Skip to content
Agentmart
Submit a tool

SK-self-iterate

赛后自我迭代学习——每次得到赛果后把复盘升级为可累积/可验证/可固化的学习(偏差定量→归因→观察池→复发自动提示→门槛回测→固化)。🔴学习≠批量更新模型:学习层仅累积与提示(不改任何计算/判定),批量更新层才动规则。由主技能复盘时必调(Step10 复盘后)+ 每批 status 汇总。

Skill by MENG-COOLMAN

IndexedMIT★ 3
Write a review

Chalk Labs

Need help deploying SK-self-iterate?

Chalk Labs sets up agents like this inside real teams: connected to your tools, tested on your data, with a human in the loop where it matters.

Book a call

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/meng-coolman-sk-self-iterate

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

赛后自我迭代学习——每次得到赛果后把复盘升级为可累积/可验证/可固化的学习(偏差定量→归因→观察池→复发自动提示→门槛回测→固化)。🔴学习≠批量更新模型:学习层仅累积与提示(不改任何计算/判定),批量更新层才动规则。由主技能复盘时必调(Step10 复盘后)+ 每批 status 汇总。

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/sk-self-iterate

What's inside

Skill name
SK-self-iterate
Description
赛后自我迭代学习——每次得到赛果后把复盘升级为可累积/可验证/可固化的学习(偏差定量→归因→观察池→复发自动提示→门槛回测→固化)。🔴学习≠批量更新模型:学习层仅累积与提示(不改任何计算/判定),批量更新层才动规则。由主技能复盘时必调(Step10 复盘后)+ 每批 status 汇总。
License
MIT
Instructions
73 lines in SKILL.md · view on GitHub

Agent Skills spec issues

  • name: only lowercase letters, digits and single hyphens allowed

Files (1)

  • SKILL.md

SKILL.md (first lines)

# SK-self-iterate —— 赛后自我迭代学习闭环(V3.5.74)

## 🔴 定位(防止混淆·用户明确要求)
| 层 | 触发 | 动作 | 是否改模型 |
|:--|:--|:--|:--|
| **学习层(本技能)** | 每次赛果 | 录入台账→归因→累积→复发自动提示→calc_all 自动消费提示 | ❌ 只提示·零计算/判定改动 |
| **批量更新层(另流程)** | 观察项达门槛 | 批量回测(时间分割·p<0.05)→ 验证通过 → 改 AGENTS/技能/数据 + check_sync | ✅ 动规则 |

**铁律**:单场永不直接改规则;学习层只积累与提示;批量更新必须门槛+回测+备份+check_sync 全绿。

## 执行流程

### ① 每场赛果后必跑(复盘的一部分)
```bash
python data/tmp/self_iterate.py learn --case <id> --real <比分> --pred "<预测方向>" \
    --anchor "<主锚>|<次锚>" --center <预测进球中心> --handi <让球> --o25 <O25>
```
输出:方向/锚命中/量级偏差(±球)+ 归因类型(多项并存)+ 复发自动升级提示。

归因类型:`超深屠杀锚量级`/`锚净低估`/`大球低估`/`小球高估`/`净胜穿透`/`平局漏判`/`方向误判`/`命中`。

### ② 每批复盘后汇总
```bash
python data/tmp/self_iterate.py status     # 各类型累积 vs 门槛 + 是否达标
python data/tmp/self_iterate.py patterns   # 偏差模式明细(最近样本)
```

### ③ 自动更新(防同一失误犯第二遍)
- 同类型**复发 ≥2** → 自动升级 `status=提示` + 生成规则文本 → 写入 `data/tmp/learned_rules.json`
- **calc_all 每场自动输出**「⚡ 学习成果提示」节(仅提示·判定权在模型)
- 达门槛(方向类 10 / 比分·量级类 20)→ `status=待回测验证`

### ④ 批量回测验证通过 → 固化
```bash
python data/tmp/self_iterate.py verify --type "<类型>" --evidence "<回测结论>" --n <样本> --p <显著性>
```
固化后进入 calc_all 提示(标注"已固化"·依据可追溯)。**固化=启动批量更新流程**(改规则须同步 AGENTS/技能/数据 + check_sync + 备份)。

