fix(openai): improve responses api tracing

py/src/braintrust/btx/span_validator.py

@@ -218,7 +218,11 @@ def validate_spans(actual_spans: list[dict[str, Any]], spec: LlmSpanSpec) -> Non
     for i, (actual_span, expected_span) in enumerate(zip(llm_spans, expected_spans)):
         span_errors: list[str] = []
         for key, exp_val in expected_span.items():
-            if key not in actual_span:
+            if key == "child_spans":
+                children = [s for s in actual_spans if actual_span.get("span_id") in (s.get("span_parents") or [])]
+                children.sort(key=lambda s: s.get("span_attributes", {}).get("exec_counter", 0))
+                _validate_value(children, exp_val, f"span[{i}].child_spans", span_errors)
+            elif key not in actual_span:
                 span_errors.append(f"  span[{i}].{key}: key not found in actual span")
             else:
                 _validate_value(actual_span[key], exp_val, f"span[{i}].{key}", span_errors)

py/src/braintrust/btx/spec-ref.txt

@@ -1 +1 @@
-v0.0.10
+v0.0.13

py/src/braintrust/btx/test_btx.py

@@ -100,7 +100,7 @@ def test_btx_spec(
 
         assert btx_project_id is not None
 
-        num_expected = len(spec.expected_brainstore_spans)
+        num_expected = sum(1 + len(span.get("child_spans", [])) for span in spec.expected_brainstore_spans)
         print(f"\n[btx] Fetching {num_expected} span(s) for root_span_id={root_span_id!r} ...")
         spans = fetch_spans(root_span_id, btx_project_id, num_expected)
         print(f"[btx] Got {len(spans)} span(s), validating...")