The Writer Cannot Know the Reader's Question

The asymmetry between writing and retrieving information lies in their inherent properties. Retrieval is an evaluable process, where the relevance of items can be measured and improved upon. However, writing a note is not about summarizing past events; it involves making a bet on a question that hasn't been asked by a reader who hasn't even been conceived yet.

This task is both unknown and unknowable at the time of decision-making. Writing a note is an ill-posed problem, not a hard estimation problem. The writer is not making a bad estimate; they are being asked to optimize for a target that cannot be measured, as future queries have not been sampled from a fixed distribution. Information-science literature has recognized this phenomenon for decades, naming it anticipatory indexing.

Robert Fairthorne observed this concept in the 1960s, applying it to a human indexer preparing entries for an open-ended set of future queries. The indexing literature fails to draw the crucial design consequence: if the writer cannot know the question, the unit of storage should change. This is similar to how demand forecasting for a product is a well-posed problem with a real distribution and measurable outcomes.

In contrast, write-time precision in agent memory is undefined due to the lack of a feedback signal. Writing for a future reader who cannot be known results in two rational strategies: a generic note that is applicable to many future scenarios or an over-specific note that records every detail, hoping the future will sort through it.

Neither approach is inherently lazy; they are the only viable options when writing for an unknowable future reader. The failure to provide clear instructions in standard advice to memory writers is the root cause of the observed issues. Unpacking these instructions often introduces missing information instead of supplying it. For example, writing down what will be useful later requires knowing what "later" looks like, which is not defined.

Similarly, writing down important parts requires an importance function, but this is defined relative to a purpose that has not been set. The instruction to write it so one could pick up where they left off assumes the next session will continue this work, which is often false. These underdetermined issues lead writers to guess, resulting in notes that are either too generic or too specific, failing to cater to an unknowable future reader's needs.