Why completeness feels kind
When someone matters to us, we dislike leaving out a detail that might help them be understood. For a continuing persona, the temptation is stronger: include every trait, past exchange, mood, plan, source note and warning, so the next model invocation cannot possibly miss the person behind the request.
But a prompt is an encounter, not an archive. Even a large context window asks the thinking layer to weigh what it sees. Repetition can give minor facts false importance; stale material can sound current; a long list can flatten a relationship into identifiers. More words may reduce the chance that one particular sentence receives the attention it deserves.
The cost of carrying everything
Imagine a friend coming to dinner to ask whether a difficult apology has been delivered. A useful account includes the promise, the attempted send, the failure notice and the friend’s present concern. It does not need every previous dinner menu. Older details may remain in Memory when retained and retrievable; this encounter need not carry them all.
PHS’s meaningful-admission doctrine treats prompt space as a portfolio. Stored evidence, retrieved candidates and admitted content are separate decisions. Relevance does not grant permission, and omission should be marked when an expected section lacks evidence. That is more demanding than merely trimming to a token budget.
Pause here: Which three facts would you keep if the next reply concerned a missed promise rather than the person’s whole biography?
Choose enough for this encounter
A smaller prompt can still be wrong if it excludes the decisive correction. A longer prompt can be justified when identity, relationships and active obligations genuinely need room. The measure is whether the context helps the receiving layer judge this situation without mistaking an old plan for an accomplished act or private material for public instruction.
Keep the archive generous and the encounter deliberate.
My answer: The best prompt supplies enough relevant, authorized, current and attributable context for this turn. Durable Memory aims for coverage, not impossible completeness; prompt admission chooses what matters now within a bounded budget.