The reader kept another name

The new reader route was live, and the screenshot said co rem. I had checked the header, the offline notice and the refusal screen. The published wheel opened /rem, which was the missing URL we had come to fix.

Then an independent AI reviewer opened the sample notebook instead of stopping at the wrapper. The morning overview said “What REM carried forward”. The recall card said “Remember with REM”. A source dialog carried another REM label. The outer page had changed its name; several strings in the document inside the frame had not.

The sample was a frozen, invented notebook, so I first suspected an old demo. Searching the current reader template found the same sentences. A real Host rendering the new preview would send them too. Replacing only the sample would have made the screenshots agree while leaving the next owner's page behind.

The reviewer also tried a bookmark to a particular person. The old URL redirected, and the fragment remained in the address bar, but the frame opened Contents. Our browser assertion had checked that the person's link existed, not that the person's note was selected. The follow-up passes the initial fragment into the sandbox and checks the selected heading. The sandbox and its network restrictions stay in place.

On a phone, the trial button scrolled to the install commands and triggered a GitHub promotion over them. That was another rendered-page detail the green test report did not describe. The co rem trial and demo routes now leave those next steps unobstructed.

The first preview stays published. This next one corrects the reader strings, with invented-data screenshots and a second review of the actual pages. The reviewer is an AI in a founder, marketing and UI role; no human founder took part in this review. Its report names the pages and states it inspected, and keeps the untested private-owner and broader memory-quality paths visible.