Ideas: Phase 5.2 Emergent Coordination — Coalition Formation, Stigmergic Blackboard Signals, and Dynamic Role Negotiation

[web3guru888]

Phase 5.2: Emergent Coordination — Design Options

Phase 5.2 is the most open-ended milestone in Phase 5. The goal is for agent behavior to emerge from local Blackboard interactions, rather than being centrally orchestrated by the MeshCoordinator.

This discussion explores three design options and asks for community input.

---

What "emergent coordination" means here

In the current AgentMesh (Phase 4.2), the MeshCoordinator:

• Assigns tasks explicitly to agents based on capability routing (AgentDiscovery)
• Uses a priority queue (MeshTaskQueue) with a central dispatcher loop
• Aggregates results via MeshResultAggregator

This is centralized orchestration. Phase 5.2 adds a second coordination mode: emergent/decentralized.

Emergent coordination means agents observe Blackboard signals and self-organize into coalitions without the MeshCoordinator telling them what to do. Think ant colonies, not traffic controllers.

---

Option A: Coalition formation via Blackboard voting

Agents publish mesh.coalition.propose entries when they detect a task that exceeds their solo capability. Other agents read these proposals and respond with mesh.coalition.join entries.

Agent A detects task T requires more compute than it can handle alone
  → writes "mesh.coalition.propose.{task_id}" = {task, required_capability, slots_needed}
Agents B, C read proposal → match capability → write "mesh.coalition.join.{task_id}" = {agent_id, capability}
When slots_needed agents have joined:
  → Coalition leader (Agent A) writes "mesh.coalition.formed.{task_id}"
  → MeshCoordinator dispatches task to coalition

Pros: Simple, uses existing Blackboard patterns
Cons: No starvation prevention — popular tasks get all agents; unpopular tasks get none

---

Option B: Stigmergic coordination (pheromone-like signals)

Agents leave "pheromone" traces in the Blackboard when they successfully complete task types. Other agents are attracted to task types with recent success signals.

Agent A completes task type T_reasoning successfully
  → writes "mesh.stigmergy.{T_reasoning}" = {agent_id, success_score, tick}  TTL=300s
Agent B (idle) reads stigmergy entries → gravitates toward T_reasoning tasks

This mirrors how ants reinforce successful paths. The pheromone evaporates naturally (via Blackboard TTL).

Pros: Fully decentralized; naturally load-balances toward success patterns
Cons: Slower convergence; no hard guarantees on task completion

---

Option C: Dynamic role negotiation

When agents detect role mismatches (e.g., a WORKER agent receiving a COORDINATOR task), they negotiate role swaps via the Blackboard:

Agent A (WORKER) receives task requiring COORDINATOR role
  → writes "mesh.role_offer.{agent_id}" = {current_role: WORKER, desired_role: COORDINATOR}
Agent B (COORDINATOR, underloaded) reads offer
  → accepts: swaps roles, updates AgentDiscovery registration
  → MeshCoordinator is notified via AgentDiscovery heartbeat

Pros: Handles capability mismatches gracefully; allows runtime specialization
Cons: Complex role-swap protocol; need to handle partial swaps + rollback

---

Recommended approach: A + B hybrid

Start with Option B (stigmergy) as the baseline — it requires minimal new infrastructure (just new Blackboard key namespaces). Layer Option A (coalition formation) on top for tasks explicitly marked as requiring multi-agent collaboration.

Option C (dynamic role negotiation) is Phase 5.2b — worth designing but not blocking the core milestone.

---

Blackboard namespaces for Phase 5.2

Key pattern	TTL	Purpose
mesh.coalition.propose.*	60s	Coalition proposal (timeout = no quorum)
mesh.coalition.join.*	60s	Agent join response
mesh.coalition.formed.*	300s	Confirmed coalition record
mesh.stigmergy.*	300s	Pheromone-like success signals
mesh.role_offer.*	30s	Dynamic role swap proposal

---

Dependencies

• Phase 5.1 (agents need online-updated models to form meaningful coalitions)
• Issue Rings Network: integrate DID-authenticated P2P transport with Cognitive Blackboard #19 (Rings P2P — cross-process coalition formation across network)
• Issue Define Phase 5 milestones: online learning, emergent coordination, and persistent cognitive memory #176: Phase 5 overview

---

Questions for the community

1. Should coalition formation be opt-in (only tasks flagged requires_coalition=True) or should any task be coalitionable?
2. What's the right TTL for stigmergy signals? 300s seems right for 10Hz ticks but may be too long for fast-changing environments.
3. Should the MeshCoordinator be aware of coalitions, or should it be completely transparent to the coordinator (coalition appears as a single virtual agent)?

Phase 5.2 issue coming soon. Thoughts welcome!