tradingSystem/app/services/strategy_runner.py

# -*- coding: utf-8 -*-
"""
个股交易方案 (策略) 运行器 —— PER_STOCK_STRATEGY_PLAN.md §四/§六/§七/§七B
============================================================================
每分钟一跳 (挂在 scheduler.intraday_exec 里, 与 run_tick 并列)。载入 ACTIVE 策略, 按类型
评估 (做T / 网格 / 跟踪止盈), 触发就**发一张短窗口指令** (window_tdays=1, is_command=True,
origin_type='strategy'), 交 executor.run_tick 用现有管线执行 —— 择时 / 规则闸 / T+1 可卖封顶 /
下发 / 账本一道不重写; autonomy=confirm 的落一条提议进「等我拍板」, 人点采纳后再由本层发指令。

**只做加法**: 本模块不改 executor / rule_gate / action_engine 任何一行, 唯一的引擎触点是
action_engine.scan 早已加好的「有 ACTIVE 策略的票跳过」那一条 skip (设计 §四)。

一条铁律定了整个框架 —— T+1 (设计 §三)
--------------------------------------------------
当日买入不可当日卖出, 所以做T / 网格 / 跟踪止盈本质都是「在一只**底仓**上、用 T+1 可卖的存量股
做买卖」= 同一自动机的三种配置。卖出一律经 run_tick 按 avail_qty 封顶, T+1 天然被挡在那里。

批次口径 (与 app/core/recon.ACTION_TO_LOT 对齐, 不另立)
--------------------------------------------------
  做T   —— 买入与卖出都用 action='T0_ROUND' → 记 T0 批次; 卖出核销次序 T0→ADD→DCA→FILL→BASE
           先把 T0 批次对冲掉, 底仓与摊薄成本不动, 做T利润自然摊入 realized_t_profit。
  网格   —— 买入 action='ADD' (记 ADD 批次), 卖出 action='TRIM' (从 avail 卖)。
  跟踪止盈 —— 卖出 action='TRIM' (部分) / 'EXIT' (全清)。只卖不买。

安全 (设计 §九)
--------------------------------------------------
  * 全局开关 PMS_STRATEGY_ENABLED (默认 False) —— 关着时本模块整体空转、一条指令都不发。
  * 挂了 ACTIVE 策略的票由 action_engine.scan 排除 (两个大脑不抢同一只)。
  * 策略动作走命令口径 (is_command=True, 过规则闸、不过研判闸)。
  * scheduler 的 @guard(session=True) 兜住: 非交易时段 / 休假模式不跑。
  * 影子/实盘由 PMS_DISPATCH_MODE 决定, 与本层无关、自动继承。
  * **双保险**: 本层 rails 先拦 (熔断 / 当日次数 / 上限 / 平回 / 下界), 规则闸再拦一道 (合规)。
  * **关键路径禁止丢弃返回值**: 发不出指令 / 落不了库一律进 out["errors"], 绝不静默当成功。

决策系统对策略票的边界 (2026-08-11 用户定)
--------------------------------------------------
挂了策略的票, 决策系统的风控卖出信号只提示、不自动清仓 (强制离场会推翻你特意设的策略);
`signal_service` 会把它落成提议、并把该票策略的**买入这一侧暂停** (不平仓、不动卖出、可页面恢复)。
暂停标记存在运行参数 `PMS_STRATEGY_BUYPAUSE` (按 ts_code 映射), 由 strategy_service 维护、
本模块只读 —— 特意不放进策略 state_json: 那份 state 每跳都被本模块重写, 放进去会被并发的
signal_digest / intraday_exec 互相覆盖 (与 portfolio.neg_streak_map 用独立参数同一个道理)。
暂停只挡「开新仓 / 加仓」这一侧, 卖出、平回、跟踪止盈照常 —— 挡的是"资金在流出、网格还在
逢跌买入", 不替你做清仓这种不可逆的事。
"""
from __future__ import annotations

import logging
from datetime import datetime

from app.core import tradedays as td
from app.core.sizer import LOT
from app.repo import pms_repo
from app.services import market, param_store, portfolio

logger = logging.getLogger("pms.strategy")

# 指令在途 (未终态) 的状态集 —— 与 executor.LIVE 一致; 本层据此判断「上一笔还没走完, 先别再发」。
LIVE_INS = ("PROPOSED", "RULE_PASSED", "DISPATCHED")

# 批次动作 (与 recon.ACTION_TO_LOT 对齐)
A_T0 = "T0_ROUND"      # 做T 买入与卖出 → T0 批次
A_GRID_BUY = "ADD"     # 网格买入 → ADD 批次
A_SELL = "TRIM"        # 网格 / 跟踪止盈 部分卖
A_EXIT = "EXIT"        # 跟踪止盈 全清 (允许零股一次性清出)

