feat(gooddata-eval): add the agentic what-if-analysis evaluator (#1799)

Scores the what-if skill end to end. MAQL correctness is scoped to adjustments on the
expected measure, so two independent checks can no longer each be satisfied by a
different wrong condition and pass the item between them.

Co-authored-by: Claude Opus 5 <noreply@anthropic.com>

Author: Tomkess

packages/gooddata-eval/src/gooddata_eval/cli/agentic_runner.py

@@ -20,6 +20,7 @@
 from gooddata_eval.core.agentic.metric_skill import evaluate_agentic_metric_skill
 from gooddata_eval.core.agentic.search_tool import evaluate_agentic_search_tool
 from gooddata_eval.core.agentic.visualization import evaluate_agentic_visualization
+from gooddata_eval.core.agentic.what_if import evaluate_agentic_what_if
 from gooddata_eval.core.config import ReasoningEffort
 from gooddata_eval.core.models import AgenticEvalOutcome, CreatedVisualization, DatasetItem
 from gooddata_eval.core.runner import EvalReport, ItemReport
@@ -47,6 +48,7 @@ class _LfKw(TypedDict, total=False):
         "agentic_guardrail",
         "agentic_conversation",
         "agentic_kda_skill",
+        "agentic_what_if",
     }
 )
 
@@ -265,6 +267,17 @@ def _dispatch_agentic(
             agent_id=agent_id,
             **lf_kw,
         )
+    elif kind == "agentic_what_if":
+        return evaluate_agentic_what_if(
+            host=host,
+            token=token,
+            workspace_id=workspace_id,
+            question=item.question,
+            expected_output=eo if isinstance(eo, dict) else {},
+            k=k,
+            agent_id=agent_id,
+            **lf_kw,
+        )
     elif kind == "agentic_conversation":
         fixture_data = eo.get("fixture") or eo if isinstance(eo, dict) else {}
         return evaluate_agentic_conversation(