diff --git a/scripts/ci/noema_review_gate.py b/scripts/ci/noema_review_gate.py
index 5ab7e830f3..16291a9a1c 100644
--- a/scripts/ci/noema_review_gate.py
+++ b/scripts/ci/noema_review_gate.py
@@ -321,14 +321,6 @@ def graphql(query: str, **fields: str | int) -> dict[str, Any]:
           }
         }
       }
-      reviews(last: 100) {
-        nodes {
-          state
-          body
-          author { login }
-          commit { oid }
-        }
-      }
       statusCheckRollup {
         contexts(first: 100) {
           nodes {
@@ -356,6 +348,40 @@ def graphql(query: str, **fields: str | int) -> dict[str, Any]:
 """
 
 
+def fetch_complete_reviews(repo: str, number: int) -> list[dict[str, Any]]:
+    """Return every pull-request review normalized to the GraphQL node shape."""
+    document = json.loads(
+        run(
+            [
+                "gh",
+                "api",
+                "--paginate",
+                "--slurp",
+                f"repos/{repo}/pulls/{number}/reviews",
+            ]
+        )
+        or "[]"
+    )
+    if not isinstance(document, list) or any(
+        not isinstance(page, list) for page in document
+    ):
+        raise RuntimeError("GitHub returned malformed paginated review evidence")
+    reviews: list[dict[str, Any]] = []
+    for page in document:
+        for review in page:
+            if not isinstance(review, dict):
+                raise RuntimeError("GitHub returned malformed review evidence")
+            reviews.append(
+                {
+                    "state": review.get("state"),
+                    "body": review.get("body"),
+                    "author": {"login": ((review.get("user") or {}).get("login"))},
+                    "commit": {"oid": review.get("commit_id")},
+                }
+            )
+    return reviews
+
+
 def fetch_pr(repo: str, number: int) -> dict[str, Any]:
     """Fetch the pull request data required for Noema review gating."""
     owner, name = split_repo(repo)
@@ -363,6 +389,7 @@ def fetch_pr(repo: str, number: int) -> dict[str, Any]:
     pr = data.get("data", {}).get("repository", {}).get("pullRequest")
     if not pr:
         raise RuntimeError(f"PR #{number} was not found in {repo}")
+    pr["reviews"] = {"nodes": fetch_complete_reviews(repo, number)}
     return pr
 
 
@@ -1779,7 +1806,7 @@ def inspect_and_review(repo: str, number: int, expected_head: str) -> int:
             "Noema requires an independent reviewer credential."
         )
     if pr.get("isDraft"):
-        print("PR is draft; Noema review skipped.")
+        print("PR is draft; Noema review skipped after primary OpenCode approval.")
         return 0
     if existing_noema_review(pr, actor):
         print("Current head already has a Noema review; nothing to do.")
diff --git a/tests/test_noema_review_gate.py b/tests/test_noema_review_gate.py
index e8a0dd6f59..235b4df9b8 100644
--- a/tests/test_noema_review_gate.py
+++ b/tests/test_noema_review_gate.py
@@ -735,13 +735,18 @@ def test_split_repo_and_graphql(monkeypatch):
 
     def fake_run(args, stdin=None):
         calls.append((args, stdin))
+        if "--paginate" in args:
+            return "[[]]"
         return '{"data":{"repository":{"pullRequest":{"number":7}}}}'
 
     monkeypatch.setattr(noema, "run", fake_run)
     assert noema.graphql("query", owner="owner", number=7)["data"]["repository"]["pullRequest"]["number"] == 7
     assert "-f" in calls[0][0]
     assert "-F" in calls[0][0]
-    assert noema.fetch_pr("owner/repo", 7) == {"number": 7}
+    assert noema.fetch_pr("owner/repo", 7) == {
+        "number": 7,
+        "reviews": {"nodes": []},
+    }
 
     monkeypatch.setattr(noema, "graphql", lambda *args, **kwargs: {"data": {"repository": {"pullRequest": None}}})
     with pytest.raises(RuntimeError, match="was not found"):
@@ -2613,6 +2618,7 @@ def test_format_review_evidence_renders_only_structured_entries():
     assert any("falsified" in line and "source trace passes" in line for line in lines)
 
 
+
 def test_parse_args_and_main(monkeypatch):
     parsed = noema.parse_args(
         ["--repo", "owner/repo", "--pr-number", "9", "--expected-head", "a" * 40]
@@ -2647,3 +2653,83 @@ def test_parse_args_and_main(monkeypatch):
         noema.main(
             ["--repo", "owner/repo", "--pr-number", "9", "--expected-head", "A" * 40]
         )
+
+
+def test_fetch_pr_keeps_exact_head_approval_older_than_one_hundred_reviews(monkeypatch):
+    """The gate must not lose a valid approval behind GitHub's review page size."""
+    head_sha = "a" * 40
+    approval = {
+        "state": "APPROVED",
+        "body": "Result: APPROVE",
+        "user": {"login": "opencode-agent"},
+        "commit_id": head_sha,
+    }
+    later_comments = [
+        {
+            "state": "COMMENTED",
+            "body": f"later review event {index}",
+            "user": {"login": "reviewer"},
+            "commit_id": head_sha,
+        }
+        for index in range(100)
+    ]
+    monkeypatch.setattr(
+        noema,
+        "graphql",
+        lambda *args, **kwargs: {
+            "data": {
+                "repository": {
+                    "pullRequest": make_pr(
+                        headRefOid=head_sha,
+                        reviews={"nodes": later_comments},
+                    )
+                }
+            }
+        },
+    )
+    calls = []
+
+    def fake_run(args, stdin=None):
+        calls.append(args)
+        return json.dumps([[approval, *later_comments]])
+
+    monkeypatch.setattr(noema, "run", fake_run)
+
+    pr = noema.fetch_pr("owner/repo", 7)
+
+    nodes = pr["reviews"]["nodes"]
+    assert len(nodes) == 101
+    assert nodes[0] == {
+        "state": "APPROVED",
+        "body": "Result: APPROVE",
+        "author": {"login": "opencode-agent"},
+        "commit": {"oid": head_sha},
+    }
+    assert calls == [
+        [
+            "gh",
+            "api",
+            "--paginate",
+            "--slurp",
+            "repos/owner/repo/pulls/7/reviews",
+        ]
+    ]
+
+
+@pytest.mark.parametrize(
+    ("payload", "message"),
+    [
+        ("{}", "malformed paginated review evidence"),
+        ("[[null]]", "malformed review evidence"),
+    ],
+)
+def test_fetch_complete_reviews_fails_closed_on_malformed_evidence(
+    monkeypatch,
+    payload,
+    message,
+):
+    """Malformed review pages must not become an empty approval history."""
+    monkeypatch.setattr(noema, "run", lambda *args, **kwargs: payload)
+
+    with pytest.raises(RuntimeError, match=message):
+        noema.fetch_complete_reviews("owner/repo", 7)