# 触发用的小额贴近带 (设计只说「近支撑 / 近压力 / 滞涨」, 未给具体数; 这里取保守小带并写明)。
NEAR_BAND = 0.005      # 现价距支撑/压力 0.5% 以内算「贴近」
OFF_HIGH_BAND = 0.003  # 距当日高点回落 0.3% 以上算「滞涨」(反T 用)


# ================================================================ 取数小工具
def _f(v, d=0.0):
    try:
        return float(v)
    except (TypeError, ValueError):
        return d


def _pos_of(view: dict, code: str) -> dict:
    for x in view.get("positions") or []:
        if x.get("ts_code") == code:
            return x
    return {}


def _round_lot(qty) -> int:
    """向下取整到一手 (100 股)。不足一手返回 0。"""
    n = int(_f(qty))
    return (n // LOT) * LOT


def _today() -> int:
    return td.ymd()


def _buypause_codes() -> set:
    """当前被暂停买入的 ts_code 集合 (决策系统对策略票的风控预警触发)。读失败按空集。"""
    try:
        from app.services import strategy_service
        return set(strategy_service.buypause_map().keys())
    except Exception as e:  # noqa: BLE001
        logger.warning("[strategy] 读取买入暂停集失败 (按空集): %s", e)
        return set()


def _macro_gate_active() -> bool:
    """宏观偏热闸 (MACRO_TIMING_PLAN.md §5.2): 生效时策略的买入开腿一并暂停。

    与 buypause 走同一个判定点 (ctx.buy_paused), 天然继承「只挡开新仓/加仓腿,
    卖出/平回/跟踪止盈照常」的既有语义 —— 反T 的买回是平回腿, 不会被拦。
    读不到按不生效: 闸的安全方向是不额外拦。"""
    try:
        from app.services import macro_service
        return bool((macro_service.gate_state() or {}).get("active"))
    except Exception as e:  # noqa: BLE001
        logger.warning("[strategy] 读宏观闸状态失败 (按不生效): %s", e)
        return False


# ================================================================ 指令 / 提议下发
def _emit_instruction(st: dict, dec: dict, *, forced: bool = False) -> str:
    """按决策发一张短窗口命令指令 (window_tdays=1, is_command=True), 交 run_tick 执行。

    limit_price 不在这里定 —— 交由 run_tick 的择时按实时行情现算 (与 materialize_plans
    同口径: 那里也是 limit_price=None)。
    """
    code = st["ts_code"]
    now = datetime.now()
    iid = f"STR{_today()}{now.strftime('%H%M%S')}_{code.replace('.', '')}_{dec['leg'][:1]}"[:40]
    is_cmd = (st.get("type") == "T0")   # 做T必须当日轧平→命令口径(含14:45强制平回);
    #                                     网格/跟踪止盈到价即成交、当日没成交则作废, 不强制
    prog = {"is_command": is_cmd, "deadline": str(_today()), "children": [],
            "origin": "strategy", "strategy_id": st["strategy_id"], "stype": st.get("type"),
            "leg": dec["leg"], "reason": dec.get("reason"), "forced": bool(forced)}
    pms_repo.insert_instruction(
        instruction_id=iid, origin_type="strategy", origin_id=st["strategy_id"],
        ts_code=code, action=dec["action"], side=dec["side"], qty=int(dec["qty"]),
        limit_price=None, window_tdays=1, status="PROPOSED", progress=prog)
    return iid


def _emit_proposal(st: dict, dec: dict) -> str:
    """autonomy=confirm: 落一条提议进「等我拍板」。hard_numbers 里带完整动作参数, 人采纳后由
    main._decide 识别 kind=strategy 再回调本层发指令 (不走通用物化, 因为 T0 买卖同 action、
    side 无法由 action 反推)。"""
    code = st["ts_code"]
    now = datetime.now()
    pid = f"STP{_today()}{now.strftime('%H%M%S')}_{code.replace('.', '')}"[:40]
    hard = {"kind": "strategy", "strategy_id": st["strategy_id"], "stype": st.get("type"),
            "side": dec["side"], "leg": dec["leg"], "action": dec["action"],
            "qty": int(dec["qty"]), "reason": dec.get("reason"), "entry": dec.get("entry")}
    expire = now.replace(hour=15, minute=0, second=0, microsecond=0)
    pms_repo.insert_proposal(proposal_id=pid, ts_code=code, action=dec["action"],
                             qty=int(dec["qty"]), hard_numbers=hard, expire_at=expire,
                             judge_verdict="STRATEGY",
                             judge_reason=f"{st.get('type')} · {dec.get('reason')}")
    return pid


