The first run should remember something

The early REM first run mapped a mailbox and local coding sessions into pages, then left most pages as empty frames. The map was useful infrastructure, but it made the person wait before they could see why a memory notebook mattered. Investigating only the owner's page helped; it still left the relationships and projects that give that page meaning untouched.

We tried widening the first run to the whole notebook. It showed that pages could be investigated in parallel, but the oldest, quietest entries could use the model before a reader saw anyone they had spoken to this week. The better question was which pages would make the notebook recognizable on opening day. We selected the owner, recent important people, active projects, and the organizations connected to them.

The first private seven-day trial selected 27 pages. Twenty-six passed review; one person page came back with ten History milestones, beyond the eight the reader accepts. That was an expensive failure to discover after the model had finished. We added a bounded repair turn for that specific error. It folded the older history and the page passed, so the trial ended with 27 cited pages.

We then installed the proposed preview wheel and ran init again against a fresh private notebook. It selected and accepted all 27 pages without a refusal. The trial is evidence that this first run can produce pages instead of empty frames; it cannot prove that every claim is right. The reader therefore keeps citations and unresolved questions in view.

The recent cohort is a starting point, not a cutoff on a relationship's past: one selected person may have two years of evidence to read. REM announces a time and token estimate before it starts. Around 20% of a weekly runner allowance guides that estimate, while selected work can continue past it. Old and quiet pages remain in the map for later investigation. We will revisit the cohort when real first runs show an important person was missed, or when a completed first view still fails to reveal a useful connection.