diff --git a/deploy/profile-scripts/mofin_db.py b/deploy/profile-scripts/mofin_db.py
index 0350e33b..9cbab104 100644
--- a/deploy/profile-scripts/mofin_db.py
+++ b/deploy/profile-scripts/mofin_db.py
@@ -2083,73 +2083,28 @@ def write_holding_strategy(conn, code: str, name: str, data: dict,
                     data['timing_signal'] = _old_sig
             except Exception:
                 pass
-        # ── 策略参数权威保护（2026-07-27 老爸：技术路径每2分钟覆写12维的Zone/SL/TP/Position→RR波动→盯盘和XMPP不一致）──
-        # 新鲜(<20h)12维分析的技术参数+仓位不允许被技术路径覆写。
-        # 2026-07-27 坑：_old_ra=None 时权威保护永不触发（很多股票的reassessed_at为空），
-        # 导致系统自动计算的仓位被反复踩回"中等仓位"。加入兜底：仓位含"%（系统按"即永保。
-        if source_trigger in _TECHNICAL_PATHS:
-            # 兜底：系统自动计算的仓位永久保护（不含 %（系统按 的不保护，即只有 LLM 仓和系统仓被保护）
-            _old_pos = conn.execute(
-                "SELECT position_advice FROM holding_strategies WHERE code=? AND status='active'",
-                (code,)).fetchone()
-            _old_pos_val = (_old_pos[0] or '') if _old_pos else ''
-            if _old_pos_val and '系统按' in str(_old_pos_val):
-                data['position_advice'] = _old_pos_val
-                # 被保护仓位触发时顺便保护参数（无论 _old_ra 是否空）
-                _op = conn.execute(
-                    "SELECT entry_low, entry_high, stop_loss, take_profit FROM holding_strategies WHERE code=? AND status='active'",
-                    (code,)).fetchone()
-        # ── per_stock 重评保护（2026-08-19 老莫：参数覆写保护）──
-        # per_stock_reassess 重评后 24h 内，其他 source_trigger 不得覆写参数
-        # 判断：最近的 per_stock_12d 快照时间戳（source_trigger='per_stock_12d'）
-        _ps = conn.execute(
-            "SELECT snapshotted_at FROM strategy_history WHERE code=? AND source_trigger='per_stock_12d' ORDER BY snapshotted_at DESC LIMIT 1",
-            (code,)).fetchone()
-        if _ps and _ps[0]:
+        # ── 策略参数权威保护（2026-08-19 重写：白名单机制，根治交叉覆写）──
+        # 只有 LLM 重评路径(per_stock_12d/batch_12d) 和 提拔(promote) 能写策略参数。
+        # 其他调用方(辅助模块/默认write_holding_strategy)：保留 DB 当前参数，不覆写。
+        # 解决：无 per_stock_12d 快照的票(如00020)参数被辅助模块(clean_watchlist/branch_scanner等)反复覆写。
+        _PARAM_WHITELIST = ('per_stock_12d', 'batch_12d', 'promote')
+        if source_trigger not in _PARAM_WHITELIST:
             try:
-                from datetime import datetime as _ddt, timedelta as _dtd
-                _ps_dt = _ddt.fromisoformat(str(_ps[0])[:19])
-                if (_ddt.now() - _ps_dt) < _dtd(hours=24):
-                    # 24h 内重评过 → 保护参数不被覆写
-                    _op2 = conn.execute(
-                        "SELECT entry_low, entry_high, stop_loss, take_profit, rr_ratio FROM holding_strategies WHERE code=? AND status='active'",
-                        (code,)).fetchone()
-                    if _op2 and _op2[0] and _op2[0] > 0:
-                        # 参数保护：当前值保留（除非调用方是 per_stock_12d 本身）
-                        if source_trigger != 'per_stock_12d':
-                            data['entry_low'] = float(_op2[0])
-                            data['entry_high'] = float(_op2[1]) if _op2[1] else _op2[0]
-                            data['stop_loss'] = float(_op2[2]) if _op2[2] else data.get('stop_loss')
-                            data['take_profit'] = float(_op2[3]) if _op2[3] else data.get('take_profit')
-                            data['rr_ratio'] = float(_op2[4]) if _op2[4] else data.get('rr_ratio')
-                            print(f"  [AUTHORITY-24H] {code} per_stock重评24h内,保留参数(区{_op2[0]}~{_op2[1]} 损{_op2[2]})", flush=True)
+                _cur = conn.execute(
+                    "SELECT entry_low, entry_high, stop_loss, take_profit, rr_ratio, position_advice "
+                    "FROM holding_strategies WHERE code=? AND status='active'", (code,)).fetchone()
+                if _cur and _cur[0] and float(_cur[0] or 0) > 0:
+                    data['entry_low'] = float(_cur[0])
+                    data['entry_high'] = float(_cur[1]) if _cur[1] else _cur[0]
+                    data['stop_loss'] = float(_cur[2]) if _cur[2] else data.get('stop_loss')
+                    data['take_profit'] = float(_cur[3]) if _cur[3] else data.get('take_profit')
+                    data['rr_ratio'] = float(_cur[4]) if _cur[4] else data.get('rr_ratio')
+                    # 仓位保护：LLM/系统算的仓位不被辅助覆盖
+                    if _cur[5] and str(_cur[5]).strip():
+                        data['position_advice'] = str(_cur[5])
+                    print(f"  [AUTHORITY-PARAM] {code} 保护策略参数(区{_cur[0]}~{_cur[1]} 损{_cur[2]}) source={source_trigger}", flush=True)
             except Exception:
                 pass
-                if _op and float(_op[0] or 0) > 0:
-                    data['entry_low'] = float(_op[0])
-                    data['entry_high'] = float(_op[1])
-                    data['stop_loss'] = float(_op[2])
-                    data['take_profit'] = float(_op[3])
-                print(f"  [AUTHORITY-POS] {code} 保护系统仓位'{_old_pos_val[:30]}'", flush=True)
-        elif _old_ra:
-                try:
-                    from datetime import datetime as _ddt3, timedelta as _dtd3
-                    _ra_dt3 = _ddt3.fromisoformat(str(_old_ra)[:19])
-                    if (_ddt3.now() - _ra_dt3) < _dtd3(hours=20):
-                        _old_params = conn.execute(
-                            "SELECT entry_low, entry_high, stop_loss, take_profit, position_advice FROM holding_strategies WHERE code=? AND status='active'",
-                            (code,)).fetchone()
-                        if _old_params:
-                            _keys = ['entry_low','entry_high','stop_loss','take_profit','position_advice']
-                            _vals = [v if v else '' for v in _old_params]
-                            for i, k in enumerate(_keys):
-                                if i < 4 and float(_vals[i] or 0) > 0:
-                                    data[k] = float(_vals[i])
-                                elif i == 4 and str(_vals[i]).strip():
-                                    data[k] = str(_vals[i])
-                            print(f"  [AUTHORITY-PARAM] {code} 保留12维参数(区{_vals[0]}~{_vals[1]} 损{_vals[2]} 盈{_vals[3]} pos={_vals[4]})", flush=True)
-                except Exception:
-                    pass
         # ── action 权限保护（与信号同一权威层级，2026-07-22）──
         # 技术路径不得覆盖新鲜(<20h)12维 action。
         # 根治：技术路径写的"盈亏比不足1:1.5不建议买入"旧 action 与12维买入分析同框矛盾。