diff --git a/agent/react.py b/agent/react.py
index 9dece69..9cea8da 100644
--- a/agent/react.py
+++ b/agent/react.py
@@ -1,5 +1,5 @@
 """
-ReAct runner: a real LLM (Grok) driving the spend tools in a loop.
+ReAct runner: a real LLM driving the spend tools in a loop.
 
 The LLM reasons in THOUGHT lines, acts with ACTION + ARGS lines, reads
 OBSERVATION lines, and finishes with ANSWER. The runner executes only
@@ -107,7 +107,7 @@ def parse_action(text: str):
 
 
 def _grok_cli() -> str:
-    """Resolve the Grok runner path from PERMIT_GROK_CLI at call time.
+    """Resolve the LLM runner path from PERMIT_GROK_CLI at call time.
 
     Raises RuntimeError with setup guidance when the variable is unset.
     """
@@ -196,7 +196,7 @@ def _learn_ids(recorded_text: str, fresh_text: str,
 
 
 def llm_turn(system: str, history: str) -> str:
-    """One Grok call. Returns the raw text (receipt stripped)."""
+    """One LLM call. Returns the raw text (receipt stripped)."""
     proc = subprocess.run(
         [sys.executable, _grok_cli(), "chat", history,
          "--system", system, "--temperature", "0.3",
diff --git a/demos/demo_six_beat.py b/demos/demo_six_beat.py
index 358e131..a8b02c5 100644
--- a/demos/demo_six_beat.py
+++ b/demos/demo_six_beat.py
@@ -1,8 +1,8 @@
 """
 Permit six-beat demo (mock PayPal rail: no PayPal credentials, no PayPal network).
 
-The human principal grants the permit; a REAL LLM agent (Grok, ReAct loop)
-decides what to spend — the agent's reasoning needs the Grok CLI (network
+The human principal grants the permit; a REAL LLM agent (ReAct loop)
+decides what to spend — the agent's reasoning needs the LLM CLI (network
 + credential) unless --replay is given. The fully offline, deterministic
 camera path is --replay demo_transcript.jsonl: the recorded reasoning is
 replayed verbatim and the recorded actions are re-executed against the
diff --git a/docs/sandbox-runbook.md b/docs/sandbox-runbook.md
index f8f132d..3b8fbbe 100644
--- a/docs/sandbox-runbook.md
+++ b/docs/sandbox-runbook.md
@@ -2,7 +2,7 @@
 
 Status: documented procedure. The automated, camera-ready beats (demos/demo_six_beat.py)
 run in **mock mode** (mock PayPal rail: no PayPal network, no PayPal
-credentials). The agent's reasoning is live (Grok CLI) unless
+credentials). The agent's reasoning is live (LLM CLI) unless
 `--replay demo_transcript.jsonl` is used — that is the fully offline,
 deterministic camera path. This document is
 the scripted procedure for the integrated SANDBOX run: every beat executed against
diff --git a/permit/durability.py b/permit/durability.py
index 0a94e9b..4b13a8d 100644
--- a/permit/durability.py
+++ b/permit/durability.py
@@ -10,7 +10,7 @@
 
 Atomic write: write to *.tmp + fsync + os.rename + fsync dir.
 
-Integrity (REA R2):
+Integrity (review R2):
   - Exclusive writer flock for the process lifetime; a second opener fails closed.
   - ledger.head binding records committed length/hash; truncation fails closed.
   - permits.json is bound to the ledger head (hash + HMAC); rewind fails closed.
diff --git a/permit/ledger.py b/permit/ledger.py
index a9bcf6b..ffc67a4 100644
--- a/permit/ledger.py
+++ b/permit/ledger.py
@@ -9,7 +9,7 @@
 a reordered sequence, or a tampered payload breaks the chain, and the
 settlement verifier refuses to proceed on a broken chain.
 
-No external format dependency (Interlock cut 2026-10-02, Grok review).
+No external format dependency (Interlock cut 2026-10-02, agent review).
 Demo keys are hardcoded and labeled as such — no key-management scope.
 """
 
diff --git a/permit/permit.py b/permit/permit.py
index 7a9b55f..7f24f68 100644
--- a/permit/permit.py
+++ b/permit/permit.py
@@ -1,7 +1,7 @@
 """
 Permit: the authorization layer. Decides what an agent is ALLOWED to spend.
 
