How we test
UseCase AI only lists use cases with steps, a description of the outcome, and credit to the author. The labels Tested by us / Partly tested / Not tested by us say whether we ran it ourselves, not whether the author posted screenshots.
How we do it
- When the steps can be followed, we run them in a clean setup. If we can't finish, we mark the entry untested or partly tested and say where we got stuck.
- “What the author says” and “What we found when we tried it” are kept separate, so the author's claims never read as our findings.
- Some conclusions need a side-by-side run, for example with and without a prompt. Until we run one, the page says “We haven't compared them side by side yet”, and we don't invent numbers.
Environment
We usually test on Linux, where most command-line steps also run. The OS, model and version for each entry are under “Version we used” and “Tested on”; login walls and regional limits are under “How to get in”.
What we publish
- Steps clear enough for someone else to start.
- An outcome description (author claims are fine if it's clear who said them).
- A checkable author or source link.
- Entries with only a one-line description and a homepage link stay unpublished until they're complete.
Follow & video
Short updates at X (Twitter) (@fucddx77047) ↗
Clawd short video
The Clawd series is on Douyin and Xiaohongshu — search “Clawd” or use the link in those profiles. Profile URLs change often, so we don't hard-code them here.