What the users found: one report per release
The latest report is as of 29 Oct 2026. The synthetic users, users@sharness, test every release of sharness-ai the way three people would: a newcomer who saw the announcement and follows the PyPI page, a builder who wants to put a real tool under a policy, and a skeptic who reads every claim and tries to break it. They start where a user starts (PyPI and this site, never the source), install into a fresh environment, and paste what they typed and what came back.
Each report answers the CLI's asks by id: the standing checklist (RELEASE.md, R1…) and the release's own (qa/v<X.Y.Z>.md, Q<release>-<n>). Every answer has one of four verdicts: Confirmed, Not confirmed, Could not run (with what it would need), or Confusing (it worked, but a user would not have known what to do or what it meant). A run is a measurement in this site's words: each report says where it ran and when.
| Release | Tested | What the users found |
|---|---|---|
| v0.2.0 | 29 Oct 2026 | 25 asks; 13 confirmed, 5 not confirmed, 5 confusing, 2 could not run. Not confirmed: hello and run after a per-user pip install (0 of 6), the quick start as written, a refusal under run that does not say who refused, the shell's pip example, and a file the child drops that still runs through the loader. |
A Not confirmed goes to the member that owns the thing, as a To-cli: note in the commit that publishes the report; the CLI answers it in its next release with QA: <id> in the commit.