def emit_from_spec(spec: dict) -> dict:
    """confirm 提议被采纳后的回调 (main._decide 调): 按 hard_numbers 里存的动作参数发指令。
    返回 {ok, instruction_id} 或 {ok:false, error}。"""
    try:
        st = pms_repo.get_strategy(spec.get("strategy_id"))
        if not st:
            return {"ok": False, "error": "策略不存在或已撤下"}
        if st.get("status") != "ACTIVE":
            return {"ok": False, "error": f"策略处于 {st.get('status')}, 不再发指令"}
        dec = {"side": spec["side"], "action": spec["action"], "qty": int(spec["qty"]),
               "leg": spec.get("leg") or "open", "reason": spec.get("reason"),
               "entry": spec.get("entry")}
        iid = _emit_instruction(st, dec)
        state = dict(st.get("state") or {})
        state["pending"] = {"iid": iid, "leg": dec["leg"], "dir": spec.get("dir"),
                            "qty": dec["qty"], "entry": dec.get("entry")}
        pms_repo.update_strategy(st["strategy_id"], state=state)
        return {"ok": True, "instruction_id": iid}
    except Exception as e:  # noqa: BLE001
        logger.exception("[strategy] 采纳提议发指令失败")
        return {"ok": False, "error": f"{type(e).__name__}: {e}"}


# ================================================================ 在途委托的收敛
def _pending_terminal(pending: dict):
    """上一笔委托是否已终态。返回 (已终态?, 指令行 or None)。无 pending 视为已终态。"""
    if not pending:
        return True, None
    iid = pending.get("iid")
    if not iid:                       # confirm 提议还没被采纳 —— 看提议是否还在等
        pid = pending.get("pid")
        if not pid:
            return True, None
        pr = pms_repo.get_proposal(pid)
        if pr and pr.get("status") == "WAIT_USER":
            return False, None        # 还在等人拍板, 先别再发
        return True, None             # 已采纳(转指令,另有pending.iid)/驳回/过期
    ins = pms_repo.get_instruction(iid)
    if not ins:
        return True, None
    return (ins.get("status") not in LIVE_INS), ins


def _reconcile(st: dict, state: dict, pos: dict):
    """把已终态的上一笔并进 state: 开仓成交 → 记 open_leg; 平回成交 → 记一次完成、清 open_leg。

    做T 用; 网格 / 跟踪止盈 的 filled_levels / high_water 在各自评估器里按现价推进, 不依赖这里。
    """
    pending = state.get("pending")
    done, ins = _pending_terminal(pending)
    if not done:
        return False                   # 上一笔未走完
    state["pending"] = None
    if not ins or not pending or not pending.get("iid"):
        return True
    filled = int(ins.get("exec_qty") or 0) > 0
    leg = pending.get("leg")
    if st.get("type") == "T0":
        if leg == "open" and filled:
            state["open_leg"] = {"dir": pending.get("dir"), "qty": int(ins.get("exec_qty") or 0),
                                 "entry": _f(pending.get("entry")) or _f(ins.get("limit_price")),
                                 "opened_at": ins.get("updated_at") and str(ins["updated_at"])}
        elif leg == "close":
            # 平回终态: 只有真成交才算一轮完成; 一股没成 (到期/被拒) 要**保留 open_leg**,
            # 让下一跳与 14:50 平回继续补平 —— 绝不能把一条还开着的 T 仓静默丢掉。
            filled_qty = int(ins.get("exec_qty") or 0)
            ol = dict(state.get("open_leg") or {})
            rem = _round_lot(_f(ol.get("qty")) - filled_qty)
            if rem >= LOT:
                ol["qty"] = rem
                state["open_leg"] = ol
                if filled_qty > 0:
                    logger.info("[strategy] %s 平回部分成交 %s, 余 %s 股待续平",
                                st.get("strategy_id"), filled_qty, rem)
            else:
                state["t_count_today"] = int(state.get("t_count_today") or 0) + 1
                state["open_leg"] = None
    return True


# ================================================================ 评估器: 做T (设计 §六)
def _eval_t0(st, pos, day, now, ctx):
    """做T: 正T (回落近支撑 → 买, 目标价差高卖) / 反T (近压力或滞涨 → 卖, 低买回)。

    rails: 当日 ≤ 3 次; 单票 / 全局当日T亏熔断后当日禁开新T (仍允许平回已开的仓);
    买入暂停 (决策系统风控预警) 时同样禁开新T, 但平回照常; 14:50 强制平回由 force_t0_close 走。
    """
    state = ctx["state"]
    price = _f(day.get("price"))
    if price <= 0:
        return None
    total = int(pos.get("total_qty") or 0)
    avail = int(pos.get("avail_qty") or 0)
    params = st.get("params") or {}
    t_ratio = min(_f(params.get("t_ratio")), param_store.get_float("PMS_T0_RATIO_MAX", 0.333))
    t_qty = _round_lot(total * t_ratio)

    open_leg = state.get("open_leg")
    round_target = param_store.get_float("PMS_T0_ROUND_TARGET", 0.015)

