@zhin.js/plugin-lottery
v1.0.7
Published
Multi-game lottery analysis for Zhin.js (Plugin Runtime)
Maintainers
Readme
@zhin.js/plugin-lottery
多玩法彩票分析插件:快乐8、双色球、大乐透、福彩3D、排列3、排列5。
功能
- 单条定时流水线:同步 → 复盘 → 推荐 → 推送(默认 18:00)
- 频率 + 遗漏 + 趋势综合统计;开奖后自动调权
- 按
games配置过滤,经 RuntimeOutboundHost推送到显式pushTargets - 手动触发:
lottery(不推送,结果在会话内返回) - IM commands 通过 owner-scoped
LotteryRuntime使用实例 DB。Agent deps 与 Outbound sender 使用 generation-owned 注册;配置关闭推送时会显式遮蔽旧 generation,回滚时再 恢复仍存活的上一代。
安装
pnpm add @zhin.js/plugin-lottery流水线
scheduleCron (18:00)
1. sync official draws → lottery_draws
2. review pending picks vs new issues → tune weights → **lottery_model_weights**
3. if no new review: replay history simulation → refresh **lottery_model_weights**
4. compute next-period picks (reads weights from DB) → lottery_predictions
5. push report to configured targets (enabled games only)权重表 lottery_model_weights 每玩法一行,字段 freq_weight / omit_weight / trend_weight + eval_count / avg_hit_rate:
有新开奖且复盘命中:
submitAgentReview增量调权并写库本期未复盘:用历史 walk-forward 模拟刷新权重后写库(
weightPersistEnabled)推荐:
loadGameWeights读库中最新权重算号代码层:官网同步、F/O/T 引擎、复盘调权
确定性报告:主流水线不依赖 legacy Plugin 或 LLM,Agent 可通过独立 tools 读取结果
配置
plugins:
lottery:
scheduleCron: "0 0 18 * * *"
scheduleEnabled: true
backtestEnabled: true
backtestWindow: 50
backtestRandomTrials: 64
backtestMinHistory: 30
pickCount: 5
historyLimit: 500
kl8:
pickCount: 5
recommendGroups: 3
groupStrategies: [balanced, hot, cold]
games: [kl8, ssq, dlt, fc3d, pl3, pl5]
pushTargets:
- adapter: sandbox
endpointId: sandbox-bot
channelType: private
channelId: operator权重训练(模拟盘)
lottery-train 对库内全部历史做 walk-forward 训练(无需下注):
- 从默认权重 F40/O35/T25 起(第
minHistory期后才有足够历史预测) - 每期:用当时权重算号 → 与库中真实开奖比对 →
tuneWeightsFromDraw调权 - 直至往期跑完 → 终权重写入
lottery_model_weights
日常 lottery 在未复盘时会用同一套全量训练刷新权重;lottery-backtest 仅抽样近 N 期做报告。
快乐8 多组推荐
同一期可输出多行推荐,每组使用不同 F/O/T 策略(均衡 / 热号 / 冷号 / 趋势);组间号码允许重复,各组独立取该策略下 Top-N。复盘仍以**组1(均衡)**写入 lottery_predictions。
kl8:
pickCount: 5 # 每组选几个号(选5)
recommendGroups: 3 # 出几组(1~10)
groupStrategies:
- balanced
- hot
- cold回测可信度
每日推荐会自动附带 walk-forward 逐期模拟:从默认 F/O/T 权重起,对每一期依次「预测 → 与开奖比对 → 调权」,在无需下注的情况下热身;近 N 期计入评分,并与固定权重、随机基线对比。
backtestAdaptive: true # 逐期调权模拟(默认开)
weightPersistEnabled: true # 模拟/复盘后写入 lottery_model_weights
weightHoldoutFallback: true # 近端 holdout 输随机 → 推荐用默认权重Commands
| Command | Description |
|---------|-------------|
| lottery [game] | Manual full pipeline (no push) |
| lottery-train [game] | Full-history weight training (DEFAULT → tune → save DB) |
| lottery-backtest [game] | Walk-forward backtest vs random baseline |
| lottery-today | Today’s published report |
| lottery-stats <game> | Single-game stats snapshot |
| lottery-history <game> [count] | Historical draws |
免责声明
本插件推荐仅供参考,不构成投注建议,不保证中奖。请理性购彩。
