fix(evals): StepEfficiency negative score on redundant calls within a step

python/fi/evals/metrics/agents/metrics.py

@@ -249,7 +249,8 @@ def compute_one(self, inputs: AgentTrajectoryInput) -> Dict[str, Any]:
                 else:
                     seen_signatures.add(call_sig)
 
-        redundancy_ratio = 1.0 - (redundant_count / total_steps) if total_steps > 0 else 1.0
+        total_calls = sum(len(step.tool_calls) for step in inputs.trajectory)
+        redundancy_ratio = 1.0 - (redundant_count / total_calls) if total_calls > 0 else 1.0
         redundancy_score = redundancy_ratio * self.redundancy_weight
         details["redundant_steps"] = redundant_count
 
@@ -259,7 +260,6 @@ def compute_one(self, inputs: AgentTrajectoryInput) -> Dict[str, Any]:
             for tc in step.tool_calls
             if not tc.success
         )
-        total_calls = sum(len(step.tool_calls) for step in inputs.trajectory)
         failure_ratio = 1.0 - (failed_calls / total_calls) if total_calls > 0 else 1.0
         failure_score = failure_ratio * self.failure_weight
         details["failed_calls"] = failed_calls

python/tests/sdk/test_agents.py

@@ -164,6 +164,31 @@ def test_redundant_steps(self):
         result = metric.compute_one(input_data)
         assert result["details"]["redundant_steps"] >= 3
 
+    def test_redundant_calls_within_single_step(self):
+        """Multiple redundant calls in one step must not push the score below 0."""
+        metric = StepEfficiency()
+        step = AgentStep(
+            step_number=1,
+            thought="Repeating the same call",
+            tool_calls=[
+                ToolCall(
+                    name="search",
+                    arguments={"query": "test"},
+                    result="result",
+                    success=True,
+                )
+                for _ in range(5)
+            ],
+            is_final=True,
+        )
+        input_data = AgentTrajectoryInput(
+            trajectory=[step],
+            task=TaskDefinition(description="Search for something")
+        )
+        result = metric.compute_one(input_data)
+        assert 0.0 <= result["output"] <= 1.0
+        assert result["details"]["redundant_steps"] == 4
+
     def test_failed_tool_calls(self):
         """Test trajectory with failed tool calls."""
         metric = StepEfficiency()