-Authority check (Grok review fix — cumulative reservation):
+Authority check (agent review fix — cumulative reservation):
     amount(a) <= remaining(P)
     AND merchant(a) IN allowlist(P)
     AND now < expiry(P)
diff --git a/settlement/verifier.py b/settlement/verifier.py
index 6a67eb2..c562db6 100644
--- a/settlement/verifier.py
+++ b/settlement/verifier.py
@@ -204,7 +204,7 @@ def _original_refusal_id(self, escrow_id: str) -> str | None:
         return original_refused or original_voided
 
     def _append_retry_denial(self, escrow_id: str, reason: str) -> Receipt:
-        """REA F6: unique denial receipt for a refused retry; never captures."""
+        """Review F6: unique denial receipt for a refused retry; never captures."""
         original = self._original_refusal_id(escrow_id)
         return self.ledger.append(
             "REFUSED",
@@ -305,7 +305,7 @@ def _verify_and_capture(
             if escrow.state == "CAPTURED":
                 return VerifyResult(True, "already_captured")
             if escrow.state in ("VOIDED", "REFUSED"):
-                # REA F6: every refused retry writes a unique denial receipt
+                # Review F6: every refused retry writes a unique denial receipt
                 # referencing the original refusal; money is never captured.
                 reason = f"already_{escrow.state.lower()}"
                 self._append_retry_denial(escrow_id, reason)
diff --git a/tests/test_agent.py b/tests/test_agent.py
index de82618..97e968b 100644
--- a/tests/test_agent.py
+++ b/tests/test_agent.py
@@ -97,7 +97,7 @@ def test_system_prompt_lists_catalog():
 
 
 def test_llm_preflight_fails_with_guidance_when_cli_missing(monkeypatch):
-    """A missing Grok runner fails fast with setup guidance — not mid-demo."""
+    """A missing LLM runner fails fast with setup guidance — not mid-demo."""
     monkeypatch.setenv("PERMIT_GROK_CLI", "/nonexistent-xyz/grok.py")
     with pytest.raises(RuntimeError) as exc:
         llm_preflight()
diff --git a/tests/test_rea_r2_pr26_attack.py b/tests/test_rea_r2_pr26_attack.py
index 8608a01..e2f0d44 100644
--- a/tests/test_rea_r2_pr26_attack.py
+++ b/tests/test_rea_r2_pr26_attack.py
@@ -1,6 +1,6 @@
-"""REA R2 attack suite vs Permit PR #26 durability (file-backed).
+"""Review R2 attack suite vs Permit PR #26 durability (file-backed).
 
-Ports the axes from the REA R2 review onto permit/durability.py.
+Ports the axes from the R2 review onto permit/durability.py.
 Each test that previously ACCEPTed a bypass must now REJECT (fail closed)
 or retain the reservation (axis E).
 """
diff --git a/tests/test_settlement.py b/tests/test_settlement.py
index eebe27a..ae1ad4e 100644
--- a/tests/test_settlement.py
+++ b/tests/test_settlement.py
@@ -183,7 +183,7 @@ def _original_refusal_hash(ledger, escrow_id: str) -> str | None:
 
 
 def test_retry_on_refused_or_voided_writes_denial_receipt():
-    """REA F6: every refused retry on an already REFUSED/VOIDED escrow must
+    """Review F6: every refused retry on an already REFUSED/VOIDED escrow must
     append a denial receipt with a unique receipt_id referencing the original
     refusal; capture must never run; hash chain stays valid."""
     for terminal, expected_reason, first_release in (
@@ -229,7 +229,7 @@ def test_retry_on_refused_or_voided_writes_denial_receipt():
 
 
 def test_retry_denial_receipts_serial_and_parallel():
-    """REA F6 concurrency: 25 serial + 25 parallel refused retries each write a
+    """Review F6 concurrency: 25 serial + 25 parallel refused retries each write a
     unique denial receipt referencing the original refusal; none capture."""
     import threading