diff --git a/.console/log.md b/.console/log.md
index de09164c..b91aee9f 100644
--- a/.console/log.md
+++ b/.console/log.md
@@ -1,3 +1,24 @@
+## 2026-07-14 — feat(reviewer): budget/cooldown-aware review — defer, don't burn (audit D1 pt1)
+
+The reviewer is part of the fleet but was claude-ONLY and consulted NO budget:
+it burned claude reviewing PRs even when over the 25% reserve (observed live
+2026-07-14 — reviewing my own PRs during a budget crunch pushed the account to
+the hard cap). Now `_process_self_review` calls `_select_review_backend`
+(reuses the controller's `select_worker_backend` ladder) BEFORE the direct
+`claude -p` verdict call: if claude is cooled or over the budget_reserve
+(`selected_backend != "claude_code"`), it DEFERS the sweep — no claude spawn, no
+budget charge, no needs-human escalation — and retries when the window drains
+(~5h). Fail-open: any selection/store error → proceed on claude (today's
+behavior); `dynamic_worker_backend_selection=False` → operator opt-out honored.
+Verdict parsing untouched (already backend-agnostic, file-based verdict.json).
+3 new tests + 150 existing reviewer tests green; ruff+ty clean.
+
+This is D1 PART 1 (stop the over-budget burn, park smart). PART 2 = actually
+review on CODEX when claude is cooled (needs live validation that codex writes a
+schema-conformant verdict.json in the empty-dir/`-p` contract — the one unknown
+from the scoping pass; until then non-claude selection = defer). See
+audit-remediation-plan memory. Next: D2 council.
+
 ## 2026-07-14 — feat(budget): operator budget signal `operations-center.sh budget` (audit D1)
 
 Voluntary operator readout (D1 part 3). A human session can't be hard-gated, so
diff --git a/src/operations_center/entrypoints/pr_review_watcher/main.py b/src/operations_center/entrypoints/pr_review_watcher/main.py
index 8214347d..b52eaad4 100644
--- a/src/operations_center/entrypoints/pr_review_watcher/main.py
+++ b/src/operations_center/entrypoints/pr_review_watcher/main.py
@@ -522,6 +522,33 @@ def _record_close_receipt(
 # ── pr review pipeline ────────────────────────────────────────────────────────
 
 
+def _select_review_backend(settings, *, usage_store=None, now=None):
+    """Pick the review backend via the shared fleet ladder (audit D1).
+
+    The reviewer is part of the fleet, so it must respect the same claude→codex
+    cooldown/budget ladder the controller uses instead of burning claude
+    unconditionally. Returns the ``WorkerBackendSelection`` (``selected_backend``
+    is ``claude_code`` when claude is runnable, another backend / ``None`` when
+    claude is cooled or over the 25% budget reserve), or ``None`` if selection
+    couldn't run — in which case the caller proceeds on claude (today's
+    behavior). The merge-gatekeeper must never crash here.
+    """
+    try:
+        from operations_center.backends.worker_backend_selector import select_worker_backend
+        from operations_center.execution.usage_store import UsageStore
+
+        team = getattr(settings, "team_executor", None)
+        return select_worker_backend(
+            preferred_backend="claude_code",
+            usage_store=usage_store or UsageStore(),
+            dynamic_enabled=bool(getattr(team, "dynamic_worker_backend_selection", True)),
+            now=now,
+        )
+    except Exception as exc:  # noqa: BLE001 — never block the gate on a store read
+        logger.warning("pr_review_watcher: backend selection failed, proceeding on claude: %s", exc)
+        return None
+
+
 def _run_direct_review(
     oc_root: Path,
     goal_text: str,
@@ -3221,6 +3248,26 @@ def _phase1(
     state.setdefault("backend_error_passes", 0)
     state.setdefault("no_verdict_escalation_count", 0)
 