    # ---- 有未平的仓 → 只找平回机会 (熔断 / 买入暂停都不挡平回) ----
    if open_leg:
        entry = _f(open_leg.get("entry")) or price
        q = _round_lot(open_leg.get("qty"))
        if open_leg.get("dir") == "long":       # 正T: 已买, 等高卖
            pressure = _f(pos.get("pressure_ref"))
            hit = price >= entry * (1 + round_target) or (pressure > 0 and price >= pressure * (1 - NEAR_BAND))
            if hit and q > 0:
                return {"side": "sell", "action": A_T0, "qty": min(q, avail), "leg": "close",
                        "reason": f"正T平回: 现价 {price} 达目标 {entry * (1 + round_target):.2f}(买价 {entry})"}
        else:                                    # 反T: 已卖, 等低买回
            support = _f(pos.get("support_ref"))
            hit = price <= entry * (1 - round_target) or (support > 0 and price <= support * (1 + NEAR_BAND))
            if hit and q > 0:
                return {"side": "buy", "action": A_T0, "qty": q, "leg": "close",
                        "reason": f"反T平回: 现价 {price} 回到目标 {entry * (1 - round_target):.2f}(卖价 {entry})"}
        return None

    # ---- 无未平的仓 → 看要不要开新的一轮 (受熔断 / 买入暂停 / 3 次 / 存量约束) ----
    if ctx.get("halted") or ctx.get("buy_paused") or int(state.get("t_count_today") or 0) >= 3:
        return None
    if t_qty < LOT:
        return None
    high = _f(day.get("high"))
    support = _f(pos.get("support_ref"))
    pressure = _f(pos.get("pressure_ref"))
    pull = param_store.get_float("PMS_T0_PULLBACK_PCT", 0.03)
    rally = param_store.get_float("PMS_T0_RALLY_PCT", 0.05)
    dayup = _f(day.get("day_chg_from_open"))

    # 正T: 距当日高点回落 ≥ 回落阈 且 (近支撑 或 无支撑参照时仅凭回落, 写明)
    if high > 0 and price <= high * (1 - pull):
        near_sup = support > 0 and price <= support * (1 + NEAR_BAND)
        if near_sup or support <= 0:
            note = "近支撑" if near_sup else "无支撑参照, 仅凭回落(偏保守)"
            return {"side": "buy", "action": A_T0, "qty": t_qty, "leg": "open", "dir": "long",
                    "entry": price,
                    "reason": f"正T开仓: 距高点 {high} 回落 {1 - price / high:.1%} 且{note}, 买 {t_qty} 股"}

    # 反T: 近压力 或 (日内涨 ≥ 反T阈 且 已从高点滞涨) —— 卖 avail 的 t_qty, 待低买回
    near_pre = pressure > 0 and price >= pressure * (1 - NEAR_BAND)
    stalled = dayup >= rally and high > 0 and price <= high * (1 - OFF_HIGH_BAND)
    if (near_pre or stalled) and avail >= LOT:
        q = min(t_qty, _round_lot(avail))
        if q >= LOT:
            why = "近压力" if near_pre else f"日内涨 {dayup:.1%} 滞涨"
            return {"side": "sell", "action": A_T0, "qty": q, "leg": "open", "dir": "short",
                    "entry": price, "reason": f"反T开仓: {why}, 卖 {q} 股待低买回"}
    return None


# ================================================================ 评估器: 网格 (设计 §七)
def _grid_levels(params: dict) -> list:
    """按 中枢/档距/上下界 生成一组网格价位 (由低到高)。档距支持百分比(step_pct)或绝对值(step)。"""
    lo, hi = _f(params.get("lower")), _f(params.get("upper"))
    center = _f(params.get("center"))
    step = _f(params.get("step"))
    step_pct = _f(params.get("step_pct"))
    if lo <= 0 or hi <= lo:
        return []
    base = center if center > 0 else (lo + hi) / 2
    if step <= 0 and step_pct > 0:
        step = base * step_pct
    if step <= 0:
        return []
    levels, p, guard = [], lo, 0
    while p <= hi + 1e-9 and guard < 200:
        levels.append(round(p, 3))
        p += step
        guard += 1
    return levels


def _band(levels: list, price: float) -> int:
    """现价所处的档位下标: 满足 levels[i] <= price 的最大 i; 低于最低档返回 -1。"""
    b = -1
    for i, lv in enumerate(levels):
        if lv <= price:
            b = i
        else:
            break
    return b


