fix(k1): settle same-process standby checkpoint

docs/runbooks/K1_OPERATOR_FLOW_INCREMENTAL_ACCEPTANCE.md

@@ -0,0 +1,79 @@
+# K1 operator flow: incremental acceptance
+
+Status: working acceptance ledger, 2026-08-14.
+
+This is the short regression anchor for changes to the existing K1 operator
+flow. The authoritative connection and safety model remains
+[`../20_K1_CONNECTION_SUPERVISION_CANON.md`](../20_K1_CONNECTION_SUPERVISION_CANON.md).
+Every K1 fix must name one row below, add a reducer test, and preserve all
+previously accepted rows. No change in this ledger authorizes a new K1 command.
+
+## Non-negotiable device boundary
+
+- START and STOP keep the captured vendor payload, topic, QoS and one-shot
+  dispatch contract. No automatic replay, substitute command or inferred ACK.
+- Reconnect and Verify are read-only. Scenario reset changes local Mission Core
+  state only; it sends no BLE, Wi-Fi, START or STOP operation.
+- `READY` / `SCAN_OVER` from a fresh, identity-bound, non-retained status is
+  physical standby truth. Local recovery state must converge to that fact
+  before a new START can be admitted.
+- An unknown command outcome is never presented as success. A later exact
+  read-only reconciliation may settle it, but may not rewrite its audit history.
+
+## Operator-state contract
+
+| State | Stable operator surface | Required feedback | Accepted exit | Forbidden |
+| --- | --- | --- | --- | --- |
+| Saved connection needs checking | Saved K1 card plus `Переподключиться` and `Подключить новый K1` | None before action | One explicit action | Hidden Scan, Verify or reconnect |
+| Reconnect in progress | The same saved K1 card and same button geometry | Disabled action with canonical `ActivityIndicator`; neutral copy `Переподключение` | Connected or an actionable terminal result | Replacing the whole panel, blank wait state, first-person copy |
+| Bluetooth search | The same connection step | Visible activity and search phase | Results or terminal no-result state | Empty disabled panel with no progress |
+| Device selected | Selected-device card and network fields | None while editing | One explicit Apply or choose another device | Background provisioning or hidden second discovery |
+| Apply in progress | The same selected-device/network form | Disabled Apply with canonical `ActivityIndicator` and named phase | `network_applied`, or a truthful terminal failure | Container-size jumps, screen substitution, automatic Apply replay |
+| Network applied, control checking | Connection card remains visible | Passive `Подтверждение управления` activity | Control ready or explicit recovery choice | Calling the provisioning result a network failure |
+| Ready to start | Project form | No activity | One explicit START | Green readiness without exact control authority |
+| START in progress | The same project form | Disabled START with named phase | Active acquisition or truthful terminal state | Second START, unexplained switch to viewer |
+| Active acquisition | Spatial viewer and one stable STOP action | Stream/visualizer status separately | One explicit STOP | STOP enabled/disabled oscillation before click |
+| STOP in progress | Same viewer and same STOP position | Disabled STOP with named phase | Physical standby or explicit outcome-unknown recovery | Second STOP or button disappearance |
+| Physical standby | Clean local idle/new-session state | None | Reconnect or new acquisition | Active recovery checkpoint blocking a proven `READY` |
+
+## Presentation invariants
+
+- Any operator-owned operation lasting longer than one rendered frame has an
+  `ActivityIndicator`, visible neutral process noun, `aria-busy`, and a stable
+  action/container position.
+- Pending copy describes the process, not the application or operator:
+  `Переподключение`, `Поиск устройства`, `Подтверждение управления`,
+  `Подготовка приёма`, `Остановка записи`.
+- One action never flashes an unrelated complete screen between its pending and
+  terminal states.
+- Wi-Fi mutation, control bootstrap, physical command and visualization are
+  separate facts. Failure of a later fact must not relabel an earlier success.
+- If decoded PCL frames exist but the browser has not admitted the exact Rerun
+  store, the viewer remains `Подключение визуального источника` and must become
+  an actionable visualization error at its deadline. It must not show
+  `Визуализатор готов` over an empty scene.
+
+## Increment gate
+
+For each patch, record four facts in the handoff:
+
+1. **Fix** — one named broken transition.
+2. **Reducer** — the smallest offline sequence that failed before the patch.
+3. **Visual acceptance** — exactly what the operator should see.
+4. **Not accepted yet** — adjacent known violations that remain out of scope.
+
+Current violations observed on 2026-08-14:
+
+- `K1-P0-CHECKPOINT`: fresh standby reconciliation left an ACTIVE recovery
+  checkpoint and blocked the next connection. Code fix and same-process/restart
+  reducers are green. Live startup convergence accepted: checkpoint revision 16
+  is `ceased`, physical head remains `physical-standby-observed`, and the public
+  policy again admits Scan plus read-only configured-device observation.
+- `K1-P0-RERUN-ADMISSION`: backend decoded and published PCL, but the browser
+  never admitted the active Rerun store; camera/counters were visible over an
+  empty point scene. Not fixed in the checkpoint increment.
+- `K1-P1-STOP-STABILITY`: STOP authority visibly oscillated before the operator
+  clicked. Not fixed in the checkpoint increment.
+- `K1-P1-PENDING-FEEDBACK`: reconnect/search/select/provision transitions replace
+  panels or show disabled controls without an activity indicator. Not fixed in
+  the checkpoint increment.

