Enhance Exercise Retrieval and Path Handling Logic

- Introduced new functions for handling exercise visibility and retrieval based on progression graph context, including `fetch_exercise_rows_by_ids_for_graph`.
- Updated `_load_supplemental_exercise_rows` to incorporate graph visibility rules, improving the accuracy of exercise retrieval.
- Enhanced `_run_path_step_retrieval` to utilize preloaded supplemental exercise rows, optimizing performance and clarity in path step processing.
- Added `exercise_title_equivalent_to_stage_goal` function to improve title matching against learning goals, enhancing exercise relevance.
- Updated tests to validate new retrieval logic and title equivalence functionality, ensuring robustness in exercise selection processes.

backend/planning_exercise_path_builder.py

@@ -6,7 +6,7 @@ planning_progression_roadmap.py und PLANNING_PROGRESSION_ROADMAP_SPEC.md.
 """
 from __future__ import annotations
 
-from typing import Any, Callable, Dict, List, Mapping, Optional, Set, Tuple
+from typing import Any, Callable, Dict, List, Mapping, Optional, Sequence, Set, Tuple
 
 from fastapi import HTTPException
 from pydantic import BaseModel, Field
@@ -298,7 +298,7 @@ def _supplemental_exercise_ids_from_body(
     cur,
     body: ProgressionPathSuggestRequest,
 ) -> List[int]:
-    """Kandidatenpool erweitern — ohne automatisches Slot-Pinning."""
+    """Kandidatenpool erweitern (Graph-Kanten, Boost, Slot-Zuordnungen)."""
     ids: List[int] = []
     for raw in body.evaluate_steps or []:
         if raw.exercise_id is not None:
@@ -308,6 +308,14 @@ def _supplemental_exercise_ids_from_body(
                 continue
             if eid > 0:
                 ids.append(eid)
+    for raw in body.slot_assignments or []:
+        if raw.exercise_id is not None:
+            try:
+                eid = int(raw.exercise_id)
+            except (TypeError, ValueError):
+                continue
+            if eid > 0:
+                ids.append(eid)
     for eid in body.retrieval_boost_exercise_ids or []:
         try:
             val = int(eid)
@@ -319,6 +327,64 @@ def _supplemental_exercise_ids_from_body(
     return list(dict.fromkeys(ids))
 
 
+def _graph_visibility_context(
+    cur,
+    progression_graph_id: Optional[int],
+) -> Tuple[str, Optional[int]]:
+    if not progression_graph_id or int(progression_graph_id) < 1:
+        return "private", None
+    cur.execute(
+        "SELECT visibility, club_id FROM exercise_progression_graphs WHERE id = %s",
+        (int(progression_graph_id),),
+    )
+    row = cur.fetchone()
+    if not row:
+        return "private", None
+    g_club = row.get("club_id")
+    return (
+        str(row.get("visibility") or "private"),
+        int(g_club) if g_club is not None else None,
+    )
+
+
+def _load_supplemental_exercise_rows(
+    cur,
+    *,
+    tenant: TenantContext,
+    progression_graph_id: Optional[int],
+    exercise_ids: Optional[Sequence[int]],
+    vis_sql: str,
+    vis_params: Sequence[Any],
+) -> List[Dict[str, Any]]:
+    """Supplemental-Übungen mit Graph-Sichtbarkeit, Fallback Library-vis_sql."""
+    ids = list(dict.fromkeys(int(x) for x in (exercise_ids or []) if int(x) > 0))
+    if not ids:
+        return []
+    if progression_graph_id and int(progression_graph_id) > 0:
+        from planning_exercise_retrieval import fetch_exercise_rows_by_ids_for_graph
+
+        gvis, gclub = _graph_visibility_context(cur, progression_graph_id)
+        graph_rows = fetch_exercise_rows_by_ids_for_graph(
+            cur,
+            ids,
+            graph_visibility=gvis,
+            graph_club_id=gclub,
+            profile_id=tenant.profile_id,
+            role=tenant.global_role,
+            exercise_allowed_fn=_exercise_allowed_in_progression_graph,
+        )
+        if graph_rows:
+            return graph_rows
+    from planning_exercise_retrieval import fetch_exercise_rows_by_ids
+
+    return fetch_exercise_rows_by_ids(
+        cur,
+        ids,
+        vis_sql=vis_sql,
+        vis_params=vis_params,
+    )
+
+
 def _planning_visibility_sql(
     cur,
     tenant: TenantContext,
@@ -508,6 +574,7 @@ def _run_path_step_retrieval(
     path_primary_topic: Optional[str] = None,
     path_technique_excludes: Optional[List[str]] = None,
     supplemental_exercise_ids: Optional[List[int]] = None,
+    priority_exercise_ids: Optional[List[int]] = None,
 ) -> Tuple[List[Dict[str, Any]], PlanningTargetProfile, Dict[str, Any], str]:
     step_query = step_query_override or step_retrieval_query(
         semantic_brief, goal_query, step_index, max_steps
@@ -617,6 +684,14 @@ def _run_path_step_retrieval(
         progression_graph_id,
     )
 
