fix: track errored/undetermined objectives in red team result counts

Previously, objectives that failed during attack execution or risk
categories with zero prepared objectives were silently dropped from the
pipeline. The result_counts.errored field always showed 0 because
_compute_result_count only counted existing output items.

Changes:
- _execution_manager.py: Record 0-objective categories as failed in
  red_team_info instead of silently skipping. Add expected_count to all
  red_team_info entries to track expected vs actual objectives.
- _result_processor.py: Add _extract_expected_total() to compute total
  expected objectives from red_team_info (de-duplicated by risk
  category). Pass expected_total to _compute_result_count() which now
  computes errored as the delta between expected and actual items. Add
  partial_failure to _determine_run_status failure detection.
- test_result_processor_errored.py: 31 new unit tests covering
  _compute_result_count with expected_total, _extract_expected_total
  de-duplication logic, and _determine_run_status failure detection.
- test_foundry.py: 3 new tests for 0-objective recording and
  expected_count propagation in FoundryExecutionManager.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Author: slister1001

sdk/evaluation/azure-ai-evaluation/azure/ai/evaluation/red_team/_foundry/_execution_manager.py

@@ -122,7 +122,22 @@ async def execute_attacks(
                 objectives = objectives_by_risk.get(risk_value, [])
 
                 if not objectives:
-                    self.logger.info(f"No objectives for {risk_value}, skipping")
+                    self.logger.info(f"No objectives for {risk_value}, recording as failed")
+                    # Record zero-objective categories so _determine_run_status
+                    # detects the failure and errored counts reflect the gap.
+                    from .._utils.formatting_utils import get_strategy_name
+
+                    if include_baseline:
+                        strategy_key = get_strategy_name(AttackStrategy.Baseline)
+                        if strategy_key not in red_team_info:
+                            red_team_info[strategy_key] = {}
+                        red_team_info[strategy_key][risk_value] = {
+                            "data_file": "",
+                            "status": "failed",
+                            "error": "No attack objectives could be prepared for this risk category",
+                            "asr": 0.0,
+                            "expected_count": 0,
+                        }
                     continue
 
                 self.logger.info(f"Processing {len(objectives)} objectives for {risk_value}")
@@ -186,6 +201,7 @@ async def execute_attacks(
                             "error": str(e),
                             "partial_failure": True,
                             "asr": 0.0,
+                            "expected_count": len(objectives),
                         }
                     else:
                         self.logger.error(f"Error executing attacks for {risk_value}: {e}")
@@ -197,6 +213,7 @@ async def execute_attacks(
                             "status": "failed",
                             "error": str(e),
                             "asr": 0.0,
+                            "expected_count": len(objectives),
                         }
                         continue
 
@@ -223,6 +240,7 @@ async def execute_attacks(
                     output_path=output_path,
                     attack_strategies=attack_strategies,
                     include_baseline=include_baseline,
+                    num_objectives=len(objectives),
                 )
 
                 for strategy_name, strategy_data in strategy_results.items():
@@ -357,6 +375,7 @@ def _group_results_by_strategy(
         output_path: str,
         attack_strategies: List[Union[AttackStrategy, List[AttackStrategy]]],
         include_baseline: bool,
+        num_objectives: int = 0,
     ) -> Dict[str, Dict[str, Any]]:
         """Group attack results by strategy for red_team_info format.
 
@@ -375,6 +394,8 @@ def _group_results_by_strategy(
         :type attack_strategies: List[Union[AttackStrategy, List[AttackStrategy]]]
         :param include_baseline: Whether baseline was included in execution
         :type include_baseline: bool
+        :param num_objectives: Number of objectives sent for this risk category
+        :type num_objectives: int
         :return: Dictionary mapping strategy name to result data
         :rtype: Dict[str, Dict[str, Any]]
         """
@@ -395,6 +416,7 @@ def _group_results_by_strategy(
                 "data_file": output_path,
                 "status": "completed",
                 "asr": overall_asr,
+                "expected_count": num_objectives,
             }
 
