Expand description
Inspection — what memory actually holds, for a person rather than a program.
Nobody trusts a memory they cannot read. graph status counts rows and
graph search answers questions, but neither answers the question a human
actually has after a week of automatic ingestion: what do you think you
know about me, and how sure are you?
Two shapes answer it.
MemoryOverview is the unprompted one: the strongest things the graph
holds, grouped by entity type, next to an honest account of how well its
relationships are believed — including the coherence tally, which is the
one number that says “some of this confidence is me agreeing with myself”.
TopicReport is the prompted one. It runs the ordinary hybrid query — no
second retrieval path, no second set of scoring weights — and then attaches
the evidence behind each hit, which retrieval itself has no reason to carry.
Structs§
- Confidence
Summary - How well the live relationships are believed.
- Known
Entity - One entity as an overview lists it.
- Memory
Overview - What the graph holds, without being asked about anything in particular.
- Topic
Entity - One retrieved entity, with the claims it takes part in.
- Topic
Report - What the graph holds about one subject.
- Type
Group - One entity type, and the strongest entities in it.
Constants§
- DOUBTFUL_
CONFIDENCE - Below this posterior mean, a relationship is barely believed at all.
- STRONG_
CONFIDENCE - At or above this posterior mean, a relationship is held firmly.