+    supplemental_rows = _load_supplemental_exercise_rows(
+        cur,
+        tenant=tenant,
+        progression_graph_id=progression_graph_id,
+        exercise_ids=supplemental_exercise_ids,
+        vis_sql=vis_sql,
+        vis_params=vis_params,
+    )
     hits, _skills_by_ex, _full_lib = run_multistage_planning_retrieval(
         cur,
         vis_sql=vis_sql,
@@ -627,9 +702,19 @@ def _run_path_step_retrieval(
         intent=intent,
         intent_weights=weights,
         pack=pack,
-        supplemental_exercise_ids=supplemental_exercise_ids,
+        supplemental_rows_preloaded=supplemental_rows,
     )
-    hits = _enrich_planning_hits_with_variant_meta(cur, hits[:32])
+    from planning_exercise_retrieval import trim_hits_preserving_priority_ids
+
+    priority_ids = list(
+        dict.fromkeys(
+            int(x)
+            for x in (priority_exercise_ids or supplemental_exercise_ids or [])
+            if int(x) > 0
+        )
+    )
+    hits = trim_hits_preserving_priority_ids(hits, priority_ids, limit=48)
+    hits = _enrich_planning_hits_with_variant_meta(cur, hits)
     return hits, target_profile, query_intent_summary, intent
 
 
@@ -727,37 +812,111 @@ def _annotate_roadmap_step(
     if stage_spec.success_criteria:
         step["success_criteria"] = list(stage_spec.success_criteria)
         step["stage_success_criteria"] = list(stage_spec.success_criteria)
-    step["roadmap_match_source"] = "stage_spec"
+    if not step.get("roadmap_match_source"):
+        step["roadmap_match_source"] = "stage_spec"
+    if step.get("exercise_id") is not None:
+        step["slot_status"] = step.get("slot_status") or (
+            "preserved" if step.get("roadmap_match_source") == "slot_reconciled" else "matched"
+        )
+    else:
+        step["slot_status"] = step.get("slot_status") or "unfilled"
     if skill_expectations:
         step["skill_expectations"] = skill_expectations
     return step
 
