What's this about?
Existing agent-memory systems store what was said. Σ-Mem stores reliability: an online memory for multi-agent systems that learns which agents can be trusted, for what, under which conditions — so an orchestrator can route work accordingly.
What exactly is "multi-agent reliability"?
When a central model delegates to many sub-agents, some are flaky at certain task types, and the orchestrator often can't verify each result directly. A reliability memory is a track record — like a manager remembering which contractor botched which kind of job — updated continuously as evidence accumulates (12 upvotes).
Why it matters
Multi-agent systems are moving to production while trust between agents is assumed, not measured; making it a first-class, learned quantity is an obvious missing piece.