docs: define peer coordination backoff

Signed-off-by: huangruiteng <14976749+huangruiteng@users.noreply.github.com>

Author: huangruiteng

docs/integrations/codex-subagent-orchestration.md

@@ -304,9 +304,11 @@ candidates. A dormant or non-resumable lane is projected under
 That blocked diagnostic does not replace the coordinator's own runnable lane or
 re-arm an activation obligation on every heartbeat. If the coordinator also
 has no in-scope runnable fallback, the final interaction mode is
-`peer_coordination_blocked`: schedulers return the bundle to its owner and stop
-the recurring heartbeat until peer capability/readiness, coordinator
-configuration, or the coordinator's own work frontier materially changes.
+`peer_coordination_blocked`: schedulers keep a no-spend observer alive with a
+10/20/30/60 minute stateful backoff. Peer capability/readiness, coordinator
+configuration, reassignment, or the coordinator's own work frontier changes
+the reset identity and restores the initial cadence; this recoverable state
+does not pause or delete the recurring heartbeat.
 
 Disable registered-peer coordination without changing peer registration or
 child-worker policy:

docs/product/core-control-plane/state-machine.md

@@ -340,7 +340,7 @@ stateDiagram-v2
 | --- | --- | --- | --- |
 | `run_now` | `active_work` | 3 / 10 minutes | Work or repair must be attempted. |
 | `backoff_waiting_for_user` | `human_gate` | 30 / 120 minutes | Concrete user/controller action is next. |
-| `backoff_until_reassigned` | `agent_scope_wait` | 10 / 60 minutes, progression 10/20/30/60 | Handoff owner or reassignment may unblock this agent. |
+| `backoff_until_reassigned` | `agent_scope_wait` or `peer_coordination_wait` | 10 / 60 minutes, progression 10/20/30/60 | Handoff owner, peer readiness, coordinator configuration, reassignment, or new local work may unblock this agent. |
 | `backoff_until_material_transition` | `monitor_wait` | 15 / 60 minutes | Monitor-only liveness without compute spend. |
 | `backoff_until_fresh_evidence` | `unchanged_noop` | 60 / 240 minutes | Wait for fresh mapped or post-handoff evidence. |
 | `backoff_until_state_change` | `quiet_wait` | 30 / 120 minutes | No specific user/monitor path is projected. |

docs/quota-allocation.md

@@ -725,11 +725,13 @@ gap, autonomy blocker, or replan obligation fail closed. This keeps recurring
 controllers alive during ordinary waits while honoring an explicit completed
 goal shutdown without another quota-spending turn.
 
-An explicit `peer_coordination_blocked` decision is also a typed host stop, not
-an ordinary quiet wait. Its `automation_liveness` and `scheduler_hint` both
-require the current recurring heartbeat to stop without spending quota. The
-host resumes only after peer activation capability, peer runtime readiness,
-coordinator configuration, or the local work frontier materially changes.
+An explicit `peer_coordination_blocked` decision is a recoverable typed wait,
+not a terminal host stop. It keeps the recurring heartbeat alive without
+spending quota and uses the existing TypeScript-owned stateful backoff
+transition with a 10/20/30/60 minute progression. Peer activation capability,
+peer runtime readiness, coordinator configuration, or newly projected local
+work changes the reset identity and restores the initial cadence. Goal stopped,
+quota paused, and validated terminal no-follow-up remain the stop cases.
 
 An individual registered peer can instead be put in `monitor_only` work mode: