fix: allow slow benchmark preflights

scripts/compare_three_systems.py

@@ -60,6 +60,7 @@
     str(Path.home() / "anaconda3" / "envs" / "swe_agent_py311"),
 )
 SWE_AGENT_RUNNER = REPO_ROOT / "swe_agent_local_runner.py"
+PREFLIGHT_TIMEOUT_SECONDS = 300
 
 
 @dataclass
@@ -315,7 +316,7 @@ def preflight_claude_endpoint(model: str) -> None:
         env=_claude_environment(model),
         capture_output=True,
         text=True,
-        timeout=90,
+        timeout=PREFLIGHT_TIMEOUT_SECONDS,
     )
     if completed.returncode != 0 or not completed.stdout.strip():
         detail = (completed.stderr or completed.stdout or "empty response")[-500:]
@@ -333,7 +334,7 @@ def preflight_swe_agent_environment() -> None:
         cwd=REPO_ROOT,
         capture_output=True,
         text=True,
-        timeout=120,
+        timeout=PREFLIGHT_TIMEOUT_SECONDS,
     )
     if completed.returncode != 0:
         detail = (completed.stderr or completed.stdout or "unknown import failure")[-1000:]

tests/test_swe_agent_local_runner.py

@@ -7,7 +7,11 @@
 import pytest
 import yaml
 
-from scripts.compare_three_systems import _claude_environment, build_goal_description
+from scripts.compare_three_systems import (
+    _claude_environment,
+    build_goal_description,
+    preflight_claude_endpoint,
+)
 from swe_agent_local_runner import (
     LocalSWEEnv,
     _install_local_environment_stubs,
@@ -161,6 +165,20 @@ def test_claude_environment_can_use_shared_benchmark_endpoint(monkeypatch):
     assert env["ANTHROPIC_MODEL"] == "deepseek-v4-flash[1m]"
 
 
+def test_claude_preflight_allows_slow_official_endpoint(monkeypatch):
+    observed = {}
+
+    def fake_run(*args, **kwargs):
+        observed.update(kwargs)
+        return SimpleNamespace(returncode=0, stdout="OK", stderr="")
+
+    monkeypatch.setattr("scripts.compare_three_systems.subprocess.run", fake_run)
+
+    preflight_claude_endpoint("deepseek-v4-flash")
+
+    assert observed["timeout"] == 300
+
+
 def test_docker_evaluator_does_not_rewrite_official_script_by_default(monkeypatch):
     monkeypatch.delenv("SWE_BENCH_PATCH_EVAL_ENV", raising=False)
     evaluator = _docker_evaluator()