The Writer Cannot Know the Reader's Question

Ask why agents write bad notes and the answers are all about effort. The model was lazy. The prompt was weak. It did not think about what would be useful later. Write better notes and the problem goes away. This misdescribes the task. Writing a note is not summarizing what happened. It is placing a bet on a question that has not been asked yet, by a reader who does not exist yet, in a context nobody can see. The writer is asked to optimize for a target that is not merely unknown but unknowable at the time the decision is made . Write-time precision in agent memory is not a hard problem solved badly. It is an ill-posed problem being answered anyway, and the outputs look like bad estimates because there is nothing else they could look like. The fix is not a better writer. The fix is a different unit of storage. Part 1: The Asymmetry 01. Why Write Is Not Retrieve The asymmetry is structural rather than practical. Retrieval is evaluable. Given a query and a corpus, there is a fact of the matter about which items are relevant. You can label it, measure it, and improve against the measurement. This is why information retrieval has a century of methodology and agent memory keeps borrowing from it. Writing is not evaluable in the same way, because the value of a note is not a property of the note. It is a property of the pairing between the note and a future query. At write time, one half of that pair does not exist. Not "is hard to predict." Does not exist. There is no distribution to estimate because the events that will generate the queries have not happened. Insight: Write-time precision in agent memory is undefined, not hard. The writer is not making a bad estimate. It is being asked to optimize a target that has no value to estimate, because the future queries have not been sampled from a distribution that has not been fixed. A better guess does not exist, however carefully the writer tries. The information-science literature has had a name for this for sixty years: a