## 🔴 一条命令(推荐·赛果自动学习总入口)
```bash
python data/tmp/learn_from_result.py --case <id> --real <比分> --league <联赛>     --oh <主赔> --od <平赔> --oa <客赔> --handi <让球> --o25 <O25>     --pred "<预测方向>" --anchor "<主锚>|<次锚>" --center <进球中心>
```
→ 一次完成【复盘学习(self_iterate·偏差归因+复发提示) + 在线学习5层(River/贝叶斯增量/误差规则/CBR/持久化) + 合并报告】。`--status` 汇总。

### 在线学习系统 5 层(data/online_learning/·方案000054实施)
| 层 | 内容 | 状态 |
|:--|:--|:--|
| L1 River 在线ML | 方向分类器/进球回归/量级分类+ADWIN漂移 | 🔴**实测无增益**(30000场·方向34.5% vs 热门48.7%·-14.2pp)→默认关闭仅记录 |
| L2 贝叶斯增量表 | 比分/让球层/进球档格子+联赛校准 | 增量独立累积·n≥50可合并(**合并走批量流程·静态表防污染**)|
| L3 误差模式 | 场景分组误差→修正规则(≥30样本) | 与 learned_rules 协同·预测时提示 |
| L4 案例库 CBR | 相似案例检索+错误案例警告 | 案例库累积·提示级 |
| L5 持久化 | JSON/pickle + 备份/回滚 | 每场自动备份·可回滚 |

## 数据文件
| 文件 | 内容 |
|:--|:--|
| `data/case-library/学习台账.csv` | 逐场学习记录(偏差/归因/状态)|
| `data/tmp/learned_rules.json` | 学习成果库(count/status/rule/evidence/verified)|

## 与既有机制的关系
- **不替代**:量级台账(预测 vs 实际对账)、观察项 md(O1-O18)、case_stats(命中率)、check_sync(一致性)
- **新增价值**:把"教训"变成**可累积计数 + 复发自动提示 + 门槛判定 + 回测固化**的闭环——同一失误第二次出现即自动进入每场提示,不再依赖人工记忆

## 首个固化成果(示例)
`超深屠杀锚量级`:主胜隐含>88%+让≥2 → 历史 n=2109 净3+ 43.2%/净4+ 24.9%/净0-1 29%(时间分割 516 场一致)→ 主锚净3·**次锚须含净4-5+ 出口**·保留净1 停止分支。

## 🔴 置信度校准复核
批量复盘后执行 `python data/tmp/conf_calib.py` → 推导 **strict/any 双口径**(不改案例库 15 列模板)→ 更新 `data/case-library/置信度校准.csv`。
🔴**实测基线(182 场)**: strict 55.5% / any 65.4% / 比分锚 28.0% / 完美 23.6%。
🔴**观察项**: MID-LOW 档 strict 34.3%(标称 45-55%·低于 10.7pp·n=35 样本不足不固化);LOW 档 any 73.9% vs strict 43.5%(any 口径虚高证实)。
🔴**纪律**: 置信度统计一律用 **strict 口径**;档位调整须达"方向类≥10场同向 + p<0.05"门槛。

From MENG-COOLMAN/PitchQuant, MIT.

More from MENG-COOLMAN/PitchQuant

Signals

Stars
3
Forks
2
Open issues
0
Last commit (this folder)
Sep 23, 2026 (1 month ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from MENG-COOLMAN

Done for you

Rather not set it up yourself?

Chalk Labs scopes, builds and runs agents for your team: the task audit, the eval harness, the integrations and the weekly ops report. Fixed scope, priced before we start.

See how we build AI agents →Read the case studies

Chalk Labs is a two-founder AI and growth studio. We build AI agents, CRMs and MVPs for Web3 and AI companies, and have shipped 50+ product and token launches.

launches shipped
50+
launches shipped
media placements
500+
media placements
founders on every account
2
founders on every account

Press we've placed: Forbes · CoinDesk · Decrypt · Cointelegraph · The Block

Rahil Jain, Co-founder, Chalk Labs. Ex-Goldman Sachs engineer. You talk to the person who builds it.

Get SK-self-iterate running in your business

Tell us what you want to automate. A founder replies within one working day with a plan, not a pitch deck.

Goes to the Chalk Labs founders, not the newsletter. Privacy

Need help deploying SK-self-iterate?

CallGet help
Cite this page

SK-self-iterate skill. Agent Mart. https://agentmart.dev/skills/meng-coolman-sk-self-iterate (last updated 2026-10-24)