Fix conversational metrics and metrics page pagination (#1421)

* fix(frontend): increase metrics fetch limit to prevent missing backend types

Metrics page only fetched 100 items, causing orgs with many metrics to
miss older DeepEval, Rhesis, and Ragas entries in the backend type tabs.

* fix(frontend): paginate metrics fetch to show all backend type tabs

The metrics page fetched only the first 100 metrics, but the backend
caps at 100 per request. Orgs with 100+ metrics lost DeepEval, Rhesis,
and Ragas tabs since older metrics were outside the first page. Added
getAllMetrics() that paginates through all pages.

* feat(frontend): add client-side pagination to metrics grid

* fix(metrics): thread conversation_history through evaluator pipeline

Conversational metrics require conversation_history in evaluate() but
the evaluator only passed single-turn params, causing TypeError. Add
conversation_history param to MetricEvaluator.evaluate(), introspect
for conversation_history and goal in _call_metric_with_introspection(),
build ConversationHistory from conversation_summary in multi-turn
evaluation, and register ConversationalJudge in native factory.

Add introspection completeness guard test that scans all metric classes
to ensure every required evaluate() param is handled by the evaluator.

* style(frontend): fix indentation in metrics grid map callback

* fix(frontend): add pagination robustness guards

- Break getAllMetrics loop on empty response to prevent infinite fetch
- Clamp page index when filtered list shrinks (e.g. after delete)

* fix(metrics): initialize _conversation_history in MetricEvaluator.__init__

Author: harry-rhesis

apps/backend/src/rhesis/backend/metrics/evaluator.py

@@ -19,6 +19,7 @@
 from rhesis.backend.metrics.score_evaluator import ScoreEvaluator
 from rhesis.backend.metrics.utils import diagnose_invalid_metric
 from rhesis.sdk.metrics import BaseMetric, MetricConfig, MetricResult
+from rhesis.sdk.metrics.conversational.types import ConversationHistory
 from rhesis.sdk.metrics.utils import backend_config_to_sdk_config
 
 # Use inline factory creation to avoid circular imports
@@ -63,6 +64,7 @@ def __init__(
         self.model = model  # Store default model for passing to metrics
         self.db = db  # Database session for fetching metric-specific models
         self.organization_id = organization_id  # For secure model lookups
+        self._conversation_history: Optional[ConversationHistory] = None
 
     @staticmethod
     def _get_config_value(
@@ -141,6 +143,7 @@ def evaluate(
         context: List[str],
         metrics: List[Union[Dict[str, Any], MetricConfig, MetricModel]],
         max_workers: int = 5,
+        conversation_history: Optional[ConversationHistory] = None,
     ) -> Dict[str, Any]:
         """
         Compute metrics using the configured backends in parallel.
@@ -176,6 +179,9 @@ def evaluate(
         Returns:
             Dictionary containing scores and details for each metric
         """
+        # Store conversation history for conversational metrics
+        self._conversation_history = conversation_history
+
         if not metrics:
             logger.warning("No metrics provided for evaluation")
             return {}
@@ -892,6 +898,10 @@ def _call_metric_with_introspection(
             kwargs["expected_output"] = expected_output
         if "context" in params:
             kwargs["context"] = context
+        if "conversation_history" in params and self._conversation_history is not None:
+            kwargs["conversation_history"] = self._conversation_history
+        if "goal" in params:
+            kwargs["goal"] = input_text
 
         logger.debug(f"Calling metric '{metric.name}' with parameters: {list(kwargs.keys())}")
 

apps/backend/src/rhesis/backend/tasks/execution/evaluation.py

@@ -19,6 +19,7 @@
 from rhesis.backend.metrics.evaluator import MetricEvaluator
 from rhesis.backend.tasks.execution.constants import MetricScope
 from rhesis.sdk.metrics import MetricConfig
+from rhesis.sdk.metrics.conversational.types import ConversationHistory
 
 from .response_extractor import extract_response_with_fallback
 
@@ -134,28 +135,32 @@ def evaluate_multi_turn_metrics(
         return {}
 
     # Evaluate each metric on the conversation using the MetricEvaluator
-    # For multi-turn, we reconstruct the conversation as a single prompt/response
-    # pair and evaluate with the standard evaluator pipeline.
-    # This approach reuses the existing evaluator infrastructure.
     metrics_evaluator = MetricEvaluator(model=model, db=db, organization_id=organization_id)
 
-    # Build a prompt/response pair from the conversation for evaluation
+    # Build ConversationHistory from conversation_summary for conversational metrics
+    conversation_summary = stored_output.get("conversation_summary", [])
+    messages = []
     conversation_text = ""
-    for turn in stored_output.get("conversation_summary", []):
+    for turn in conversation_summary:
         penelope_msg = turn.get("penelope_message", "")
         target_resp = turn.get("target_response", "")
         if penelope_msg:
+            messages.append({"role": "user", "content": penelope_msg})
             conversation_text += f"User: {penelope_msg}\n"
         if target_resp:
+            messages.append({"role": "assistant", "content": target_resp})
             conversation_text += f"Assistant: {target_resp}\n"
 
+    conversation_history = ConversationHistory.from_messages(messages) if messages else None
+
     try:
         results = metrics_evaluator.evaluate(
             input_text=goal,
             output_text=conversation_text.strip(),
             expected_output="",
             context=[],
             metrics=metric_configs,
+            conversation_history=conversation_history,
         )
     except Exception as e:
         logger.warning(f"Error evaluating multi-turn metrics: {str(e)}")

apps/frontend/src/app/(protected)/metrics/components/MetricsClient.tsx

@@ -136,9 +136,7 @@ export default function MetricsClientComponent({
             sort_by: 'created_at',
             sort_order: 'desc',
           }),
-          metricsClient.getMetrics({
-            skip: 0,
-            limit: 100,
+          metricsClient.getAllMetrics({
             sort_by: 'created_at',
             sort_order: 'desc',
           }),
@@ -148,7 +146,7 @@ export default function MetricsClientComponent({
         const behaviorsData = behaviorsWithMetricsData;
 
         // Use all metrics from the dedicated metrics endpoint
-        const metricsData = allMetricsData.data || [];
+        const metricsData = allMetricsData;
 
         // Add behavior IDs to each metric for compatibility
         const metricsWithBehaviors = metricsData.map(metric => {