src/k1link/device_plugins/xgrids_k1/active_acquisition_recovery_checkpoint.py

@@ -2627,6 +2627,7 @@ def _require_gap_cessation_evidence_session(
     failed_binding: ActiveAcquisitionRecoveryTransportBinding,
     failed_evidence_session_id: str,
     status_proof: ActiveAcquisitionRecoveryStatusProof,
+    allow_same_runtime_successor_evidence: bool = False,
 ) -> None:
     crosses_runtime_instance = (
         status_proof.binding.runtime_instance_id
@@ -2640,7 +2641,11 @@ def _require_gap_cessation_evidence_session(
             "terminal recovery across a runtime instance requires a successor "
             "evidence session"
         )
-    if not crosses_runtime_instance and not preserves_evidence_session:
+    if (
+        not crosses_runtime_instance
+        and not preserves_evidence_session
+        and not allow_same_runtime_successor_evidence
+    ):
         raise ActiveAcquisitionRecoveryCheckpointTransitionError(
             "same-runtime terminal recovery must preserve the exact evidence session"
         )
@@ -3316,16 +3321,26 @@ def _require_active_reconciled_standby_shape(
         raise ActiveAcquisitionRecoveryCheckpointTransitionError(
             "restart standby physical proof must advance the active lineage"
         )
+    recovery_binding = cessation_status_proof.binding
+    same_runtime = (
+        recovery_binding.runtime_instance_id == failed_binding.runtime_instance_id
+    )
     if (
-        cessation_status_proof.binding.runtime_instance_id
-        == failed_binding.runtime_instance_id
-        or cessation_status_proof.binding.control_session_id
-        == failed_binding.control_session_id
+        recovery_binding.control_session_id == failed_binding.control_session_id
         or cessation_status_proof.evidence_session_id
         == failed_evidence_session_id
+        or (
+            same_runtime
+            and (
+                recovery_binding.host_path_epoch == failed_binding.host_path_epoch
+                or recovery_binding.producer_generation
+                == failed_binding.producer_generation
+            )
+        )
     ):
         raise ActiveAcquisitionRecoveryCheckpointTransitionError(
-            "restart standby settlement requires new runtime, control and evidence sessions"
+            "standby settlement requires a new control/evidence boundary; "
+            "same-runtime recovery also requires a new host path and producer generation"
         )
 
     gap_started = _validated_timestamp(
@@ -4982,6 +4997,9 @@ def _validate_checkpoint_semantics(
                     prior_active_status.evidence_session_id
                 ),
                 status_proof=status,
+                allow_same_runtime_successor_evidence=(
+                    receipts[-1].kind == "cease-active-reconciled-standby"
+                ),
             )
         if receipts[-1].kind == "cease-active-reconciled-standby":
             gap_started_at_utc = checkpoint.last_gap_started_at_utc

tests/test_xgrids_active_acquisition_checkpoint_integration.py

@@ -455,11 +455,17 @@ def test_repeated_reset_reopen_fresh_ready_ceases_old_active_checkpoint(
     assert service._active_acquisition_checkpoint_trust == "trusted"  # noqa: SLF001
 