def _eval_grid(st, pos, day, now, ctx):
    """网格(逐档穿越): 价每向下跌破一个新档买一手(只买中枢下方), 向上涨破一个档就把下面对应
    档买的那手卖掉(从 avail)。一跳只走一档; 跌破下界=继续持有不再买; 买入暂停只停买、卖出照常;
    filled_levels 每跳与真实 ADD 持仓对账, 被对账冲销后收敛 —— 绝不卖幻影档(误卖底仓)。"""
    state = ctx["state"]
    params = st.get("params") or {}
    price = _f(day.get("price"))
    if price <= 0:
        return None
    levels = _grid_levels(params)
    if not levels:
        return None
    lo = levels[0]
    per_lot = _round_lot(params.get("per_lot")) or LOT
    max_capital = _f(params.get("max_capital"))
    filled = {int(k): dict(v) for k, v in (state.get("filled_levels") or {}).items()}
    invested = _f(state.get("invested"))
    avail = int(pos.get("avail_qty") or 0)
    actual_add = int(pos.get("add_qty") or 0)   # 网格买入记 ADD 批次, 这是真实网格持仓

    # —— 对账收敛: filled 声称的网格股 > 真实(被 RECON 冲销) → 收敛; 真实为0则清空重来 ——
    claimed = sum(int(v.get("qty") or 0) for v in filled.values())
    if claimed > actual_add:
        if actual_add <= 0:
            if filled:
                ctx["notes"].append(f"{st['ts_code']} 网格持仓已被对账冲销(真实网格股0), 清空网格档位重来")
            filled, invested = {}, 0.0
            state["last_band"] = None
        else:
            for k in sorted(filled.keys()):
                if claimed <= actual_add:
                    break
                q = int(filled[k].get("qty") or 0)
                invested = max(0.0, invested - _f(filled[k].get("price")) * q)
                claimed -= q
                del filled[k]
            ctx["notes"].append(f"{st['ts_code']} 网格档位与真实持仓对齐(真实网格股 {actual_add})")
        state["filled_levels"] = {str(k): v for k, v in filled.items()}
        state["invested"] = invested

    cur = _band(levels, price)
    last = state.get("last_band")

    # 跌破下界: 停买、保留已买、告警; 记基准档但不交易
    if price < lo:
        if not state.get("below_floor"):
            state["below_floor"] = True
            ctx["notes"].append(f"{st['ts_code']} 跌破网格下界 {lo}, 已停止网格买入(继续持有已买档)")
        state["last_band"] = cur
        return None
    state["below_floor"] = False

    # 首跳: 只记基准档, 不交易 (等价格真正穿越档位才动)
    if last is None:
        state["last_band"] = cur
        return None

    # —— 上行: 价涨破 → 卖掉离开的这一档买的那手 (从 avail, 且确有网格股, 才卖) ——
    if cur > last:
        k = last
        if k in filled and avail >= LOT and actual_add >= LOT:
            q = min(per_lot, _round_lot(avail))
            if q >= LOT:
                state["last_band"] = last + 1
                info = filled[k]
                return {"side": "sell", "action": A_SELL, "qty": q, "leg": f"grid_sell:{k}",
                        "grid_sell_idx": k,
                        "reason": f"网格卖: 现价 {price} 涨破档{k}(买价 {info.get('price')}), 卖 {q} 股"}
            return None                          # 有档可卖但量不足, 先不推进, 下跳再试
        state["last_band"] = last + 1            # 该档无网格持仓可卖, 只随价上移
        return None

    # —— 下行: 价跌破新档 → 买这一档 (只买中枢下方; 受下界/暂停/上限约束) ——
    if cur < last:
        k = last - 1                             # 刚跌破的这一档
        if k < 0:
            state["last_band"] = cur
            return None
        if k in filled:
            state["last_band"] = last - 1        # 已买, 只推进
            return None
        if ctx.get("buy_paused"):
            return None                          # 暂停买入: 原地等, 不推进
        need = per_lot * price
        if max_capital > 0 and invested + need > max_capital + 1e-6:
            if not state.get("cap_hit"):
                state["cap_hit"] = True
                ctx["notes"].append(f"{st['ts_code']} 网格已达最大投入 {max_capital:.0f} 元, 暂停买入")
            return None                          # 触顶: 不推进, 下次重试
        state["cap_hit"] = False
        state["last_band"] = last - 1
        return {"side": "buy", "action": A_GRID_BUY, "qty": per_lot, "leg": f"grid_buy:{k}",
                "grid_buy_idx": k, "grid_buy_price": price,
                "reason": f"网格买: 现价 {price} 跌破档{k}(档价 {levels[k]}), 买 {per_lot} 股"}

