen, not how often it will in your project.
- Discovia is an independent publication. It is not affiliated with, endorsed by, or sponsored by Anthropic. Claude is a trademark of Anthropic, PBC.
Ready to read one? Browse all guides.
New guides and tools, in your inbox.
No hype. Real results, including the ones that did not work. Unsubscribe in one click.