fix(ci): scope OpenAPI evidence to required tests

scripts/collect-capability-baseline.py

@@ -90,26 +90,52 @@ def playwright_verdict(report: dict[str, Any], required_titles: list[str]) -> tu
         return "fail", counts
     tests = list(iter_tests(report))
     if not tests:
+        if required_titles:
+            counts["failed"] = len(set(required_titles))
+            return "fail", counts
         counts["not_run"] = 1
         return "not_run", counts
     required = {title: False for title in required_titles}
     for test, title in tests:
+        if required and title not in required:
+            continue
         status = test.get("status")
-        results = test.get("results") if isinstance(test.get("results"), list) else []
-        result_statuses = [result.get("status") for result in results if isinstance(result, dict)]
-        retries = [result.get("retry", 0) for result in results if isinstance(result, dict)]
-        if len(result_statuses) != len(results):
+        results = test.get("results")
+        if not isinstance(results, list):
             counts["not_run"] += 1
             continue
-        if status == "flaky" or any(isinstance(retry, int) and retry > 0 for retry in retries):
-            counts["flaky"] += 1
+        result_statuses: list[Any] = []
+        retries: list[int] = []
+        malformed = False
+        for result in results:
+            if not isinstance(result, dict):
+                malformed = True
+                break
+            result_status = result.get("status")
+            retry = result.get("retry", 0)
+            if result_status not in ("passed", "failed", "timedOut", "skipped", "interrupted") \
+                or type(retry) is not int or retry < 0:
+                malformed = True
+                break
+            result_statuses.append(result_status)
+            retries.append(retry)
+        if malformed:
+            counts["not_run"] += 1
+            continue
+        final_status = result_statuses[-1] if result_statuses else None
+        if status in ("unexpected", "failed", "timedOut", "interrupted") \
+            or final_status in ("failed", "timedOut", "interrupted"):
+            counts["failed"] += 1
+        elif status == "flaky":
+            if final_status == "passed" and any(retry > 0 for retry in retries):
+                counts["flaky"] += 1
+            else:
+                counts["not_run"] += 1
         elif status == "skipped" or (not results and status in ("skipped", "expected")):
             counts["skipped"] += 1
-        elif status in ("unexpected", "failed", "timedOut", "interrupted") or any(
-            result_status in ("failed", "timedOut", "interrupted") for result_status in result_statuses
-        ):
-            counts["failed"] += 1
-        elif results and all(result_status == "passed" for result_status in result_statuses):
+        elif any(retry > 0 for retry in retries):
+            counts["flaky"] += 1
+        elif status == "expected" and results and all(result_status == "passed" for result_status in result_statuses):
             counts["passed"] += 1
             if title in required:
                 required[title] = True