 
-def _match_roadmap_slot(
+def _try_reconcile_slot_assignment(
     cur,
     *,
+    assignment: EvaluateStepPayload,
+    stage_spec: StageSpecArtifact,
+    major_step: Optional[MajorStep],
     tenant: TenantContext,
+    progression_graph_id: Optional[int],
+    stage_match_brief: Optional[PlanningSemanticBrief],
+    stage_goal: str,
+    stage_anti: Optional[List[str]],
+    path_primary: str,
+    path_tech_excludes: Optional[List[str]],
+) -> Optional[Dict[str, Any]]:
+    """
+    Bestehende Slot-Zuordnung behalten, wenn sie noch zum Stufen-Lernziel passt.
+
+    Validiert gegen dieselben Gates wie Match/QA (relaxed), inkl. Titel-Äquivalenz.
+    """
+    from planning_exercise_semantics import (
+        exercise_passes_stage_fit,
+        exercise_title_equivalent_to_stage_goal,
+    )
+
+    step = _path_step_from_slot_assignment(
+        cur,
+        assignment=assignment,
+        stage_spec=stage_spec,
+        major_step=major_step,
+        tenant=tenant,
+        progression_graph_id=progression_graph_id,
+    )
+    if not step:
+        return None
+
+    title = str(step.get("title") or "").strip()
+    summary = str(step.get("summary") or "").strip()
+    goal = ""
+    cur.execute("SELECT goal FROM exercises WHERE id = %s", (int(step["exercise_id"]),))
+    grow = cur.fetchone()
+    if grow:
+        goal = str(grow.get("goal") or "").strip()
+
+    lg = (stage_goal or stage_spec.learning_goal or "").strip()
+    if exercise_title_equivalent_to_stage_goal(title, lg):
+        step["roadmap_match_source"] = "slot_reconciled"
+        step["slot_status"] = "preserved"
+        step["reasons"] = ["Bestehende Zuordnung (Titel = Lernziel)"] + list(step.get("reasons") or [])[:2]
+        return _annotate_roadmap_step(
+            step,
+            stage_spec=stage_spec,
+            major_step=major_step,
+            anti_patterns_override=stage_anti,
+        )
+
+    if exercise_passes_stage_fit(
+        learning_goal=lg,
+        title=title,
+        summary=summary,
+        goal=goal,
+        stage_brief=stage_match_brief,
+        anti_patterns=stage_anti,
+        path_primary_topic=path_primary or None,
+        path_technique_excludes=path_tech_excludes,
+        relaxed=True,
+    ):
+        step["roadmap_match_source"] = "slot_reconciled"
+        step["slot_status"] = "preserved"
+        step["reasons"] = ["Bestehende Zuordnung (Stufen-Fit)"] + list(step.get("reasons") or [])[:2]
+        return _annotate_roadmap_step(
+            step,
+            stage_spec=stage_spec,
+            major_step=major_step,
+            anti_patterns_override=stage_anti,
+        )
+    return None
+
+
+def _stage_validation_context_for_spec(
+    cur,
+    *,
     body: ProgressionPathSuggestRequest,
     goal_query: str,
-    max_steps: int,
     semantic_brief: PlanningSemanticBrief,
     path_target_profile: PlanningTargetProfile,
-    path_intent: str,
     roadmap_ctx: ProgressionRoadmapContext,
     stage_spec: StageSpecArtifact,
     step_index: int,
     stage_count: int,
-    planned_ids: List[int],
-    anchor_id: Optional[int],
-    anchor_variant_id: Optional[int],
-    used: Set[int],
-) -> Tuple[Optional[Dict[str, Any]], Optional[StageSpecArtifact]]:
-    """Einzelnen Roadmap-Slot matchen (Initial-Build und Auto-Rematch)."""
-    major_by_index: Dict[int, MajorStep] = {}
-    if roadmap_ctx.roadmap:
-        major_by_index = {m.index: m for m in roadmap_ctx.roadmap.major_steps}
-    major = major_by_index.get(stage_spec.major_step_index)
-
+    major: Optional[MajorStep],
+) -> Dict[str, Any]:
+    """Gemeinsamer Kontext für Reconcile + Match eines Roadmap-Slots."""
     ga_dump = (
         roadmap_ctx.goal_analysis.model_dump() if roadmap_ctx.goal_analysis else None
     )
@@ -770,34 +929,6 @@ def _match_roadmap_slot(
         structured=roadmap_ctx.resolved_structured,
         goal_analysis=roadmap_ctx.goal_analysis,
     )
-    brief_summary = (
-        roadmap_ctx.semantic_brief
-        if roadmap_ctx.semantic_brief
-        else brief_to_summary_dict(semantic_brief)
-    )
-
-    stage_spec_dict = stage_spec.model_dump()
-    if major:
-        stage_spec_dict["phase"] = major.phase
-    stage_inp = expectation_input_from_progression_stage(
-        goal_query=goal_query,
-        goal_analysis=ga_dump,
-        resolved_structured=rs_dump,
-        stage_spec=stage_spec_dict,