+    # D1: consult the shared fleet ladder BEFORE spawning a claude review. If
+    # claude is on cooldown or over the 25% budget reserve, don't burn it —
+    # defer this sweep and retry when it returns (degrade-never-halt applied to
+    # review; the budget window drains within ~5h). No claude spawn, no budget
+    # charge, no needs-human escalation for a transient wait. (Codex-fallback
+    # for the review itself is a validated follow-up; until then any non-claude
+    # selection means claude — the only review backend — is unavailable.)
+    _selection = _select_review_backend(settings)
+    if _selection is not None and _selection.selected_backend != "claude_code":
+        _resets = [r for r in _selection.cooldowns.values() if r is not None]
+        _reset_at = max(_resets) if _resets else None
+        logger.info(
+            "pr_review_watcher: PR #%d review DEFERRED — claude unavailable "
+            "(selected=%s, reset≈%s); not burning budget, will retry when it returns.",
+            pr_number,
+            _selection.selected_backend,
+            _reset_at.isoformat() if _reset_at else "unknown",
+        )
+        return
+
     try:
         verdict = _run_direct_review(oc_root, goal_text, state_key)
     except OCSourceTreeUncleanError as exc:
diff --git a/tests/test_pr_review_watcher.py b/tests/test_pr_review_watcher.py
index 782028d0..1f95e9d5 100644
--- a/tests/test_pr_review_watcher.py
+++ b/tests/test_pr_review_watcher.py
@@ -3468,3 +3468,66 @@ def test_branch_protection_gate_refuses_on_api_error() -> None:
     assert not watcher._branch_protection_ok(
         gh, "o", "r", "main", _settings_require_protection(True)
     )
+
+
+# ── D1: reviewer respects the fleet backend ladder (budget/cooldown aware) ─────
+
+
+def _ladder_settings(dynamic: bool = True):
+    from types import SimpleNamespace
+
+    return SimpleNamespace(
+        team_executor=SimpleNamespace(dynamic_worker_backend_selection=dynamic)
+    )
+
+
+def _cool_claude(store, now):
+    from datetime import timedelta
+
+    store.record_worker_backend_cooldown(
+        worker_backend="claude_code",
+        reset_at=now + timedelta(hours=2),
+        now=now,
+        limit_kind="session_5h",  # account-wide → claude fully cooled
+        model=None,
+    )
+
+
+def test_select_review_backend_available_when_no_cooldown(monkeypatch, tmp_path):
+    from datetime import datetime, timezone
+
+    from operations_center.execution.usage_store import UsageStore
+
+    monkeypatch.setenv("OPERATIONS_CENTER_EXECUTION_USAGE_PATH", str(tmp_path / "u.json"))
+    sel = watcher._select_review_backend(
+        _ladder_settings(), usage_store=UsageStore(), now=datetime.now(timezone.utc)
+    )
+    assert sel is not None and sel.selected_backend == "claude_code"
+
+
+def test_select_review_backend_defers_when_claude_over_budget(monkeypatch, tmp_path):
+    from datetime import datetime, timezone
+
+    from operations_center.execution.usage_store import UsageStore
+
+    monkeypatch.setenv("OPERATIONS_CENTER_EXECUTION_USAGE_PATH", str(tmp_path / "u.json"))
+    now = datetime.now(timezone.utc)
+    store = UsageStore()
+    _cool_claude(store, now)
+    sel = watcher._select_review_backend(_ladder_settings(), usage_store=store, now=now)
+    # claude cooled/over-budget → not selected → the reviewer will DEFER instead of burning it
+    assert sel is not None and sel.selected_backend != "claude_code"
+
+
+def test_select_review_backend_respects_dynamic_disabled(monkeypatch, tmp_path):
+    from datetime import datetime, timezone
+
+    from operations_center.execution.usage_store import UsageStore
+
+    monkeypatch.setenv("OPERATIONS_CENTER_EXECUTION_USAGE_PATH", str(tmp_path / "u.json"))
+    now = datetime.now(timezone.utc)
+    store = UsageStore()
+    _cool_claude(store, now)
+    sel = watcher._select_review_backend(_ladder_settings(dynamic=False), usage_store=store, now=now)
+    # operator opted out of the ladder globally → always the preferred backend
+    assert sel is not None and sel.selected_backend == "claude_code"