Excerpt from a settled run on a real dev-tool site. Names withheld until the owners agree. Details changed just enough to hide the site; the failure shapes are real.
Alpha. We don't know yet how well this matches real visitors. Each finding is a risk signal to check on your own page — not a measurement of your traffic.
Cited by 3 of 10 sessions. The plan selector read like a confirmation page, so prospects stopped instead of continuing to payment.
Check it yourself: open signup step 2 in a fresh window and watch whether the primary button reads as continue or done.
Cited by 2 of 10 sessions. The footer docs link on /pricing returned 404, so two prospects gave up looking for the API reference.
Check it yourself: click every footer link on /pricing in a fresh session.
Cited by 2 of 10 sessions. After Google OAuth the prospect landed back on the marketing page with no workspace, and walked out.
Check it yourself: sign up with a throwaway Google account and note where you land.
Ten prospects — core customers, adjacent roles, people outside the target — each drove a real browser one viewport at a time until they finished, walked out, or ran out of patience. Cheap models voted; a filter kept only what more than one session cited; a verifier reviewed; a stronger model re-walked the hardest prospect. Every step logged, every run recorded.
The videos, the session logs, a full REPORT.md, a one-page AGGREGATE.md, and FIXES.md with check-it-yourself steps. Requested runs land the same report in your mail.