-        semantic_brief_summary=brief_summary,
-        major_step=major.model_dump() if major else None,
-    )
-    stage_exp = build_planning_skill_expectations(cur, stage_inp, semantic_brief=semantic_brief)
-    step_target = apply_expectations_to_target(path_target_profile, stage_exp)
-    skill_exp_api = stage_exp.to_api_dict() if stage_exp.items else None
-
-    step_query = stage_spec_retrieval_query(
-        semantic_brief=semantic_brief,
-        goal_query=goal_query,
-        stage_spec=stage_spec,
-        major_step=major,
-    )
-    step_kind = resolve_step_exercise_kind_filter(stage_spec, body.exercise_kind_any)
     stage_goal = (stage_spec.learning_goal or "").strip()
     stage_start = (stage_spec.start_state or "").strip()
     stage_target = (stage_spec.target_state or "").strip()
@@ -855,6 +986,105 @@ def _match_roadmap_slot(
         path_target_state=path_target or None,
         contextualized_learning_goal=contextual_goal or None,
     )
+    return {
+        "stage_goal": stage_goal,
+        "stage_anti": stage_anti,
+        "path_primary": path_primary,
+        "path_tech_excludes": path_tech_excludes,
+        "stage_match_brief": stage_match_brief,
+        "path_context_note": path_context_note,
+        "path_anti": path_anti,
+        "path_start": path_start,
+        "path_target": path_target,
+        "ga_dump": ga_dump,
+        "rs_dump": rs_dump,
+    }
+
+
+def _match_roadmap_slot(
+    cur,
+    *,
+    tenant: TenantContext,
+    body: ProgressionPathSuggestRequest,
+    goal_query: str,
+    max_steps: int,
+    semantic_brief: PlanningSemanticBrief,
+    path_target_profile: PlanningTargetProfile,
+    path_intent: str,
+    roadmap_ctx: ProgressionRoadmapContext,
+    stage_spec: StageSpecArtifact,
+    step_index: int,
+    stage_count: int,
+    planned_ids: List[int],
+    anchor_id: Optional[int],
+    anchor_variant_id: Optional[int],
+    used: Set[int],
+    slot_priority_exercise_id: Optional[int] = None,
+) -> Tuple[Optional[Dict[str, Any]], Optional[StageSpecArtifact]]:
+    """Einzelnen Roadmap-Slot matchen (Initial-Build und Auto-Rematch)."""
+    major_by_index: Dict[int, MajorStep] = {}
+    if roadmap_ctx.roadmap:
+        major_by_index = {m.index: m for m in roadmap_ctx.roadmap.major_steps}
+    major = major_by_index.get(stage_spec.major_step_index)
+
+    ctx = _stage_validation_context_for_spec(
+        cur,
+        body=body,
+        goal_query=goal_query,
+        semantic_brief=semantic_brief,
+        path_target_profile=path_target_profile,
+        roadmap_ctx=roadmap_ctx,
+        stage_spec=stage_spec,
+        step_index=step_index,
+        stage_count=stage_count,
+        major=major,
+    )
+    stage_goal = ctx["stage_goal"]
+    stage_anti = ctx["stage_anti"]
+    path_primary = ctx["path_primary"]
+    path_tech_excludes = ctx["path_tech_excludes"]
+    stage_match_brief = ctx["stage_match_brief"]
+    path_context_note = ctx["path_context_note"]
+    ga_dump = ctx["ga_dump"]
+    rs_dump = ctx["rs_dump"]
+
+    brief_summary = (
+        roadmap_ctx.semantic_brief
+        if roadmap_ctx.semantic_brief
+        else brief_to_summary_dict(semantic_brief)
+    )
+
+    stage_spec_dict = stage_spec.model_dump()
+    if major:
+        stage_spec_dict["phase"] = major.phase
+    stage_inp = expectation_input_from_progression_stage(
+        goal_query=goal_query,
+        goal_analysis=ga_dump,
+        resolved_structured=rs_dump,
+        stage_spec=stage_spec_dict,
+        semantic_brief_summary=brief_summary,
+        major_step=major.model_dump() if major else None,
+    )
+    stage_exp = build_planning_skill_expectations(cur, stage_inp, semantic_brief=semantic_brief)
+    step_target = apply_expectations_to_target(path_target_profile, stage_exp)
+    skill_exp_api = stage_exp.to_api_dict() if stage_exp.items else None
+
+    step_query = stage_spec_retrieval_query(
+        semantic_brief=semantic_brief,
+        goal_query=goal_query,
+        stage_spec=stage_spec,
+        major_step=major,
+    )
+    step_kind = resolve_step_exercise_kind_filter(stage_spec, body.exercise_kind_any)
+
+    supplemental_ids = _supplemental_exercise_ids_from_body(cur, body)
+    priority_ids = list(
+        dict.fromkeys(
+            x
+            for x in [slot_priority_exercise_id, *(body.retrieval_boost_exercise_ids or [])]
+            if x is not None and int(x) > 0
+        )
+    )
 
