Speed up fabric path BFS with lazy edge loads and score orbit total by verdict weights.

Only expand adjacency for visited nodes, and rank objective=total with crossing 0.18 + clearance 0.08 (fallback to crossings when clearance is skipped).

Co-authored-by: Cursor <cursoragent@cursor.com>

netx_api/topology_fabric_nodes.py

@@ -503,12 +503,36 @@ def find_fabric_paths(
     max_hops = max(1, min(12, int(max_hops or 6)))
     layer_v = str(layer or "physical").strip() or "physical"
 
-    edges = db.query(TopoFabricEdge).filter(TopoFabricEdge.layer == layer_v).all()
+    # Lazy adjacency: only fetch edges for nodes the BFS actually expands
-    edge_map: dict[str, TopoFabricEdge] = {e.id: e for e in edges}
+    # (avoids loading the entire fabric layer on large graphs).
-    adj: dict[str, list[tuple[str, str]]] = {}
+    adj_cache: dict[str, list[tuple[str, str]]] = {}
-    for e in edges:
+    adj_loaded: set[str] = set()
-        adj.setdefault(e.a_node_id, []).append((e.b_node_id, e.id))
+    edge_map: dict[str, TopoFabricEdge] = {}
-        adj.setdefault(e.b_node_id, []).append((e.a_node_id, e.id))
+
+    def _ensure_adj(node_ids: set[str]) -> None:
+        missing = [n for n in node_ids if n not in adj_loaded]
+        if not missing:
+            return
+        batch = (
+            db.query(TopoFabricEdge)
+            .filter(
+                TopoFabricEdge.layer == layer_v,
+                or_(
+                    TopoFabricEdge.a_node_id.in_(missing),
+                    TopoFabricEdge.b_node_id.in_(missing),
+                ),
+            )
+            .all()
+        )
+        for nid in missing:
+            adj_cache[nid] = []
+            adj_loaded.add(nid)
+        for e in batch:
+            edge_map[e.id] = e
+            if e.a_node_id in missing:
+                adj_cache[e.a_node_id].append((e.b_node_id, e.id))
+            if e.b_node_id in missing:
+                adj_cache[e.b_node_id].append((e.a_node_id, e.id))
 
     # BFS for simple paths so shorter hops are found first; cap expansions on dense graphs.
     _EXPLORE_CAP = 5000
@@ -519,7 +543,8 @@ def find_fabric_paths(
         node, edge_path, visited = queue.popleft()
         if len(edge_path) >= max_hops:
             continue
-        for nbr, eid in adj.get(node, []):
+        _ensure_adj({node})
+        for nbr, eid in adj_cache.get(node, []):
             if nbr in visited:
                 continue
             explored += 1

packages/netx-topology-mcp/src/netx_topology_mcp/layout_ops/orbit_sweep.py

@@ -184,38 +184,86 @@ def _score_key(c: dict[str, Any]) -> tuple:
     )
 
 
+# Verdict weights for the components a single-node orbit move mainly affects.
+_W_CROSS = 0.18
+_W_CLR = 0.08
+
+
+def _crossing_part_score(crossings: int, *, n_links: int, n_nodes: int) -> float:
+    """Match layout_stats crossing sub-score in [0,1] from raw crossing count."""
+    cpl = float(crossings) / max(int(n_links), 1)
+    if n_nodes <= 50:
+        cpl_ok, cpl_bad = 0.05, 0.20
+    elif n_nodes <= 200:
+        cpl_ok, cpl_bad = 0.10, 0.25
+    else:
+        cpl_ok, cpl_bad = 0.16, 0.30
+    if cpl <= cpl_ok:
+        return 1.0
+    if cpl >= cpl_bad:
+        return 0.0
+    return 1.0 - (cpl - cpl_ok) / max(cpl_bad - cpl_ok, 1e-9)
+
+
+def _verdict_partial(crossings: int, clearance_score: float, *, n_links: int, n_nodes: int) -> float:
+    """Weighted crossing+clearance slice of verdict.total (higher is better)."""
+    return (
+        _W_CROSS * _crossing_part_score(crossings, n_links=n_links, n_nodes=n_nodes)
+        + _W_CLR * max(0.0, min(1.0, float(clearance_score)))
+    )
+
+
 def _rerank_by_total(
     scored: list[dict[str, Any]],
     nid: str,
     pos: dict[str, tuple[float, float]],
     names: dict[str, str],
     links: list[tuple[str, str]],
-    adj: dict[str, set[str]],
     *,
+    global0: int,
     re_rank_n: int = 20,
-) -> list[dict[str, Any]]:
+) -> tuple[list[dict[str, Any]], bool, float]:
-    """Re-rank top candidates by crossing + edge_clearance (multi-objective).
+    """Re-rank top candidates by weighted crossing+clearance (verdict.total slice).
 
