The generous shelf
A full archive can rescue details that a short summary lost. The exact day of an agreement, a change of mind, a joke that mattered to two people—there are good reasons to keep such things. An empty shelf is no cure for a crowded one.
Let the details compete
Imagine the system preparing for a visitor who once asked it not to use a certain name. It has a thousand pages of shared conversation, ten cheerful summaries about their rapport, and one brief correction about the name. Give it the thousand pages without choosing or weighing them, and that correction may disappear among the agreeable stories.
More history was available; less of the necessary history governed the reply. A short, well-sourced selection containing the correction could do better. Yet selection can fail too. If it keeps only flattering evidence, it may make the relationship appear smoother than it was.
Pause here: Which would you rather have in the next exchange: every pleasant line, or the one uncomfortable line that changes how you should be addressed?
Keep widely; bring forward carefully
The PHS memory-selection math drafts explore limited prompt space: relevance, coverage, provenance, redundancy, and diversity all matter. This is a proposed way to reason about selection, not proof that a live selector always gets it right. The archive and the material carried into a particular response are different things.
My answer: More stored history can make better memory possible. Better memory depends on finding the relevant evidence, including corrections and contradictions, and using it with care at the right time.