     hits, _, _, _ = _run_path_step_retrieval(
         cur,
@@ -882,7 +1112,8 @@ def _match_roadmap_slot(
         path_context_note=path_context_note,
         path_primary_topic=path_primary or None,
         path_technique_excludes=path_tech_excludes or None,
-        supplemental_exercise_ids=_supplemental_exercise_ids_from_body(cur, body),
+        supplemental_exercise_ids=supplemental_ids,
+        priority_exercise_ids=priority_ids,
     )
 
     hit = _pick_best_path_hit(
@@ -907,6 +1138,7 @@ def _match_roadmap_slot(
         skill_expectations=skill_exp_api,
         anti_patterns_override=stage_anti,
     )
+    step["slot_status"] = "matched"
     return step, None
 
 
@@ -949,7 +1181,8 @@ def _normalize_roadmap_steps_coverage(
                 "roadmap_major_step_index": midx,
                 "roadmap_phase": major.phase if major else None,
                 "roadmap_learning_goal": goal or None,
-                "roadmap_match_source": "stage_spec",
+                "roadmap_match_source": "unfilled",
+                "slot_status": "unfilled",
                 "reasons": [],
             }
         )
@@ -1063,34 +1296,54 @@ def _build_steps_roadmap_first(
     anchor_variant_id: Optional[int] = None
     unfilled: List[Tuple[int, StageSpecArtifact]] = []
     stage_count = len(stage_specs)
-    assignments = (
-        _slot_assignments_by_major_index(body.slot_assignments)
-        if body.preserve_slot_assignments
-        else {}
-    )
+    assignments = _slot_assignments_by_major_index(body.slot_assignments)
     majors_by_index: Dict[int, MajorStep] = {}
     if roadmap_ctx.roadmap:
         majors_by_index = {m.index: m for m in roadmap_ctx.roadmap.major_steps}
 
     for step_index, stage_spec in enumerate(stage_specs):
         major_idx = stage_spec.major_step_index
+        major = majors_by_index.get(major_idx)
+        slot_priority_id: Optional[int] = None
+
         if major_idx in assignments:
-            pinned = _path_step_from_slot_assignment(
+            ctx = _stage_validation_context_for_spec(
+                cur,
+                body=body,
+                goal_query=goal_query,
+                semantic_brief=semantic_brief,
+                path_target_profile=path_target_profile,
+                roadmap_ctx=roadmap_ctx,
+                stage_spec=stage_spec,
+                step_index=step_index,
+                stage_count=stage_count,
+                major=major,
+            )
+            reconciled = _try_reconcile_slot_assignment(
                 cur,
                 assignment=assignments[major_idx],
                 stage_spec=stage_spec,
-                major_step=majors_by_index.get(major_idx),
+                major_step=major,
                 tenant=tenant,
                 progression_graph_id=body.progression_graph_id,
+                stage_match_brief=ctx["stage_match_brief"],
+                stage_goal=ctx["stage_goal"],
+                stage_anti=ctx["stage_anti"],
+                path_primary=ctx["path_primary"],
+                path_tech_excludes=ctx["path_tech_excludes"],
             )
-            if pinned:
-                steps.append(pinned)
-                eid = int(pinned["exercise_id"])
+            if reconciled:
+                steps.append(reconciled)
+                eid = int(reconciled["exercise_id"])
                 used.add(eid)
                 planned_ids.append(eid)
                 anchor_id = eid
-                anchor_variant_id = pinned.get("variant_id")
+                anchor_variant_id = reconciled.get("variant_id")
                 continue
+            try:
+                slot_priority_id = int(assignments[major_idx].exercise_id)
+            except (TypeError, ValueError):
+                slot_priority_id = None
 
         step, unfilled_spec = _match_roadmap_slot(
             cur,
@@ -1109,6 +1362,7 @@ def _build_steps_roadmap_first(
             anchor_id=anchor_id,
             anchor_variant_id=anchor_variant_id,
             used=used,
+            slot_priority_exercise_id=slot_priority_id,
         )
         if not step:
             unfilled.append((step_index, unfilled_spec or stage_spec))

backend/planning_exercise_path_qa.py

@@ -435,7 +435,14 @@ def detect_off_topic_steps(
     for idx, step in enumerate(steps):
         if step.get("is_ai_proposal") or step.get("exercise_id") is None:
             continue
+        stage_goal_early = (step.get("roadmap_learning_goal") or "").strip()
         bundle = _load_exercise_text_bundle(cur, int(step["exercise_id"]))
+        from planning_exercise_semantics import exercise_title_equivalent_to_stage_goal
+
+        if stage_goal_early and exercise_title_equivalent_to_stage_goal(
+            bundle["title"], stage_goal_early
+        ):
+            continue
         blob = _blob_from_fields(
             bundle["title"],
             bundle["summary"],

backend/planning_exercise_retrieval.py

@@ -58,6 +58,21 @@ def _normalize_exercise_kind_filter(exercise_kind_any: Optional[List[str]]) -> L
     return out
 