 
-def test_restart_ready_after_dispatched_stop_ceases_active_checkpoint(
+@pytest.mark.parametrize(
+    "same_process",
+    (False, True),
+    ids=("process-restart", "same-process-new-control-epoch"),
+)
+def test_ready_after_dispatched_stop_ceases_active_checkpoint(
     tmp_path: Path,
     monkeypatch: pytest.MonkeyPatch,
+    same_process: bool,
 ) -> None:
-    """Fresh READY settles a dispatched STOP whose outcome crossed restart."""
+    """Fresh READY settles a dispatched STOP across either recovery boundary."""
 
     service = _service(tmp_path, monkeypatch)
     original = _connection(
@@ -509,9 +515,10 @@ def test_restart_ready_after_dispatched_stop_ceases_active_checkpoint(
             observed_at_utc="2026-08-13T12:01:01.000Z",
         ),
     )
-    service._snapshot_runtime_id = (  # noqa: SLF001
-        "snapshot-runtime-checkpoint-dispatched-stop-successor"
-    )
+    if not same_process:
+        service._snapshot_runtime_id = (  # noqa: SLF001
+            "snapshot-runtime-checkpoint-dispatched-stop-successor"
+        )
     restarted = _connection(
         control_session_id="checkpoint-dispatched-stop-restarted-control",
         host_path_epoch=2,
@@ -561,6 +568,7 @@ def test_restart_ready_after_dispatched_stop_ceases_active_checkpoint(
         lambda: None,
     )
 
+    settle_checkpoint = service._settle_restart_checkpoint_after_verified_standby  # noqa: SLF001
     monkeypatch.setattr(
         service,
         "_settle_restart_checkpoint_after_verified_standby",
@@ -583,10 +591,23 @@ def test_restart_ready_after_dispatched_stop_ceases_active_checkpoint(
     assert checkpoint is not None
     assert checkpoint.state == "active"
 
-    # Simulate the exact live process boundary: the physical ledger fsync is
-    # durable, while checkpoint cessation did not commit in the prior process.
-    restarted_service = _service(tmp_path, monkeypatch)
-    assert restarted_service._active_acquisition_checkpoint_reason is None  # noqa: SLF001
+    # The physical ledger fsync is durable while checkpoint cessation did not
+    # commit. Recovery may happen after a process restart or, as observed in
+    # production, later in the same process under a fresh control/path epoch.
+    if same_process:
+        monkeypatch.setattr(
+            service,
+            "_settle_restart_checkpoint_after_verified_standby",
+            settle_checkpoint,
+        )
+        token = service._validate_active_acquisition_checkpoint_lineage()  # noqa: SLF001
+        assert token is not None
+        assert service._settle_durable_prepared_stop_standby_checkpoint(  # noqa: SLF001
+            token
+        ), service._active_acquisition_checkpoint_reason  # noqa: SLF001
+    else:
+        restarted_service = _service(tmp_path, monkeypatch)
+        assert restarted_service._active_acquisition_checkpoint_reason is None  # noqa: SLF001
     checkpoint = ActiveAcquisitionRecoveryCheckpointStore(tmp_path).snapshot().checkpoint
     assert checkpoint is not None
     assert checkpoint.state == "ceased"

tests/test_xgrids_active_acquisition_recovery_checkpoint.py

@@ -1312,6 +1312,7 @@ def test_cease_active_reconciled_standby_rejects_inexact_restart_proofs(
         replace(
             status.binding,
             runtime_instance_id=binding.runtime_instance_id,
+            host_path_epoch=binding.host_path_epoch,
         ),
         replace(
             status.binding,