memfleet-conflict-resolution

MemFleet — Conflict Resolution

Conflict classes

Class	Meaning	Resolution
A	Additive — new symbols, no removals	No action. Never blocks.
B	Modification — body / type / rename	LWW by reference_time. Loser replans with replan_hint.
C	Breaking — removes live symbol or changes signature with active callers	Blocked. Human review.

Class B — you lost

Your edit's reference_time was older than the winning edit's. Read replan_hint:

{
  "won": false,
  "winner_episode_id": "01H…",
  "winner_intent_kind": {"Refactor": {"pattern": "RenameSymbol"}},
  "replan_hint": {
    "kind": "ApplyToRenamedTarget",
    "new_name": "UserRoleV2"
  }
}

Map your edit onto the new target (e.g. apply the NullHandling fix to UserRoleV2 instead of UserRole) and record_episode again with the new touched_nodes.

Intent-kind auto-merge

Refactor(RenameSymbol) + BugFix(NullHandling) on the same symbol are orthogonal — the broker can apply both. This is Phase-2 behavior; Phase-1 returns replan_hint and lets you do it.

Class C — blocked

Your intent removes a live symbol OR changes a signature that has active callers. publish_intent returns:

{
  "active_conflicts": [{
    "cause": "SignatureChangeWithActiveCallers",
    "callers": ["UserRoleRepo.find", "UserSessionService.check"],
    "blocker_agent": "agent-alice"
  }]
}

Do NOT attempt to edit. Options:

1. Narrow the scope — edit only internals, keep signature stable
2. Coordinate out-of-band with blocker_agent
3. Call resolve_conflict (Phase-2) with prose rationale — the ONLY place prose is required

Retry pattern

r = publish_intent(...)
while r.active_conflicts ≠ ∅:
   wait(exp_backoff)
   ns = get_node_state(symbol)
   if ns.active_intents == [] → retry publish_intent
   if elapsed > deadline → escalate