 
+_EXERCISE_ROW_SELECT = """
+        SELECT e.id, e.title, e.summary, e.method_archetype,
+               e.visibility, e.club_id, e.created_by,
+               (
+                 SELECT fa.name FROM exercise_focus_areas efa
+                 JOIN focus_areas fa ON fa.id = efa.focus_area_id
+                 WHERE efa.exercise_id = e.id
+                 ORDER BY efa.is_primary DESC NULLS LAST, fa.name ASC
+                 LIMIT 1
+               ) AS primary_focus_name,
+               0.0::float AS ft_rank
+        FROM exercises e
+"""
+
+
 def fetch_exercise_rows_by_ids(
     cur,
     exercise_ids: Sequence[int],
@@ -71,16 +86,7 @@ def fetch_exercise_rows_by_ids(
         return []
     ph = ",".join(["%s"] * len(ids))
     sql = f"""
-        SELECT e.id, e.title, e.summary, e.method_archetype,
-               (
-                 SELECT fa.name FROM exercise_focus_areas efa
-                 JOIN focus_areas fa ON fa.id = efa.focus_area_id
-                 WHERE efa.exercise_id = e.id
-                 ORDER BY efa.is_primary DESC NULLS LAST, fa.name ASC
-                 LIMIT 1
-               ) AS primary_focus_name,
-               0.0::float AS ft_rank
-        FROM exercises e
+        {_EXERCISE_ROW_SELECT.strip()}
         WHERE e.id IN ({ph})
           AND ({vis_sql})
           AND COALESCE(e.status, '') <> %s
@@ -90,6 +96,67 @@ def fetch_exercise_rows_by_ids(
     return [dict(r) for r in cur.fetchall()]
 
 
+def fetch_exercise_rows_by_ids_for_graph(
+    cur,
+    exercise_ids: Sequence[int],
+    *,
+    graph_visibility: str,
+    graph_club_id: Optional[int],
+    profile_id: int,
+    role: str,
+    exercise_allowed_fn,
+) -> List[Dict[str, Any]]:
+    """
+    Lädt Übungen nach ID mit Graph-Sichtbarkeitsregeln (nicht Library-vis_sql).
+
+    Ermöglicht Re-Match für im Graph verankerte private Übungen auf Club-Graphen
+    (eigene private) bzw. alle graph-konformen Übungen.
+    """
+    ids = sorted({int(x) for x in exercise_ids if int(x) > 0})
+    if not ids:
+        return []
+    ph = ",".join(["%s"] * len(ids))
+    sql = f"""
+        {_EXERCISE_ROW_SELECT.strip()}
+        WHERE e.id IN ({ph})
+          AND COALESCE(e.status, '') <> %s
+    """
+    cur.execute(sql, [*ids, "archived"])
+    out: List[Dict[str, Any]] = []
+    for row in cur.fetchall() or []:
+        if exercise_allowed_fn(
+            row,
+            graph_visibility=graph_visibility,
+            graph_club_id=graph_club_id,
+            profile_id=profile_id,
+            role=role,
+        ):
+            out.append(dict(row))
+    return out
+
+
+def trim_hits_preserving_priority_ids(
+    hits: Sequence[Mapping[str, Any]],
+    priority_ids: Optional[Sequence[int]],
+    *,
+    limit: int = 48,
+) -> List[Dict[str, Any]]:
+    """Behält priorisierte Graph-/Slot-Übungen im Kandidatenpool (vor pick_best_path_hit)."""
+    priority_set = {int(x) for x in (priority_ids or []) if int(x) > 0}
+    if not priority_set:
+        return list(hits)[:limit]
+    by_id: Dict[int, Dict[str, Any]] = {}
+    for hit in hits:
+        try:
+            by_id[int(hit["id"])] = dict(hit)
+        except (TypeError, ValueError, KeyError):
+            continue
+    priority_hits = [by_id[eid] for eid in sorted(priority_set) if eid in by_id]
+    rest = [dict(h) for h in hits if int(h.get("id") or 0) not in priority_set]
+    merged = priority_hits + rest
+    return merged[: max(limit, len(priority_hits))]
+
+
 def merge_supplemental_exercise_rows(
     rows: Sequence[Dict[str, Any]],
     supplemental: Sequence[Dict[str, Any]],
@@ -485,6 +552,7 @@ def run_multistage_planning_retrieval(
     intent_weights: Mapping[str, float],
     pack: Mapping[str, Any],
     supplemental_exercise_ids: Optional[Sequence[int]] = None,
+    supplemental_rows_preloaded: Optional[Sequence[Dict[str, Any]]] = None,
 ) -> Tuple[List[Dict[str, Any]], Dict[int, Set[int]], bool]:
     """Orchestriert S1b-0 → S1b-1 (Voll-Library-Ranking)."""
     rows = fetch_all_visible_exercise_rows(
@@ -494,7 +562,9 @@ def run_multistage_planning_retrieval(
         query=pack.get("retrieval_query") or query,
         exercise_kind_any=exercise_kind_any,
     )
-    if supplemental_exercise_ids:
+    if supplemental_rows_preloaded:
+        rows = merge_supplemental_exercise_rows(rows, supplemental_rows_preloaded)
+    elif supplemental_exercise_ids:
         extra = fetch_exercise_rows_by_ids(
             cur,
             supplemental_exercise_ids,

backend/planning_exercise_semantics.py

@@ -153,6 +153,48 @@ def _normalize_phrase(text: str) -> str:
     return re.sub(r"\s+", " ", (text or "").strip().lower())
 
