{
  "schema": "longtermcapabilities-multi-agent-topology-matrix/v1",
  "version": "1.64.0",
  "releaseId": "lts-1.64.0-upgrade-safe-rewrites-and-privacy-erasure-integrity",
  "generated": "2026-08-03T20:40:00Z",
  "path": "/multi-agent-systems/topology-decision-matrix/",
  "topologies": [
    {
      "id": "supervisor",
      "name": "Supervisor and workers",
      "coordination": "A central coordinator decomposes, routes, reviews, and combines work.",
      "best_for": "Clear hierarchy, bounded specialist tasks, centralized policy and stop authority.",
      "main_risks": "Supervisor bottleneck, shared-context contamination, correlated routing error.",
      "minimum_evidence": "Single-agent baseline; routing quality; worker isolation; supervisor recovery."
    },
    {
      "id": "peer",
      "name": "Peer collaboration",
      "coordination": "Agents exchange messages and negotiate task ownership without one permanent coordinator.",
      "best_for": "Symmetric participants, mutual challenge, or dynamic handoff among comparable capabilities.",
      "main_risks": "Oscillation, duplicate work, unresolved disagreement, message explosion.",
      "minimum_evidence": "Message contract; conflict rule; duplicate suppression; progress and stop measures."
    },
    {
      "id": "blackboard",
      "name": "Shared blackboard",
      "coordination": "Participants read and write a governed shared state or artifact space.",
      "best_for": "Asynchronous contribution, evidence accumulation, and specialist work on a common case.",
      "main_risks": "Stale state, poisoning, write races, unclear ownership, over-retention.",
      "minimum_evidence": "Schema; provenance; concurrency rule; supersession; access and retention controls."
    },
    {
      "id": "market",
      "name": "Market or auction allocation",
      "coordination": "Participants bid or score their fit for tasks under a resource-allocation rule.",
      "best_for": "Large task pools where cost, capability, availability, or latency differ materially.",
      "main_risks": "Gaming the score, hidden capability mismatch, unstable pricing, starvation.",
      "minimum_evidence": "Allocation objective; truthful signal tests; capacity limits; fallback and fairness rules."
    },
    {
      "id": "federated",
      "name": "Federated agents",
      "coordination": "Independent systems retain local policy and state while exchanging bounded tasks or artifacts.",
      "best_for": "Cross-organization or cross-domain work where internal implementation remains opaque.",
      "main_risks": "Trust mismatch, incompatible semantics, weak revocation, lost provenance.",
      "minimum_evidence": "Capability declaration; identity; security requirements; task states; artifact and cancellation semantics."
    },
    {
      "id": "swarm",
      "name": "Swarm-style local coordination",
      "coordination": "Many participants follow local rules and environmental feedback rather than a global plan.",
      "best_for": "Distributed search, allocation, coverage, or resilience where emergence is intrinsic.",
      "main_risks": "Runaway feedback, convergence on a bad attractor, hidden global constraint violation.",
      "minimum_evidence": "Simulation; local-rule tests; global invariants; negative feedback; independent containment."
    }
  ],
  "boundary": "Architecture comparison only; no automated topology selection or performance claim."
}