The direction
The work asks how a retrieval foundation might support long-term personal memory without collapsing distinct sources into an untraceable summary. Its intended evidence types include events, temporal context, people, documents and related records.
What is being investigated
The architectural questions are how to represent memory, preserve source and time context, retrieve relevant evidence under uncertainty, and make an answer’s basis understandable to the person using it.
Why the boundary matters
Personal memory systems make privacy, consent, correction and evidence provenance first-order design concerns. This direction is not framed as an always-on assistant or as a substitute for a person’s judgment and relationships.
Next step
Turn the architectural intent into bounded experiments around memory representation, provenance, temporal retrieval and evaluation criteria.