    return None                                  # cur == last, 同档不动


# ================================================================ 评估器: 跟踪止盈 (设计 §七B)
def _eval_trail(st, pos, day, now, ctx):
    """跟踪止盈: 创新高抬止盈线, 从高点回落 ≥ giveback 就卖 (从 avail); 命中硬止盈目标直接全清。
    只卖不买 —— 纯离场保护, 不受买入暂停影响。"""
    state = ctx["state"]
    params = st.get("params") or {}
    price = _f(day.get("price"))
    if price <= 0:
        return None
    avg = _f(pos.get("avg_cost"))
    avail = int(pos.get("avail_qty") or 0)
    profit = _f(pos.get("cushion_pct")) if pos.get("cushion_pct") is not None else (
        (price / avg - 1) if avg > 0 else 0.0)

    start_line = _f(params.get("start_line")) or param_store.get_float("PMS_CUSHION_SOLID", 0.03)
    giveback = _f(params.get("giveback"))
    sell_ratio = _f(params.get("sell_ratio")) or 1.0
    hard_target = _f(params.get("hard_target"))

    # 高水位每跳只抬不降
    hw = max(_f(state.get("high_water")), price)
    state["high_water"] = round(hw, 3)
    if not state.get("armed") and profit >= start_line:
        state["armed"] = True
        ctx["notes"].append(f"{st['ts_code']} 跟踪止盈已启动(浮盈 {profit:.1%} ≥ 启动线 {start_line:.1%})")

    if avail < LOT:
        return None                     # 无 T+1 可卖, 只更新高水位

    # 硬止盈目标: 直接全清
    if hard_target > 0 and profit >= hard_target:
        return {"side": "sell", "action": A_EXIT, "qty": _round_lot(avail) or avail, "leg": "trail_hard",
                "reason": f"跟踪止盈-硬目标: 浮盈 {profit:.1%} ≥ {hard_target:.1%}, 全清 avail {avail} 股"}

    # 已武装且从高点回落到设定比例 → 卖
    if state.get("armed") and hw > 0 and price <= hw * (1 - giveback) and giveback > 0:
        q = _round_lot(avail * sell_ratio) if sell_ratio < 1 else (_round_lot(avail) or avail)
        if q >= LOT or (sell_ratio >= 1 and q > 0):
            act = A_EXIT if sell_ratio >= 1 else A_SELL
            return {"side": "sell", "action": act, "qty": q, "leg": "trail_sell",
                    "reason": f"跟踪止盈: 现价 {price} 自高点 {hw} 回落 {1 - price / hw:.1%} ≥ {giveback:.1%}, 卖 {q} 股"}
    return None


EVALUATORS = {"T0": _eval_t0, "GRID": _eval_grid, "TRAIL": _eval_trail}


# ================================================================ 供 action_engine 排除
def active_codes() -> set:
    """有 ACTIVE 策略的 ts_code —— 供 action_engine 排除。读库失败按空集 (不误排除全体持仓)。"""
    try:
        return pms_repo.active_strategy_codes()
    except Exception as e:  # noqa: BLE001
        logger.warning("[strategy] 读取 ACTIVE 策略集失败 (按空集): %s", e)
        return set()


# ================================================================ 每分钟主跳
def _roll_day(state: dict, pos: dict, today: int) -> dict:
    """做T 的按日重置: 当日次数归零, 记下当日 realized_t_profit 基线 (算当日T盈亏用)。
    网格 filled_levels / 跟踪止盈 high_water 是跨日的, 不在这里动。"""
    if int(state.get("day") or 0) != today:
        state["day"] = today
        state["t_count_today"] = 0
        state["rt_base"] = _f(pos.get("realized_t_profit"))
        # 隔夜后原则上不该留未平的仓 (14:50 已平回); 万一留了, 清掉 open_leg 交对账兜底
        state["open_leg"] = None
    return state


def _day_t_pnl(state: dict, pos: dict) -> float:
    return _f(pos.get("realized_t_profit")) - _f(state.get("rt_base"))


def tick(*, now=None, dry_run: bool = False) -> dict:
    """盘中每分钟一跳: 载入 ACTIVE 策略 → 逐只评估 → 触发就发短窗口指令 / 落提议 → 更新 state。

    dry_run=True 只算不发不落库 (页面「试算」用)。
    """
    now = now or datetime.now()
    out = {"enabled": False, "checked": 0, "fired": [], "queued": [], "skipped": [],
           "notes": [], "errors": [], "dry_run": dry_run}
    if not param_store.get_bool("PMS_STRATEGY_ENABLED", False):
        out["skipped"].append("PMS_STRATEGY_ENABLED=False, 策略层整体停用")
        return out
    out["enabled"] = True

