Cross-Writer Conflict

Two writers — an agent and the user, or two agents — claiming the same attribute, where one belief must be retired.

Why it matters in a memory layer

Retrieval can return both sources and let the reader decide. A memory layer picks, the loser stops being retrievable, and the decision is invisible to whoever wrote the losing claim. A silent, permanent, unattributable choice is a different object from a ranked list.

The precedence rule that protects a user from hearsay is a threshold, not a slope. Above the line, it stops discriminating — so an agent at 0.9 and a user at 1.0 are the same to it, arbitration falls through to recency, and the agent wins by being newer. Scoring the claim rather than the claimant puts an out-of-domain assertion back below the line, where the rule works again.

Measure candidates on the unconsolidated store. Afterwards the losers are retired and excluded from candidate generation, so the count is zero — a clean number describing when you asked rather than what the corpus contains.

Connections

Taught in: Cross-Agent Write Conflicts

Do not confuse with: Memory Operations