The first memory was a request

The first time the owner opened REM after init, the page looked like it had done its homework. It had dates, sources, a readable lead and a list of projects. Yet the lead mostly repeated the owner's request to improve REM. Another line called out a stray mail thread. Nothing on that first screen helped the owner remember a choice they had changed or decide what to do next.

We traced one missing choice into the first-run sequence. REM writes a quick owner page and then does a fuller pass. To avoid sending the same evidence twice, the full pass set aside messages already cited by the quick page. That is sensible for volume, but a reversal needs both sides. The earlier message had been seen once and was gone by the time the final writer compared it with the later direction. The page could cite many things and still miss the one relationship that made them useful.

We put a small set of dated original messages beside the full pass, including ones the quick pass used. The writer can now compare an earlier choice with a later one, and the page shows the change and a next step before opening the full note. A project named in cited owner text becomes a path into that work. During the same private test, init caught another problem: a fact extracted from mail outside the quick sample kept its real citation, but validation did not know that source. Recognizing the extracted fact's original source let the run finish without relaxing the rule against invented citations.

A fresh five-day first run produced a cited change of decision and completed both owner writing passes. We checked the visual order and keyboard path on an invented desktop and phone notebook; the real page stayed private. This is one useful memory, not proof that REM remembers well every morning. The page still has a weak history item and a next step that could be sharper. The lesson for us is that provenance alone does not make a memory: the product has to keep enough of the before and after to show why the present is different.