From 017f947e87936de84263e5b6d8249827086f37b1 Mon Sep 17 00:00:00 2001
From: Robert Lippmann <robert.lippmann.development@gmail.com>
Date: Wed, 18 Mar 2026 01:41:08 -0400
Subject: [PATCH] fix: exit nonzero on mediated demo regressions

---
 demos/run_demo.py      |  3 +++
 tests/test_run_demo.py | 31 ++++++++++++++++++++++++++++---
 2 files changed, 31 insertions(+), 3 deletions(-)

diff --git a/demos/run_demo.py b/demos/run_demo.py
index 2f24fdc..01f8a62 100644
--- a/demos/run_demo.py
+++ b/demos/run_demo.py
@@ -188,6 +188,8 @@ def main() -> None:
                     f"→ {report['compiled_prompt_length']} chars "
                     f"({report['prompt_reduction_percent']}% reduction)"
                 )
+        if compiler_regressions > 0:
+            raise SystemExit(1)
         return
 
     try:
@@ -206,6 +208,7 @@ def main() -> None:
         raise SystemExit(2) from exc
     if args.demo in SCORED_DEMOS and result is not None and _is_compiler_regression(result):
         _print_compiler_regression_warning()
+        raise SystemExit(1)
 
 
 if __name__ == "__main__":
diff --git a/tests/test_run_demo.py b/tests/test_run_demo.py
index b98c489..aa99e84 100644
--- a/tests/test_run_demo.py
+++ b/tests/test_run_demo.py
@@ -57,9 +57,11 @@ def fake_run(
     monkeypatch.setattr(run_demo, "_run", fake_run)
     monkeypatch.setattr("sys.argv", ["run_demo", "1"])
 
-    run_demo.main()
+    with pytest.raises(SystemExit) as exc_info:
+        run_demo.main()
     output = capsys.readouterr().out
 
+    assert exc_info.value.code == 1
     assert "result:" in output
     assert "⚠️ MEDIATED REGRESSION" in output
     assert "baseline succeeded but compiler-mediated version failed" in output
@@ -97,9 +99,11 @@ def fake_run(
     monkeypatch.setattr(run_demo, "_run", fake_run)
     monkeypatch.setattr("sys.argv", ["run_demo", "all"])
 
-    run_demo.main()
+    with pytest.raises(SystemExit) as exc_info:
+        run_demo.main()
     output = capsys.readouterr().out
 
+    assert exc_info.value.code == 1
     assert "Baseline results: 1 passed, 0 failed" in output
     assert "Compiler results: 0 passed, 1 failed" in output
     assert "*** 1 MEDIATED REGRESSION DETECTED ***" in output
@@ -137,9 +141,11 @@ def fake_run(
     monkeypatch.setattr(run_demo, "_run", fake_run)
     monkeypatch.setattr("sys.argv", ["run_demo", "all"])
 
-    run_demo.main()
+    with pytest.raises(SystemExit) as exc_info:
+        run_demo.main()
     output = capsys.readouterr().out
 
+    assert exc_info.value.code == 1
     assert "Baseline results: 2 passed, 0 failed" in output
     assert "Compiler results: 0 passed, 2 failed" in output
     assert "*** 2 MEDIATED REGRESSIONS DETECTED ***" in output
@@ -277,6 +283,25 @@ def fake_run(
     assert "Compiler results: 1 passed, 0 failed" in output
 
 
+def test_single_scored_demo_without_mediated_regression_exits_zero(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    def fake_run(
+        path: Path, *, verbose: bool, llm_delay: float
+    ) -> tuple[run_demo.DemoReport | None, run_demo.InfoReport | None]:
+        assert path.name == "fake_01.py"
+        assert not verbose
+        assert llm_delay == 0
+        return _demo_report(baseline_pass=True, compiler_pass=True), None
+
+    monkeypatch.setattr(run_demo, "DEMO_FILES", {"1": "fake_01.py"})
+    monkeypatch.setattr(run_demo, "SCORED_DEMOS", {"1"})
+    monkeypatch.setattr(run_demo, "_run", fake_run)
+    monkeypatch.setattr("sys.argv", ["run_demo", "1"])
+
+    run_demo.main()
+
+
 def test_compaction_demo_reports_sane_metrics() -> None:
     consume_last_info_report()