diff --git a/loopx/capabilities/reward_memory/outcome_lifecycle.py b/loopx/capabilities/reward_memory/outcome_lifecycle.py
index a5e363165..9fc7d9087 100644
--- a/loopx/capabilities/reward_memory/outcome_lifecycle.py
+++ b/loopx/capabilities/reward_memory/outcome_lifecycle.py
@@ -61,6 +61,7 @@
     "evidence_refs",
 }
 _REFLECTION_FIELDS = _LEGACY_REFLECTION_FIELDS | {"experience"}
+_MAX_REFLECTION_BYTES = 16 * 1024
 
 
 def _base(
@@ -230,7 +231,12 @@ def _reflection(value: object) -> dict[str, Any] | None:
     text = str(value or "").strip()
     if not text:
         return None
-    if len(text) > 2_400:
+    # The v1 reflection embeds a bounded procedural-experience contract. Its
+    # independently bounded fields can legitimately exceed the legacy 2,400
+    # character envelope, especially for multi-byte text. Keep a whole-packet
+    # denial-of-service guard, but measure the serialized transport in bytes
+    # and leave semantic compaction to the field-level validators below.
+    if len(text.encode("utf-8")) > _MAX_REFLECTION_BYTES:
         raise ValueError("reward memory reflection exceeds its bounded contract")
     try:
         raw = json.loads(text)
diff --git a/tests/capabilities/test_reward_memory_codex_app_outcome.py b/tests/capabilities/test_reward_memory_codex_app_outcome.py
index ad2567361..c626f0f18 100644
--- a/tests/capabilities/test_reward_memory_codex_app_outcome.py
+++ b/tests/capabilities/test_reward_memory_codex_app_outcome.py
@@ -58,6 +58,26 @@ def _reflection() -> str:
     )
 
 
+def _large_but_valid_reflection() -> str:
+    payload = json.loads(_reflection())
+    experience = payload["experience"]
+    payload["content_summary"] = "C" * 500
+    payload["reasoning_summary"] = "R" * 500
+    experience["applicability"] = ["A" * 500, "B" * 500]
+    experience["observed_outcome"] = "O" * 500
+    experience["attribution"] = "T" * 500
+    experience["future_behavior"] = {
+        "trigger": "G" * 500,
+        "action": "X" * 500,
+        "validation": "V" * 500,
+        "stop_condition": "S" * 500,
+    }
+    experience["limitations"] = ["L" * 500, "M" * 500]
+    serialized = json.dumps(payload)
+    assert len(serialized) > 2_400
+    return serialized
+
+
 def _validator(path: Path, *, attest: bool) -> list[str]:
     body = (
         "import json,sys; r=json.load(sys.stdin); "
@@ -198,3 +218,72 @@ def test_app_refresh_without_exact_validator_attestation_stays_pending(
     assert staged["validation_bound"] is False
     assert staged["provider_sync_count"] == 0
     assert staged["external_writes_performed"] is False
+
+
+def test_app_refresh_accepts_large_valid_reflection_without_validator(
+    tmp_path: Path,
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    monkeypatch.setattr(
+        codex_app_outcome,
+        "_validation_declaration",
+        lambda **_kwargs: None,
+    )
+    monkeypatch.setattr(
+        codex_app_outcome,
+        "_goal_repo",
+        lambda *_args, **_kwargs: tmp_path,
+    )
+
+    staged = stage_codex_app_turn_outcome_candidate(
+        registry_path=tmp_path / "registry.json",
+        runtime_root=tmp_path / "runtime",
+        goal_id="goal",
+        agent_id="pilot",
+        todo_id="todo_without_validator",
+        turn_instance_id="turn-large-reflection",
+        effect_id="effect:large-reflection",
+        state_file=tmp_path / "ACTIVE_GOAL_STATE.md",
+        validation_workspace=tmp_path,
+        reflection_json=_large_but_valid_reflection(),
+        observed_at="2026-09-21T15:02:02+00:00",
+    )
+
+    assert staged["status"] == "awaiting_evidence_validation"
+    assert staged["reason_code"] == "reflection_not_bound_to_independent_validation"
+    assert staged["candidate_id"]
+    assert staged["reflection_digest"].startswith("sha256:")
+    assert staged["validation_bound"] is False
+    assert staged["provider_sync_count"] == 0
+    assert staged["external_writes_performed"] is False
+
+
+def test_app_refresh_rejects_truly_oversized_reflection(
+    tmp_path: Path,
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    monkeypatch.setattr(
+        codex_app_outcome,
+        "_goal_repo",
+        lambda *_args, **_kwargs: tmp_path,
+    )
+    payload = json.loads(_reflection())
+    payload["content_summary"] = "X" * (17 * 1024)
+
+    with pytest.raises(
+        ValueError,
+        match="reward memory reflection exceeds its bounded contract",
+    ):
+        stage_codex_app_turn_outcome_candidate(
+            registry_path=tmp_path / "registry.json",
+            runtime_root=tmp_path / "runtime",
+            goal_id="goal",
+            agent_id="pilot",
+            todo_id="todo_oversized",
+            turn_instance_id="turn-oversized-reflection",
+            effect_id="effect:oversized-reflection",
+            state_file=tmp_path / "ACTIVE_GOAL_STATE.md",
+            validation_workspace=tmp_path,
+            reflection_json=json.dumps(payload),
+            observed_at="2026-09-21T15:02:02+00:00",
+        )