Competence

What a given writer is authoritative about — trust attached to the claim rather than to the claimant.

Why it matters in a memory layer

A retrieval source is reputable across its output; a memory writer is not. A calendar agent is excellent about calendars and, about a user's diet, no better informed than a stranger. Provenance.authority is a constant on the row, so it cannot express that.

Competence is a short table of writer → slots, and it needs three outcomes rather than two. Within the domain, keep the authority. Naming a slot the writer does not own, discount it. Naming no modelled slot at all, keep the authority and flag it — because that claim is outside the vocabulary, not outside the writer's domain, and discounting it would record a gap in your slot table as a fact about the writer's reliability.

On this course's corpus that distinction is the whole result: the calendar agent's entire output is unnameable, so a two-state policy drops the reliable writer to 0.3 and leaves the unreliable one exactly where it was.

Connections

Taught in: Provenance and Trust

Used in: Cross-Agent Write Conflicts · Memory Access Control · Memory Attacks

Do not confuse with: Provenance