Small, honest open-source tools. Every claim in them points at its evidence: a quoted policy sentence, a recorded run, a test that fails when reality disagrees.
| Project | What it is | |
|---|---|---|
| Will it run on a Mac? | Paste a shell script, see every flag that does not exist on macOS or BusyBox, with that platform's own man page line and a real run as proof. Every flag of 81 tools and 86 find primaries is executed on real Ubuntu, macOS and Alpine runners weekly. | open |
| Why didn't my job run? | Paste a GitHub Actions workflow, pick the event, see which jobs run and why. GitHub's own workflow parser and expression engine, running in your browser; checked against 65 recorded real runs. | open |
| PrivacyMatrix | Does this AI app train on your chats? 28 assistants × 14 privacy questions; every cell quotes the vendor's own policy and is re-checked against its source weekly. | open |
| AgentMatrix | A cited feature matrix of AI coding agent CLIs; every cell quotes the vendor's documentation and is re-checked weekly. | open |
wdmjr-fixtures holds the real GitHub Actions runs the simulator is tested against; every run is visible under its Actions tab.
How these are built. I direct, decide what to verify and what to ship; Claude writes most of the code. What makes that acceptable is the harness around it: quotes that must be found at their URL, runs that must replay, weekly checks that demote a cell when its source changes. When the tool and reality disagree, reality wins and becomes a test.
Nothing here has a server, a key or a bill. Fork it, run it with your own credentials, open an issue when a result is wrong.