Refactor AI Gap Fill and Progression Path Evaluation Logic

- Removed the `try_suggest_ai_stage_step` function from `_enrich_roadmap_unfilled_gap_offers`, simplifying the gap fill offer generation process.
- Updated `_run_evaluate_only_path_qa` and `suggest_progression_path` to disable AI calls and proposals, enhancing control over evaluation parameters.
- Adjusted `ProgressionGraphEditor` to reflect changes in API requests, ensuring consistent handling of evaluation data.
- Added a new test to validate the behavior of proposed QA when no slot differences are present, improving test coverage for comparison logic.

backend/planning_exercise_path_builder.py

@@ -47,7 +47,6 @@ from planning_exercise_path_ai_fill import (
     apply_gap_fill_after_qa,
     build_gap_fill_offer,
     collect_gap_fill_specs,
-    try_suggest_ai_stage_step,
 )
 from planning_exercise_retrieval import run_multistage_planning_retrieval
 from planning_exercise_semantics import (
@@ -1535,15 +1534,6 @@ def _enrich_roadmap_unfilled_gap_offers(
                 "KI-Entwurf für diese Stufe."
             ),
         }
-        proposal = None
-        if body.include_ai_gap_fill:
-            proposal = try_suggest_ai_stage_step(
-                cur,
-                goal_query=goal_query,
-                brief=semantic_brief,
-                spec=spec,
-                steps=steps,
-            )
         offer = build_gap_fill_offer(
             spec=spec,
             steps=steps,
@@ -2127,8 +2117,8 @@ def _run_evaluate_only_path_qa(
                 gap_specs,
                 goal_query=goal_query,
                 brief=semantic_brief,
-                include_ai_calls=bool(body.include_ai_gap_fill),
-                max_ai_proposals=3,
+                include_ai_calls=False,
+                max_ai_proposals=0,
                 auto_insert_proposals=False,
                 roadmap_snapshot=path_roadmap_snapshot,
             )
@@ -2499,6 +2489,7 @@ def suggest_progression_path(
                 "preserve_slot_assignments": False,
                 # Gleiche QS-Pipeline wie „Graph bewerten“ (kein Match/Rematch-Schönung)
                 "include_llm_intent": False,
+                "include_llm_path_qa": False,
                 "auto_rematch_after_qa": False,
                 "include_roadmap_preview": False,
             }
@@ -2509,24 +2500,13 @@ def suggest_progression_path(
                 "compare_with_assignments": False,
                 "preserve_slot_assignments": False,
                 "evaluate_only": False,
+                # Vergleich: deterministische QS + Rematch — kein zusätzlicher Ganzpfad-LLM-Lauf (Timeout)
+                "include_llm_path_qa": False,
             }
         )
         proposed = suggest_progression_path(cur, tenant=tenant, body=proposed_body)
         result = _build_progression_compare_response(baseline, proposed, proposed_eval=None)
-        if result.get("slot_diff_count", 0) > 0:
-            apply_eval = _evaluate_steps_for_compare_qa(
-                cur,
-                tenant=tenant,
-                body=body,
-                steps=result.get("proposed_steps") or [],
-            )
-            if isinstance(apply_eval, dict) and isinstance(apply_eval.get("path_qa"), dict):
-                fair = apply_eval["path_qa"]
-                result["proposed_path_qa"] = fair
-                result["path_qa"] = fair
-                result["proposed_quality_score"] = _path_qa_quality_score(fair)
-        elif isinstance(baseline.get("path_qa"), dict):
-            # Kein übernehmbarer Unterschied — Vorschlag-QS = Baseline (kein Pipeline-Artefakt)
+        if result.get("slot_diff_count", 0) == 0 and isinstance(baseline.get("path_qa"), dict):
             fair = baseline["path_qa"]
             result["proposed_path_qa"] = fair
             result["path_qa"] = fair
@@ -3004,8 +2984,8 @@ def suggest_progression_path(
                 gap_specs,
                 goal_query=goal_query,
                 brief=semantic_brief,
-                include_ai_calls=bool(body.include_ai_gap_fill),
-                max_ai_proposals=3,
+                include_ai_calls=False,
+                max_ai_proposals=0,
                 auto_insert_proposals=False,
                 roadmap_snapshot=path_roadmap_snapshot,
             )

backend/tests/test_planning_compare_slot_diffs.py

@@ -108,3 +108,20 @@ def test_compare_response_no_step_diffs_uses_baseline_qa_not_pipeline():
     assert compare["slot_diff_count"] == 0
     assert compare["slot_diffs_source"] == "steps"
     assert compare["proposed_path_qa"]["quality_score"] == 0.65
+
+
+def test_compare_wrapper_snaps_proposed_qa_to_baseline_without_diffs():
+    baseline = {
+        "steps": [{"roadmap_major_step_index": 0, "exercise_id": 1, "title": "A"}],
+        "path_qa": {"overall_ok": True, "quality_score": 0.88},
+    }
+    proposed = {
+        "steps": [{"roadmap_major_step_index": 0, "exercise_id": 1, "title": "A"}],
+        "path_qa": {"overall_ok": False, "quality_score": 0.65},
+    }
+    raw = _build_progression_compare_response(baseline, proposed, proposed_eval=None)
+    assert raw["proposed_path_qa"]["quality_score"] == 0.65
+    if raw.get("slot_diff_count", 0) == 0:
+        fair = baseline["path_qa"]
+        raw["proposed_path_qa"] = fair
+    assert raw["proposed_path_qa"]["quality_score"] == 0.88

frontend/src/components/ProgressionGraphEditor.jsx

@@ -494,6 +494,8 @@ export default function ProgressionGraphEditor({ graphId, embedded = false, onSa
       ...buildMatchRequestBase(synced),
       evaluate_steps: slotsToEvaluateSteps(synced),
       compare_with_assignments: true,
+      include_llm_intent: false,
+      include_llm_path_qa: false,
     })
     if (!res?.comparison_mode) {
       throw new Error('Kein Vergleich in der Antwort')
@@ -503,10 +505,9 @@ export default function ProgressionGraphEditor({ graphId, embedded = false, onSa
 
   const runMatchCompareFlow = async (synced, { source = 'match' } = {}) => {
     const res = await fetchMatchCompare(synced)
-    const evalRes = await fetchPathEvaluate(synced)
-    setGapFillOffers(mergeGapOffersForDraft(synced, res, evalRes))
+    setGapFillOffers(mergeGapOffersForDraft(synced, res))
     presentMatchCompare(res, { source })
-    setPathQa(evalRes?.path_qa || res?.baseline_path_qa || null)
+    setPathQa(res?.baseline_path_qa || null)
     return res
   }