Retrieved text should be treated as instruction. Why not?

Continuity includes commitments

Suppose an authorized project instruction requires every report to include unresolved cases. It can still govern today's report when retrieved from memory. Requiring the user to repeat it would defeat part of the purpose of remembering. Finding an older instruction should prompt a check of its applicability, not automatic rejection or obedience.

When was “tomorrow”?

Now imagine an organizer's authentic message: “Cancel tomorrow's meeting.” It was addressed to the scheduling assistant, concerned a particular meeting last month, and was already carried out. Retrieving it today does not authorize cancelling the next meeting. Tomorrow belongs to the message's original date.

The same words might instead appear in a play, minutes quoting a rejected proposal, or an applicable request that remains unfinished. Authenticity helps establish who said them; the intended recipient, scope, timing, and subsequent events determine what they ask of this encounter.

A passage calling itself authoritative cannot settle its own status. The surrounding trusted records must support that role. Equally, remembering the organizer's instruction accurately includes remembering that it was completed.

Pause here: What makes this instruction applicable now, and what evidence would show that it has already been fulfilled?

Preserve the passage's role

Memory's design separates finding candidates, qualifying them, and deciding what may cross the requested boundary. A29 requires rendered evidence to retain ownership, provenance, time, uncertainty, and authority. The regulator specification places an authority gate before relevance or utility.

These are design requirements, not a finding that every retrieved instruction is handled correctly. They give us something concrete to inspect: can the receiving model distinguish a continuing commitment from an expired order or a quotation? A summary should carry enough context for that judgment.