Method
How we stamp a receipt
The short version: quote the launch, read the users, link everything, and say “too early” when we don’t know yet.
Companies and reviewers show launch day. We show what happens once real people use it. We’re not here to undermine anyone; we add one more opinion, with every source shown.
1. Pick the launch
We cover major AI launches people actually use to build: models, coding tools, app builders and agents. Launches come from vendor blogs and changelogs, release notes, and what’s trending on Hacker News and in developer forums.
2. Quote the claims, word for word
We open the official launch page and copy three to five concrete, checkable claims exactly as written. We don’t paraphrase claims and we don’t take them from press coverage. Footnote markers are the only thing we drop. Every claim links to the page it came from.
3. Collect real user reports
- Sources: GitHub issues, vendor forums (Discourse), Reddit, X and Hacker News comments.
- We open every page we cite. A search snippet never counts as a source.
- We paraphrase in our own words and leave out usernames, emails and ticket or request IDs.
- We note the date, the platform, visible signals (comments, reactions, views) and whether the vendor replied.
- Where a verified fix exists in the data of Unstuck (a separate product), we link it too.
Issue trackers and support forums lean negative: people post when things break. We say so, and we count praise when we find it.
4. Mark each claim
- ✓ Holds: solid reports back it up.
- ~ Mixed: it shipped, but reports show real gaps or caveats.
- ✗ Fails: reports show it does not work as written.
- ? Unclear: not enough evidence either way. List prices and benchmark scores usually stay here until users report on them.
5. Stamp the verdict
The headline claims hold up in most solid user reports.
It shipped, but some claims fail or carry real caveats for a meaningful share of users.
The headline claim does not hold for most users who report on it, or isn't available as described.
Fewer than about five solid user reports. We wait instead of guessing.
“A solid report” is one we opened ourselves that describes first-hand use of the launch, with enough detail to check. Fewer than about five of those means Too early, however loud the launch was.
6. Re-stamp and correct
We re-check receipts as new reports arrive. Every receipt shows when it was first stamped and last re-stamped. If a verdict or fact changes because we got something wrong, we log it publicly on the corrections page.