    try:
        strategies = pms_repo.active_strategies()
    except Exception as e:  # noqa: BLE001
        logger.exception("[strategy] 载入 ACTIVE 策略失败")
        return {**out, "ok": False, "errors": [f"载入失败: {type(e).__name__}: {e}"]}
    if not strategies:
        out["ok"] = True
        return out

    try:
        view = portfolio.positions_view()
    except Exception as e:  # noqa: BLE001
        logger.exception("[strategy] 取持仓快照失败")
        return {**out, "ok": False, "errors": [f"取持仓失败: {type(e).__name__}: {e}"]}
    scale = _f(view.get("totals", {}).get("scale"))
    today = _today()
    paused_codes = _buypause_codes()   # 决策系统风控预警暂停买入的票 (本模块只读)
    macro_gated = _macro_gate_active()  # 宏观偏热闸: 全体策略买开腿暂停, 闸解除自动恢复
    if macro_gated:
        out["notes"].append("宏观偏热闸生效: 策略买入开腿本轮暂停 (卖出与平回照常)")

    # 全局当日T亏熔断: 汇总所有做T策略的当日T盈亏 (realized_t_profit 相对日初基线的增量)
    global_t_pnl = 0.0
    for st in strategies:
        if st.get("type") == "T0":
            global_t_pnl += _day_t_pnl(st.get("state") or {}, _pos_of(view, st["ts_code"]))
    global_loss_cap = param_store.get_float("PMS_T0_GLOBAL_DAY_LOSS", 0.01) * scale
    global_halt = scale > 0 and global_t_pnl <= -global_loss_cap
    if global_halt:
        out["notes"].append(f"全局当日T亏 {global_t_pnl:.0f} 元 达熔断线 {global_loss_cap:.0f} 元, 今日不再开新T")

    for st in strategies:
        out["checked"] += 1
        code = st.get("ts_code")
        fn = EVALUATORS.get(st.get("type"))
        if not fn:
            out["skipped"].append({"strategy_id": st.get("strategy_id"),
                                   "why": f"未知策略类型 {st.get('type')}"})
            continue
        try:
            pos = _pos_of(view, code)
            if not pos or int(pos.get("total_qty") or 0) <= 0:
                out["skipped"].append({"strategy_id": st.get("strategy_id"),
                                       "why": f"{code} 已无持仓, 策略空转 (可撤下)"})
                continue
            state = dict(st.get("state") or {})
            if st.get("type") == "T0":
                state = _roll_day(state, pos, today)

            # 上一笔还没走完就别再发; 走完了先把结果并进 state
            if not _reconcile(st, state, pos):
                if not dry_run:
                    pms_repo.update_strategy(st["strategy_id"], state=state)
                out["skipped"].append({"strategy_id": st["strategy_id"], "why": "上一笔还在途, 等它走完"})
                continue

            day = market.day_snapshot(code)
            if not day or not day.get("price"):
                out["skipped"].append({"strategy_id": st["strategy_id"],
                                       "why": f"{code} 无实时行情 (停牌/盘前), 顺延"})
                if not dry_run:
                    pms_repo.update_strategy(st["strategy_id"], state=state)
                continue

            # 单票当日T亏熔断 (做T)
            halted = global_halt
            if st.get("type") == "T0" and scale > 0:
                stock_cap = param_store.get_float("PMS_T0_STOCK_DAY_LOSS", 0.003) * scale
                if _day_t_pnl(state, pos) <= -stock_cap:
                    halted = True
                    out["notes"].append(f"{code} 当日T亏达单票熔断线 {stock_cap:.0f} 元, 今日不再开新T")

            ctx = {"state": state, "scale": scale, "halted": halted, "notes": out["notes"],
                   "buy_paused": (code in paused_codes) or macro_gated}
            dec = fn(st, pos, day, now, ctx)

            if dec:
                if not dry_run:
                    _apply_grid_state(state, dec)
                    if (st.get("autonomy") or "auto") == "confirm":
                        pid = _emit_proposal(st, dec)
                        state["pending"] = {"pid": pid, "leg": dec["leg"], "dir": dec.get("dir"),
                                            "qty": dec["qty"], "entry": dec.get("entry")}
                        out["queued"].append({"strategy_id": st["strategy_id"], "proposal_id": pid,
                                              "reason": dec["reason"]})
                    else:
                        iid = _emit_instruction(st, dec, forced=False)
                        state["pending"] = {"iid": iid, "leg": dec["leg"], "dir": dec.get("dir"),
                                            "qty": dec["qty"], "entry": dec.get("entry")}
                        out["fired"].append({"strategy_id": st["strategy_id"], "instruction_id": iid,
                                            "side": dec["side"], "qty": dec["qty"], "reason": dec["reason"]})
                else:
                    out["fired"].append({"strategy_id": st["strategy_id"], "dry_run": True,
                                        "side": dec["side"], "qty": dec["qty"], "reason": dec["reason"]})