         # Add entries for special strategies that were executed (e.g., IndirectJailbreak via XPIA)
@@ -407,6 +429,7 @@ def _group_results_by_strategy(
                     "data_file": output_path,
                     "status": "completed",
                     "asr": overall_asr,
+                    "expected_count": num_objectives,
                 }
 
         # Add baseline entry if it was included
@@ -415,6 +438,7 @@ def _group_results_by_strategy(
                 "data_file": output_path,
                 "status": "completed",
                 "asr": overall_asr,
+                "expected_count": num_objectives,
             }
 
         # Fallback if no strategies produced results
@@ -423,6 +447,7 @@ def _group_results_by_strategy(
                 "data_file": output_path,
                 "status": "completed",
                 "asr": overall_asr,
+                "expected_count": num_objectives,
             }
 
         return results

sdk/evaluation/azure-ai-evaluation/azure/ai/evaluation/red_team/_result_processor.py

@@ -1424,16 +1424,52 @@ def _format_thresholds_for_output(self) -> Dict[str, Any]:
         return formatted_thresholds
 
     @staticmethod
-    def _compute_result_count(output_items: List[Dict[str, Any]]) -> Dict[str, int]:
+    def _extract_expected_total(red_team_info: Optional[Dict]) -> Optional[int]:
+        """Extract the total expected objective count from red_team_info.
+
+        Each strategy/risk-category entry may carry an ``expected_count`` field
+        set by the execution layer.  Since the same risk category can appear
+        under multiple strategies (e.g. baseline + crescendo), we de-duplicate
+        by risk category to avoid double-counting.
+
+        :param red_team_info: The red_team_info tracking dictionary
+        :return: Total expected objectives, or None if information is unavailable
+        """
+        if not isinstance(red_team_info, dict):
+            return None
+
+        # Collect the max expected_count per risk category across strategies
+        per_risk: Dict[str, int] = {}
+        for risk_data in red_team_info.values():
+            if not isinstance(risk_data, dict):
+                continue
+            for risk_category, details in risk_data.items():
+                if not isinstance(details, dict):
+                    continue
+                count = details.get("expected_count")
+                if count is not None:
+                    per_risk[risk_category] = max(per_risk.get(risk_category, 0), int(count))
+
+        if not per_risk:
+            return None
+
+        return sum(per_risk.values())
+
+    @staticmethod
+    def _compute_result_count(
+        output_items: List[Dict[str, Any]],
+        expected_total: Optional[int] = None,
+    ) -> Dict[str, int]:
         """Aggregate run-level pass/fail counts from individual output items.
 
         Counts reflect attack success rate (ASR) semantics:
         - passed: attacks that were unsuccessful (system defended successfully)
         - failed: attacks that were successful (system was compromised)
-        - errored: rows that failed to process due to errors
+        - errored: rows that failed to process due to errors, plus any
+          objectives that never produced output items (expected vs actual delta)
         """
 
-        total = len(output_items)
+        actual_total = len(output_items)
         passed = failed = errored = 0
 
         for item in output_items:
@@ -1474,6 +1510,15 @@ def _compute_result_count(output_items: List[Dict[str, Any]]) -> Dict[str, int]:
             else:
                 errored += 1
 
+        # Account for objectives that never produced output items (e.g., attack
+        # execution errors that were caught and swallowed, or risk categories
+        # with zero objectives prepared).
+        if expected_total is not None and expected_total > actual_total:
+            missing = expected_total - actual_total
+            errored += missing
+
+        total = expected_total if expected_total is not None and expected_total > actual_total else actual_total
+
         return {
             "total": total,
             "passed": passed,
@@ -1675,7 +1720,7 @@ def _determine_run_status(
                     if not isinstance(details, dict):
                         continue
                     status = details.get("status", "").lower()
-                    if status in ("incomplete", "failed", "timeout", "pending", "running"):
+                    if status in ("incomplete", "failed", "timeout", "pending", "running", "partial_failure"):
                         return "failed"
 
         return "completed"
@@ -1769,7 +1814,10 @@ def _build_results_payload(
         if run_name is None:
             run_name = scan_name or f"redteam-run-{run_id[:8]}"
 
