From b8b71523b551225c40b752d53ecc3d09b53ac92b Mon Sep 17 00:00:00 2001
From: fengxiangdebaozi <10845145+fengxiangdebaozi@user.noreply.gitee.com>
Date: Wed, 17 Jun 2026 23:18:05 +0800
Subject: [PATCH] =?UTF-8?q?fix:=20=E6=B5=81=E7=A8=8B=E5=9B=BE=E8=BF=9E?=
 =?UTF-8?q?=E7=BA=BF=E6=8E=A8=E5=AF=BC=E5=8A=A0=E5=9B=BA=E2=80=94=E2=80=94?=
 =?UTF-8?q?=E4=B8=AA=E5=88=AB=E8=8A=82=E7=82=B9=E6=BC=8F=20group=20?=
 =?UTF-8?q?=E6=97=B6=E6=8C=89=E5=8F=91=E5=B0=84=E9=A1=BA=E5=BA=8F=E5=B0=B1?=
 =?UTF-8?q?=E8=BF=91=E5=BD=92=E5=B1=82(carry-forward)?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

弱模型可能漏标个别节点的 group，原逻辑一漏标就整体放弃推导导致节点平铺。改为漏 group 的
节点沿用前一节点的层级，只有开头无 group / 全程无 group 才回退模型连边。

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 .../backend/app/gateway/routers/flow_extract.py   | 15 ++++++++++++---
 1 file changed, 12 insertions(+), 3 deletions(-)

diff --git a/offline-backend-20260512/backend/app/gateway/routers/flow_extract.py b/offline-backend-20260512/backend/app/gateway/routers/flow_extract.py
index 213f5d21..ae8cac9e 100644
--- a/offline-backend-20260512/backend/app/gateway/routers/flow_extract.py
+++ b/offline-backend-20260512/backend/app/gateway/routers/flow_extract.py
@@ -346,18 +346,27 @@ def _derive_layered_edges(nodes: list[dict[str, Any]]) -> list[dict[str, Any]] |
     返回 None 表示「无法可靠推导」（无 group / 只有一层 / 存在不规则层对），调用方回退模型连边。
     """
     # 按 group 分层，保持每层内节点的原始顺序 + 层的首次出现顺序。
+    # 加固：个别节点漏 group（弱模型可能漏标）→ 按发射顺序**就近归入前一个层级**（carry-forward），
+    # 不再因一个漏标就整体放弃；只有连开头节点都没 group / 全程无任何 group 才放弃。
     order: list[str] = []
     layers: dict[str, list[str]] = {}
+    last_group = ""
+    any_group = False
     for n in nodes:
         g = _as_str(n.get("group"))
+        if g:
+            any_group = True
+            last_group = g
+        else:
+            g = last_group  # 漏 group → 沿用前一节点的层级
         if not g:
-            return None  # 有节点没层级 → 无法分层，回退模型连边
+            return None  # 开头节点就没 group，无法锚定层级 → 回退模型连边
         if g not in layers:
             layers[g] = []
             order.append(g)
         layers[g].append(n["id"])
-    if len(order) < 2:
-        return None  # 只有一层，没有层间连线可推
+    if not any_group or len(order) < 2:
+        return None  # 全程无 group / 只有一层 → 回退模型连边
     out: list[dict[str, Any]] = []
     for k in range(len(order) - 1):
         a = layers[order[k]]