            if not dry_run:
                pms_repo.update_strategy(st["strategy_id"], state=state)
        except Exception as e:  # noqa: BLE001 —— 单策略异常不拖垮整轮
            logger.exception("[strategy] 评估失败 %s", st.get("strategy_id"))
            out["errors"].append(f"{st.get('strategy_id')}: {type(e).__name__}: {e}")

    out["ok"] = not out["errors"]
    return out


def _apply_grid_state(state: dict, dec: dict):
    """网格发单同时更新 filled_levels / invested (买入占用一档, 卖出释放一档)。"""
    if "grid_buy_idx" in dec:
        filled = dict(state.get("filled_levels") or {})
        filled[str(dec["grid_buy_idx"])] = {"price": _f(dec.get("grid_buy_price")),
                                            "qty": int(dec["qty"])}
        state["filled_levels"] = filled
        state["invested"] = _f(state.get("invested")) + _f(dec.get("grid_buy_price")) * int(dec["qty"])
    elif "grid_sell_idx" in dec:
        filled = dict(state.get("filled_levels") or {})
        info = filled.pop(str(dec["grid_sell_idx"]), None)
        state["filled_levels"] = filled
        if info:
            state["invested"] = max(0.0, _f(state.get("invested")) - _f(info.get("price")) * int(info.get("qty") or 0))


# ================================================================ 14:50 强制平回 (设计 §六 rails)
def force_t0_close(*, now=None) -> dict:
    """做T 强制平回 (scheduler.t0_close 在 PMS_T0_CLOSE_TIME 调): 对每只有未平仓的做T策略,
    立刻发反方向的平回委托把当日T仓打平, 绝不过夜。已在途的不重复 —— 这里只补「还没平」的。

    平回不受买入暂停影响: 暂停挡的是开新仓, 平回是把已开的打平, 必须放行 (反T 平回是买回)。"""
    out = {"closed": [], "skipped": [], "errors": []}
    if not param_store.get_bool("PMS_STRATEGY_ENABLED", False):
        out["skipped"].append("PMS_STRATEGY_ENABLED=False")
        return out
    try:
        strategies = [s for s in pms_repo.active_strategies() if s.get("type") == "T0"]
    except Exception as e:  # noqa: BLE001
        return {**out, "errors": [f"载入做T策略失败: {type(e).__name__}: {e}"]}
    if not strategies:
        return out
    try:
        view = portfolio.positions_view()
    except Exception as e:  # noqa: BLE001
        return {**out, "errors": [f"取持仓失败: {type(e).__name__}: {e}"]}

    for st in strategies:
        code = st["ts_code"]
        try:
            state = dict(st.get("state") or {})
            pos = _pos_of(view, code)
            _reconcile(st, state, pos)
            open_leg = state.get("open_leg")
            if not open_leg:
                out["skipped"].append({"strategy_id": st["strategy_id"], "why": "无未平仓"})
                pms_repo.update_strategy(st["strategy_id"], state=state)
                continue
            if state.get("pending"):
                out["skipped"].append({"strategy_id": st["strategy_id"], "why": "平回委托已在途"})
                continue
            q = _round_lot(open_leg.get("qty"))
            if q < LOT:
                out["skipped"].append({"strategy_id": st["strategy_id"], "why": "残量不足一手"})
                continue
            if open_leg.get("dir") == "long":       # 正T 已买 → 卖平
                dec = {"side": "sell", "action": A_T0, "qty": min(q, int(pos.get("avail_qty") or 0)),
                       "leg": "close", "reason": "14:50 强制平回(正T)"}
            else:                                    # 反T 已卖 → 买平
                dec = {"side": "buy", "action": A_T0, "qty": q, "leg": "close",
                       "reason": "14:50 强制平回(反T)"}
            if int(dec["qty"]) < LOT:
                out["skipped"].append({"strategy_id": st["strategy_id"], "why": "可平量不足 (avail 不够)"})
                continue
            iid = _emit_instruction(st, dec, forced=True)
            state["pending"] = {"iid": iid, "leg": "close", "dir": open_leg.get("dir"),
                                "qty": dec["qty"], "entry": open_leg.get("entry")}
            pms_repo.update_strategy(st["strategy_id"], state=state)
            out["closed"].append({"strategy_id": st["strategy_id"], "instruction_id": iid,
                                  "side": dec["side"], "qty": dec["qty"]})
        except Exception as e:  # noqa: BLE001
            logger.exception("[strategy] 强制平回失败 %s", st.get("strategy_id"))
            out["errors"].append(f"{st.get('strategy_id')}: {type(e).__name__}: {e}")
    out["ok"] = not out["errors"]
    return out