-    A move may increase crossings but resolve several edge-clearance hits;
+    Returns ``(ranked, clearance_ok, base_partial)``. When edge_clearance is
-    ranking by ``crossings + edge_clearance_hits`` aligns with verdict.total.
+    skipped on large graphs, ``clearance_ok`` is False and ranking is unchanged.
     """
     from netx_topology_mcp.layout_metrics import compute_edge_clearance
 
+    n_nodes = len(pos)
+    n_links = len(links)
+    ec0 = compute_edge_clearance(pos, links, names=names, top_n=1)
+    if ec0.get("edge_clearance_skipped"):
+        return scored, False, 0.0
+
+    base_clr = float(ec0.get("edge_clearance_score") or 1.0)
+    base_partial = _verdict_partial(int(global0), base_clr, n_links=n_links, n_nodes=n_nodes)
+
     n = min(re_rank_n, len(scored))
     for c in scored[:n]:
         trial = dict(pos)
         trial[nid] = (float(c["x"]), float(c["y"]))
         ec = compute_edge_clearance(trial, links, names=names, top_n=1)
+        if ec.get("edge_clearance_skipped"):
+            return scored, False, base_partial
+        clr_s = float(ec.get("edge_clearance_score") or 1.0)
         c["edge_clearance_hits"] = int(ec.get("edge_clearance_hits") or 0)
+        c["edge_clearance_score"] = clr_s
+        c["verdict_partial"] = _verdict_partial(
+            int(c["crossings"]["global"]),
+            clr_s,
+            n_links=n_links,
+            n_nodes=n_nodes,
+        )
     head = sorted(
         scored[:n],
         key=lambda c: (
-            int(c["crossings"]["global"]) + int(c.get("edge_clearance_hits", 0)),
+            -float(c.get("verdict_partial") or 0.0),
             int(c["crossings"]["incident"]),
             float(c.get("stretch") or 1.0),
         ),
     )
-    return head + scored[n:]
+    return head + scored[n:], True, base_partial
 
 
 def _diversify_top(
@@ -488,22 +536,22 @@ def orbit_sweep_node(
             scored.append(c)
 
     scored.sort(key=_score_key)
-    # Multi-objective re-rank: optimize total score, not just crossings
+    # Multi-objective re-rank: weighted crossing+clearance slice of verdict.total
     use_total = objective == "total" and len(scored) > 1
-    base_clearance_hits = 0
+    base_partial = 0.0
     if use_total:
-        from netx_topology_mcp.layout_metrics import compute_edge_clearance
+        scored, clr_ok, base_partial = _rerank_by_total(
-
+            scored, nid, pos, names, links, global0=int(global0)
-        ec0 = compute_edge_clearance(pos, links, names=names, top_n=1)
+        )
-        base_clearance_hits = int(ec0.get("edge_clearance_hits") or 0)
+        if not clr_ok:
-        scored = _rerank_by_total(scored, nid, pos, names, links, adj)
+            # Large-graph clearance skip → fall back to crossing-only ranking.
+            use_total = False
     # Prefer improving moves; still return best even if none improve.
     if use_total:
-        base_total = int(global0) + base_clearance_hits
         improving = [
             c
             for c in scored
-            if int(c["crossings"]["global"]) + int(c.get("edge_clearance_hits", 0)) < base_total
+            if float(c.get("verdict_partial") or 0.0) > base_partial
         ]
     else:
         improving = [c for c in scored if c["delta"]["global"] < 0]
@@ -523,7 +571,9 @@ def orbit_sweep_node(
         "improving_n": len(improving),
         "max_jump": jump,
         "angle_step": angle_step,
-        "objective": objective,
+        "objective": "total" if use_total else (
+            "crossing" if objective != "total" else "crossing_fallback"
+        ),
         "y_band": (
             None if (y_min is None and y_max is None)
             else [y_min, y_max]
@@ -531,8 +581,15 @@ def orbit_sweep_node(
         "hint": (
             "prefer rank1 unless util/label concern; then pick 2/3. "
             "apply with params.pick=1|2|3 or updateTopologyViewPositions."
-            + (" objective=total: rank by crossing+edge_clearance, may trade crossings for clearance."
+            + (
-               if objective == "total" else "")
+                " objective=total: rank by weighted crossing+edge_clearance (verdict slice)."
+                if use_total
+                else (
+                    " objective=total skipped clearance (graph too large); ranked by crossings."
+                    if objective == "total"
+                    else ""
+                )
+            )
         ),
     }
 

packages/netx-topology-mcp/tests/test_orbit_sweep.py

@@ -200,6 +200,9 @@ def test_orbit_objective_total_ranks_clearance_trade() -> None:
     by_total = orbit_sweep_node(st, "h", max_jump=400, nn_floor=20.0, objective="total")
     assert by_cross["ok"] is True and by_total["ok"] is True
     assert by_total.get("objective") == "total"
+    # Tiny graph: clearance runs, so ranked candidates carry verdict_partial.
+    assert by_total.get("candidates")
+    assert all("verdict_partial" in c for c in by_total["candidates"])
     # y_band plumbing
     banded = orbit_sweep_node(
         st, "h", max_jump=400, nn_floor=20.0, objective="total", y_min=0.0, y_max=80.0
@@ -208,3 +211,23 @@ def test_orbit_objective_total_ranks_clearance_trade() -> None:
     assert banded.get("y_band") == [0.0, 80.0]
     for c in banded.get("candidates") or []:
         assert 0.0 <= float(c["y"]) <= 80.0
+
+
+def test_verdict_partial_weights_match_layout_stats() -> None:
+    from netx_topology_mcp.layout_ops.orbit_sweep import (
+        _W_CLR,
+        _W_CROSS,
+        _crossing_part_score,
+        _verdict_partial,
+    )
+
+    assert abs(_W_CROSS - 0.18) < 1e-9
+    assert abs(_W_CLR - 0.08) < 1e-9
+    # Zero crossings on small graph → crossing part 1.0
+    assert _crossing_part_score(0, n_links=10, n_nodes=10) == 1.0
+    # Perfect clearance + perfect crossing
+    assert abs(_verdict_partial(0, 1.0, n_links=10, n_nodes=10) - (_W_CROSS + _W_CLR)) < 1e-9
+    # Worse crossings lower the partial
+    assert _verdict_partial(5, 1.0, n_links=10, n_nodes=10) < _verdict_partial(
+        0, 1.0, n_links=10, n_nodes=10
+    )