孩子一学就喊累？用胜任感（心效力）构建可工程化的正反馈模型

# slice.py —— 把大任务切到「一定能起步」的颗粒度
from dataclasses import dataclass
from datetime import date
from enum import Enum

class Tracker(str, Enum):
    DIRECTING  = "心航力"     # 方向：知不知道该做什么
    FOCUS      = "心定力"     # 专注：能不能待在任务上
    RESILIENCE = "心韧力"     # 抗挫：卡住之后多久能回来
    EFFICACY   = "心效力"     # 胜任：信不信自己能做成

@dataclass
class TaskUnit:
    child_id: str
    day: date
    name: str                 # 单元名
    tracker: Tracker          # 这个单元主要练哪个力
    est_minutes: int          # 预估用时（先由孩子自己估）
    can_start: bool = False   # 颗粒度自检：是否小到「一定能起步」
    started: bool = False     # 行为事实：孩子当天真的动手了没
    done: bool = False

def slice_task(child_id, day, raw, est_minutes,
               tracker=Tracker.EFFICACY, max_unit=15):
    """按预估时长把大任务切成最小可完成单元。

    规则：单个单元预估 ≤ max_unit 分钟；切不动时宁粗勿碎，不再硬拆。
    """
    if est_minutes <= max_unit:
        return [TaskUnit(child_id, day, raw, tracker, est_minutes)]
    n = max(2, -(-est_minutes // max_unit))      # 向上取整，至少切 2 段
    base, extra = divmod(est_minutes, n)
    units = []
    for i in range(n):
        minutes = base + (1 if i < extra else 0)
        units.append(
            TaskUnit(child_id, day, f"{raw}（第 {i + 1}/{n} 步）", tracker, minutes))
    return units

def mark_startable(unit: TaskUnit, first_step: str) -> TaskUnit:
    """「一定能起步」自检：单元预估 ≤15 分钟，且孩子能说清第一步做什么。"""
    unit.can_start = unit.est_minutes <= 15 and bool(first_step.strip())
    return unit

# evidence.py —— 只记「做成」的行为事实，不记表扬
import sqlite3

SCHEMA = """
CREATE TABLE IF NOT EXISTS task_units(
  child_id TEXT, day TEXT, name TEXT, tracker TEXT,
  est_minutes INTEGER, can_start INTEGER, started INTEGER, done INTEGER
);
CREATE TABLE IF NOT EXISTS evidence(
  child_id TEXT, day TEXT, unit_name TEXT, tracker TEXT, created_at TEXT
);
"""

def init(conn):
    conn.executescript(SCHEMA)
    conn.commit()

def save_unit(conn, unit, created_at):
    conn.execute(
        "INSERT INTO task_units VALUES(?,?,?,?,?,?,?,?)",
        (unit.child_id, unit.day.isoformat(), unit.name, unit.tracker.value,
         unit.est_minutes, int(unit.can_start), int(unit.started), int(unit.done)))
    conn.commit()

def record_evidence(conn, unit, created_at):
    """只在单元完成时落一条「做成」证据——记行为，不记分数，不写表扬语。"""
    if not unit.done:
        return False
    conn.execute(
        "INSERT INTO evidence VALUES(?,?,?,?,?)",
        (unit.child_id, unit.day.isoformat(), unit.name,
         unit.tracker.value, created_at))
    conn.commit()
    return True

# attribution.py —— 把「人身定性」改写为「过程归因」
PERSON_LABELS = ("笨", "傻", "不行", "差劲", "没用", "没救")

def rewrite_attribution(raw_text: str) -> dict:
    """孩子自评一句话 → 受控改写。

    规则：命中人身定性词块，一律不评价人格，只把话头拉回
    「这次是哪一步没准备好」，并落回一个明天能试的小动作。
    没命中（本来就说的过程）就原样保留，不做多余加工。
    """
    hit = [w for w in PERSON_LABELS if w in raw_text]
    if not hit:
        return {"kind": "process", "text": raw_text, "action": ""}
    return {
        "kind": "person",
        "text": "这次是哪一步没准备好？",
        "action": "把今天的单元再拆小一档，明天先做最容易起步的那个",
    }

# loop.py —— 每周把证据回读成三条曲线，形成正反馈回路
import datetime

def moving_avg(values, window=3):
    out = []
    for i in range(len(values)):
        seg = [v for v in values[max(0, i - window + 1): i + 1] if v is not None]
        out.append(round(sum(seg) / len(seg), 3) if seg else None)
    return out

def weekly_loop(conn, child_id: str, day: str, window: int = 3):
    """聚合一周：能起步率、单元完成率、做成证据条数 → 三条曲线 + 本周可见证据。"""
    since = (datetime.date.fromisoformat(day) - datetime.timedelta(days=6)).isoformat()
    rows = conn.execute(
        "SELECT day, SUM(started), SUM(done), COUNT(*) FROM task_units "
        "WHERE child_id=? AND day>=? GROUP BY day ORDER BY day",
        (child_id, since)).fetchall()
    start_rate = [r[1] / r[3] if r[3] else None for r in rows]   # 能起步率
    done_rate = [r[2] / r[3] if r[3] else None for r in rows]    # 单元完成率
    evidence = conn.execute(
        "SELECT day, unit_name FROM evidence WHERE child_id=? AND day>=? "
        "ORDER BY created_at", (child_id, since)).fetchall()
    return {
        "days": [r[0] for r in rows],
        "start_rate": start_rate,
        "done_rate": done_rate,
        "start_rate_ma": moving_avg(start_rate, window),
        "done_rate_ma": moving_avg(done_rate, window),
        "evidence_count": len(evidence),
        "visible_evidence": evidence,
    }