 
+_STAGE_TITLE_STOP = frozenset(
+    {"für", "fur", "und", "der", "die", "das", "mit", "im", "in", "am", "an", "zur", "zum", "den", "dem", "des"}
+)
+
+
+def _stage_title_tokens(text: str) -> List[str]:
+    return [
+        tok
+        for tok in _normalize_phrase(text).split()
+        if tok not in _STAGE_TITLE_STOP and len(tok) > 1
+    ]
+
+
+def exercise_title_equivalent_to_stage_goal(title: str, learning_goal: str) -> bool:
+    """
+    Titel entspricht dem Stufen-Lernziel (wortgleich oder nahezu identisch).
+
+    Deckt Graph-Slots ab, bei denen die Übung gezielt zum Lernziel angelegt wurde,
+    ohne dass die Pfad-Haupttechnik im Übungstext vorkommt.
+    """
+    t = _normalize_phrase(title)
+    lg = _normalize_phrase(learning_goal)
+    if len(t) < 3 or len(lg) < 3:
+        return False
+    if t == lg:
+        return True
+    shorter, longer = (t, lg) if len(t) <= len(lg) else (lg, t)
+    if shorter in longer and len(shorter) >= 8 and len(shorter) / max(len(longer), 1) >= 0.72:
+        return True
+    t_tok = _stage_title_tokens(title)
+    lg_tok = _stage_title_tokens(learning_goal)
+    if len(t_tok) >= 2 and t_tok == lg_tok:
+        return True
+    if len(t_tok) >= 2 and len(lg_tok) >= 2:
+        t_set = set(t_tok)
+        lg_set = set(lg_tok)
+        overlap = len(t_set & lg_set)
+        if overlap >= 2 and overlap / max(len(t_set), len(lg_set)) >= 0.85:
+            return True
+    return False
+
+
 def _normalize_query(text: str) -> str:
     return re.sub(r"\s+", " ", (text or "").strip())
 
@@ -1059,6 +1101,9 @@ def exercise_passes_stage_fit(
     if constraints.exclude_phrases and _blob_matches_stage_excludes(blob, constraints.exclude_phrases):
         return False
 
+    if exercise_title_equivalent_to_stage_goal(title, learning_goal or lg):
+        return True
+
     primary_path = (path_primary_topic or "").strip()
     if not primary_path and lg:
         hit = _find_technique_in_text(_normalize_phrase(lg))
@@ -1327,6 +1372,7 @@ __all__ = [
     "build_stage_match_brief",
     "enrich_brief_with_path_constraints",
     "exercise_passes_stage_fit",
+    "exercise_title_equivalent_to_stage_goal",
     "resolve_path_primary_topic",
     "resolve_path_anti_patterns",
     "exercise_passes_stage_learning_goal_gate",

backend/tests/test_planning_exercise_path_builder.py

@@ -18,7 +18,7 @@ class _FakeCur:
         return []
 
 
-def test_supplemental_boost_uses_retrieval_boost_not_slot_pins():
+def test_supplemental_boost_includes_slot_assignments_and_retrieval_boost():
     body = ProgressionPathSuggestRequest(
         query="Mawashi Geri Progression",
         slot_assignments=[
@@ -27,7 +27,7 @@ def test_supplemental_boost_uses_retrieval_boost_not_slot_pins():
         retrieval_boost_exercise_ids=[42, 7],
     )
     ids = _supplemental_exercise_ids_from_body(_FakeCur(), body)
-    assert 99 not in ids
+    assert 99 in ids
     assert 42 in ids
     assert 7 in ids
 

backend/tests/test_planning_roadmap_stage_match.py

@@ -340,6 +340,32 @@ def test_technique_scope_rejects_kumite_when_only_stage_goal_mentions_mawashi():
     )
 
