fix: resolve IndexError in exporter and add timeout flag to run command

assayer/cli/main.py

@@ -61,6 +61,7 @@ def cli() -> None:
 @click.option(
     "--judge-criteria", default=None, help="Comma-separated evaluation criteria."
 )
+@click.option("--timeout", type=float, default=30.0, help="Per-model timeout in seconds (default: 30).")
 def run(
     prompt: str | None,
     models: str,
@@ -73,6 +74,7 @@ def run(
     score: bool,
     judge: str | None,
     judge_criteria: str | None,
+    timeout: float,
 ) -> None:
     if prompt_file:
         with open(prompt_file) as f:
@@ -113,6 +115,7 @@ def run(
             system=system,
             temperature=temperature,
             max_tokens=max_tokens,
+            timeout=timeout,
         )
     )
     similarity = compute_similarity(results) if score else None

assayer/exporter.py

@@ -4,6 +4,16 @@
 
 from assayer.models import ModelResult
 
+_FIELDS = [
+    "model",
+    "output",
+    "tokens_input",
+    "tokens_output",
+    "latency_seconds",
+    "cost_usd",
+    "error",
+]
+
 
 def _to_dict(result: ModelResult) -> dict:
     return {
@@ -23,9 +33,10 @@ def export(results: list[ModelResult], path: str) -> None:
 
     if dest.suffix.lower() == ".csv":
         with dest.open("w", newline="", encoding="utf-8") as f:
-            writer = csv.DictWriter(f, fieldnames=list(records[0].keys()))
+            writer = csv.DictWriter(f, fieldnames=_FIELDS)
             writer.writeheader()
-            writer.writerows(records)
+            if records:
+                writer.writerows(records)
     else:
         dest.write_text(
             json.dumps(records, indent=2, ensure_ascii=False), encoding="utf-8"

assayer/runner.py

@@ -8,9 +8,6 @@
 from assayer.providers.openai import OpenAIProvider
 
 
-_TIMEOUT = 30.0
-
-
 def _make_provider(model: str) -> BaseProvider:
     if model.startswith("ollama/"):
         return OllamaProvider(model)
@@ -27,6 +24,7 @@ async def _run_one(
     system: str | None,
     temperature: float | None,
     max_tokens: int | None,
+    timeout: float = 30.0,
 ) -> ModelResult:
     provider = _make_provider(model)
     try:
@@ -37,17 +35,17 @@ async def _run_one(
                 temperature=temperature,
                 max_tokens=max_tokens,
             ),
-            timeout=_TIMEOUT,
+            timeout=timeout,
         )
     except TimeoutError:
         return ModelResult(
             model=model,
             output="",
             tokens_input=0,
             tokens_output=0,
-            latency_seconds=30.0,
+            latency_seconds=timeout,
             cost_usd=0.0,
-            error="Request timed out after 30 seconds",
+            error=f"Request timed out after {timeout} seconds",
         )
 
 
@@ -57,9 +55,10 @@ async def run_all(
     system: str | None = None,
     temperature: float | None = None,
     max_tokens: int | None = None,
+    timeout: float = 30.0,
 ) -> list[ModelResult]:
     tasks = [
-        _run_one(model, prompt, system, temperature, max_tokens)
+        _run_one(model, prompt, system, temperature, max_tokens, timeout)
         for model in models
     ]
     return list(await asyncio.gather(*tasks))

tests/test_exporter.py

@@ -109,3 +109,26 @@ def test_export_unknown_extension_writes_json(tmp_path):
 
     data = json.loads(path.read_text(encoding="utf-8"))
     assert len(data) == 2
+
+def test_export_csv_empty_results_does_not_crash(tmp_path):
+    path = tmp_path / "results.csv"
+    export([], str(path))  # should not raise
+
+
+def test_export_csv_empty_results_has_header(tmp_path):
+    path = tmp_path / "results.csv"
+    export([], str(path))
+
+    with path.open(encoding="utf-8") as f:
+        reader = csv.DictReader(f)
+        assert set(reader.fieldnames) == _EXPECTED_FIELDS
+        rows = list(reader)
+    assert rows == []
+
+
+def test_export_json_empty_results(tmp_path):
+    path = tmp_path / "results.json"
+    export([], str(path))
+
+    data = json.loads(path.read_text(encoding="utf-8"))
+    assert data == []

tests/test_runner.py

@@ -116,9 +116,8 @@ async def _slow_run(self, prompt, **kwargs):
         )
 
     monkeypatch.setattr(OpenAIProvider, "run", _slow_run)
-    monkeypatch.setattr("assayer.runner._TIMEOUT", 0.05)
 
-    result = await _run_one("gpt-4o-mini", "test", None, None, None)
+    result = await _run_one("gpt-4o-mini", "test", None, None, None, timeout=0.05)
 
     assert result.error is not None
     assert "timed out" in result.error