-        result_count = self._compute_result_count(output_items)
+        result_count = self._compute_result_count(
+            output_items,
+            expected_total=self._extract_expected_total(red_team_info),
+        )
         per_testing_results = self._compute_per_testing_criteria(output_items)
         data_source = self._build_data_source_section(parameters, red_team_info)
         status = self._determine_run_status(scan_result, red_team_info, output_items)

sdk/evaluation/azure-ai-evaluation/tests/unittests/test_redteam/test_foundry.py

@@ -1749,7 +1749,7 @@ def test_group_results_by_strategy_with_indirect_jailbreak(
 
     @pytest.mark.asyncio
     async def test_execute_attacks_empty_objectives(self, mock_credential, mock_azure_ai_project, mock_logger):
-        """Test execute_attacks with no objectives."""
+        """Test execute_attacks with no objectives for any risk category."""
         manager = FoundryExecutionManager(
             credential=mock_credential,
             azure_ai_project=mock_azure_ai_project,
@@ -1762,12 +1762,75 @@ async def test_execute_attacks_empty_objectives(self, mock_credential, mock_azur
         result = await manager.execute_attacks(
             objective_target=mock_target,
             risk_categories=[RiskCategory.Violence],
-            attack_strategies=[AttackStrategy.Base64],
+            attack_strategies=[AttackStrategy.Baseline],
             objectives_by_risk={},  # No objectives
         )
 
-        # Should return empty dict when no objectives
-        assert result == {}
+        # When no objectives are available at all the category should be
+        # recorded as failed in red_team_info so _determine_run_status can
+        # detect the gap.
+        assert "baseline" in result
+        assert "violence" in result["baseline"]
+        entry = result["baseline"]["violence"]
+        assert entry["status"] == "failed"
+        assert entry["expected_count"] == 0
+        assert "error" in entry
+
+    @pytest.mark.asyncio
+    async def test_execute_attacks_zero_objectives_records_failed(
+        self, mock_credential, mock_azure_ai_project, mock_logger
+    ):
+        """When a risk category has zero objectives, it should be recorded as failed
+        in red_team_info so that _determine_run_status marks the run as failed."""
+        manager = FoundryExecutionManager(
+            credential=mock_credential,
+            azure_ai_project=mock_azure_ai_project,
+            logger=mock_logger,
+            output_dir="/test/output",
+        )
+
+        mock_target = MagicMock()
+
+        result = await manager.execute_attacks(
+            objective_target=mock_target,
+            risk_categories=[RiskCategory.Violence, RiskCategory.SelfHarm],
+            attack_strategies=[AttackStrategy.Baseline],
+            objectives_by_risk={
+                "violence": [],  # explicitly empty
+                # self_harm not present at all
+            },
+        )
+
+        # Both risk categories should be recorded as failed
+        assert "baseline" in result
+        assert result["baseline"]["violence"]["status"] == "failed"
+        assert result["baseline"]["self_harm"]["status"] == "failed"
+
+    def test_group_results_by_strategy_includes_expected_count(
+        self, mock_credential, mock_azure_ai_project, mock_logger
+    ):
+        """Verify _group_results_by_strategy includes expected_count in entries."""
+        manager = FoundryExecutionManager(
+            credential=mock_credential,
+            azure_ai_project=mock_azure_ai_project,
+            logger=mock_logger,
+            output_dir="/test/output",
+        )
+
+        mock_orchestrator = MagicMock()
+        mock_orchestrator.calculate_asr.return_value = 0.5
+
+        results = manager._group_results_by_strategy(
+            orchestrator=mock_orchestrator,
+            risk_value="violence",
+            output_path="/test/output.jsonl",
+            attack_strategies=[AttackStrategy.Baseline],
+            include_baseline=True,
+            num_objectives=32,
+        )
+
+        assert "baseline" in results
+        assert results["baseline"]["expected_count"] == 32
 
     @pytest.mark.asyncio
     async def test_execute_attacks_filters_multi_turn_without_adversarial(