It told me it had read 46 things

I opened REM's invented test notebook with a question about Mara. The first thing it told me was that the night pass had read 46 items and rewritten four pages. The answer about Mara was there, but I had to go looking for it. The reader was proud of its workload; I wanted to know what I should carry into the day.

That seemed like a layout problem until we tried moving a sentence from Mara's page into the first card. The sentence said she had moved from feedback calls to asking about expansion. It looked like the kind of insight a memory product ought to greet you with. Then we checked what the run log actually recorded: Mara's page had been updated. It did not say when that sentence first became true, or whether REM had learned it last night. Calling it a new discovery would make the screen more exciting and less trustworthy.

The card now makes the smaller, supported claim. It says the page was updated, shows what that page currently says, and gives a path back to its sources. The night's counts are still available, but they no longer get the first word. Below the card, an older relationship appears as a question. For a moment the reader waits, so the person can try to remember before seeing what REM wrote.

The picture of a night pass is useful. The point of that pass is the morning after. We can make the morning feel more alive only as far as the evidence allows; the next step is to record what a claim was before and after the pass, so REM can finally say exactly what changed.