 
+def test_title_equivalent_to_stage_goal():
+    from planning_exercise_semantics import exercise_title_equivalent_to_stage_goal
+
+    assert exercise_title_equivalent_to_stage_goal(
+        "Hüftmobilität für Mae Geri",
+        "Hüftmobilität für Mae Geri",
+    )
+    assert exercise_title_equivalent_to_stage_goal(
+        "Hüftmobilität Mae Geri",
+        "Hüftmobilität für Mae Geri",
+    )
+    assert not exercise_title_equivalent_to_stage_goal("Kumite", "Hüftmobilität für Mae Geri")
+
+
+def test_stage_fit_passes_for_title_equivalent_despite_missing_path_technique():
+    stage_goal = "Koordination Absprung ohne Kick"
+    assert exercise_passes_stage_fit(
+        learning_goal=stage_goal,
+        title=stage_goal,
+        summary="",
+        goal="",
+        path_primary_topic="mawashi geri",
+        path_technique_excludes=["kumite"],
+    )
+
+
 def test_pick_roadmap_relaxed_with_path_primary_when_strict_fails():
     """Bestehende Graph-Übungen: relaxed Gate auch bei gesetztem path_primary_topic."""
     stage_goal = "Hüftmobilität für Mawashi Geri"

frontend/src/components/ProgressionGraphEditor.jsx

@@ -42,6 +42,7 @@ import {
   slotsAsPathStepRows,
   slotsToEvaluateSteps,
   draftRetrievalBoostExerciseIds,
+  slotsToSlotAssignments,
   syncProgressionRoadmapFromSlots,
   syncSlotPhasesFromRoadmap,
 } from '../utils/progressionGraphDraft'
@@ -415,6 +416,7 @@ export default function ProgressionGraphEditor({ graphId, embedded = false, onSa
         include_llm_roadmap: false,
         roadmap_first: true,
         roadmap_override: override,
+        slot_assignments: slotsToSlotAssignments(synced),
         retrieval_boost_exercise_ids: draftRetrievalBoostExerciseIds(synced),
         progression_graph_id: Number(graphId),
         ...roadmapStructuredPayload(draft.startSituation, draft.targetState, draft.roadmapNotes),

frontend/src/utils/progressionGraphDraft.js

@@ -712,7 +712,22 @@ export function draftSiblingEdgePairs(draft) {
   return pairs
 }
 
-/** Bereits zugeordnete Bibliotheks-Übungen — nur Retriever-Boost, kein Pinning. */
+/** Slot-Zuordnungen für Backend-Reconciliation (validiert, nicht blind gepinnt). */
+export function slotsToSlotAssignments(draft) {
+  return (draft.slots || [])
+    .filter((slot) => slot.primary?.kind === 'library' && slot.primary.exerciseId != null)
+    .map((slot) => ({
+      exercise_id: slot.primary.exerciseId,
+      variant_id: slot.primary.variantId || null,
+      title: slot.primary.exerciseTitle || null,
+      is_ai_proposal: false,
+      roadmap_major_step_index: slot.majorStepIndex,
+      roadmap_phase: slot.phase || null,
+      roadmap_learning_goal: slot.learning_goal || null,
+    }))
+}
+
+/** Alle Graph-Übungs-IDs für Retriever-Boost (Slots + Geschwister). */
 export function draftRetrievalBoostExerciseIds(draft) {
   const ids = new Set()
   for (const slot of draft.slots || []) {
@@ -784,7 +799,12 @@ export function applyMatchStepsToSlots(draft, apiSteps) {
     const isProposal = Boolean(step.is_ai_proposal) || step.exercise_id == null
     const hasAiPayload = Boolean(step.ai_suggestion) || Boolean(step.proposal_key)
     if (isProposal && !hasAiPayload) {
-      nextSlots[idx].primary = emptySlotExercise()
+      const wasLibrary =
+        nextSlots[idx].primary?.kind === 'library' && nextSlots[idx].primary.exerciseId != null
+      const mustClear = step.slot_status === 'unfilled' || step.slot_status === 'stripped'
+      if (!wasLibrary || mustClear) {
+        nextSlots[idx].primary = emptySlotExercise()
+      }
     } else if (isProposal) {
       nextSlots[idx].primary = proposalSlotExercise({
         title: step.title || nextSlots[idx].learning_goal,
@@ -800,12 +820,6 @@ export function applyMatchStepsToSlots(draft, apiSteps) {
     }
   }
 
-  for (let i = 0; i < nextSlots.length; i += 1) {
-    if (!touchedMajors.has(i)) {
-      nextSlots[i].primary = emptySlotExercise()
-    }
-  }
-
   return syncProgressionRoadmapFromSlots({ ...draft, slots: nextSlots, dirty: true })
 }