LineageCycle

A group that lies on a topology cycle — walking its live inbound edges root-ward returns to the group itself instead of reaching a root. Like DualActiveInbound this is a fork the merge deliberately retains rather than resolving by arrival order (§5.2): two replicas each attached an inbound edge, and together the records close a loop. The whole cycle is quarantined — EntitlementLedger.holdings is zero at every group on it and at every group below it — so this report is what makes that quarantine visible (§10.11: quarantine ⟺ explicit report).

Reported once per cycle member, not per descendant. A group merely below a cycle is quarantined too, but it is not itself a loop member and would only flood the report with the whole subtree; only groups the walk re-enters at their own starting point are listed.

Not a delivery artifact. Records are grow-only, so a cycle observed on any state is a cycle in the merged topology — a partially-delivered replica can only ever see fewer edges, never a loop that is not really there. It can nonetheless be transient in the honest control plane: an inverting reparent (attaching G under H while H is still live under G) closes a real loop for the window between the new edge activating and the old one retiring. That window is a real quarantine, so reporting it is the point; it clears when one of the loop's edges reaches Lifecycle.RETIRED.

Resolution is a control-plane concern, exactly as for DualActiveInbound: a quarantined generation has zero holdings and so cannot be drained, and the loop is broken by retiring-and-abandoning one of its edges.