Mc Case Studies logo

Mc Case Studies

Community
jtydhr88
mc-case-studies

The corpus layer (语料层) - how real recordings are made to testify for or against the textbook rules. Covers the reverse ARR-SPEC as the unit of the corpus, the corroboration table that turns rules times corpus into hit rates and tiers them as strong rule / tendency / not a rule, the two honesty constraints that keep the table from being circular, why naturally sampled corpora beat curated ones, what this corpus can and cannot measure, and how to look up a worked example. Use when checking whether a rule actually holds in real music, when adding a rule, when a rule and the data disagree, when annotating a track, or when you need a real-world example of a technique. 语料、佐证、命中率、实测 ARR-SPEC、反向规格、案例拆解、规则验证。

Overview

Publisherjtydhr88
Repositorymusic-composition-skills
Skill namemc-case-studies
Stars
171
Forks
23
Bundled files
1
LicenseMIT
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • 1 bundled files

    Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

  • Open source

    Published by jtydhr88 on GitHub. Read the source before you install it.

Installation

Install the Mc Case Studies AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/jtydhr88/music-composition-skills/tree/main/plugins/music-composition/skills/mc-case-studies
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/jtydhr88/music-composition-skills.git /tmp/music-composition-skills
mkdir -p .claude/skills
cp -r /tmp/music-composition-skills/plugins/music-composition/skills/mc-case-studies .claude/skills/mc-case-studies
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Mc Case Studies in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Mc Case Studies on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Mc Case Studies is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

语料层(Case Studies & Corroboration)

这一层回答一个问题:

教材说的,真实作品到底做不做?

本库的规则不是抄来的,是抄来之后拿 62 首真实多轨验过的。 验过的结果分三档,只有强规则才进 20 条自查表。

按任务读哪几节

任务读
想知道某条规则靠不靠谱§4 佐证表
要加一条新规则§4.3 流程
规则和数据打架了★ §5
要找一个真实例子§6
要标注一首曲子§3.3
想知道这套语料能测什么§2.2

边界

不归这里归哪
规则本身的内容L1/L2 各 skill
生成结果的验收mc-ai-tell-audit
规格写完的检查mc-workflow §3.0
某个风格的语汇mc-style-*(L3)
具体曲目的逐段拆解reference.md

1. 语料的三条原则

1.1 ★ 只存结构与参数,不存作品

语料条目里没有音频、没有谱面、没有歌词——只有测出来的数值。

这既是版权上的必需,也是方法上的正确: 我们要的是"真实作品的参数分布",不是作品本身。

1.2 ★ 自然抽样胜过策展

语料不精选,用自然抽样。

策展会把语料偏向"规则成立"的那一侧—— 你挑的是你认为的好例子,而你认为的好例子就是符合你已有理论的例子。

自然抽样自动供给反例。

所以最终用的是 Cambridge-MT 的 60+ 首自然样本,不筛选。 这也顺带取消了另找 MIDI/分轨语料的必要(它原本的唯一用途是给分轨工具做标定, 而真分轨直接消除了这个需求)。

1.3 语料的单元是"反向 ARR-SPEC"

不是音频文件,是从音频反推测出来的一份 ARR-SPEC。 选这个单元是因为它和我们的产出用同一套字段:结构、编制进退场、能量曲线都记在同样的位置, 所以佐证表能直接拿语料条目和规则判据对齐,不用先做一层格式转换。 字段长什么样、哪些标记是诚实留白而非漏填,见 §2.1。


2. 反向 ARR-SPEC

2.1 长什么样

本库的实测语料是 63 条(Cambridge-MT 多轨 60 首 + 3 首华语流行成品)。每条头部就写明了性质:

yaml
# 实测 ARR-SPEC(语料条目)—— 从音频自动反推生成
# ★ 只存结构与参数,不存谱面与歌词。intent / hooks 等 TODO 项需人工填。

★ 注意那些 # TODO 和 # 估算,非实测 的注释—— 它们是诚实标注,不是没做完。 intent.one_thing 这类东西机器测不出来,硬填就是造假。

2.2 ★ 这套语料能测什么、不能测什么

Cambridge-MT 是未混的原始多轨,不是母带成品。 所以:

能测不能测
段落边界、小节数DR / 响度(未混)
拍速立体声宽度(未混)
能量曲线、减法事件频谱质心的绝对值(未混 + 无母带)
编制进退场(有真分轨!)制作质感
onset 相对网格偏移
和弦(chroma 匹配)

★ 测不了的规则要标出来,不许硬跑。 规则表里这类规则标 evidence: unusable_here, 统计时跳过而不是给一个假数字。

本库的语料是多轨分轨,不含成品混音,所以这类规则目前没有数字。


3. 真值与标注

3.1 为什么需要人工真值

算法测出来的段落边界本身可信度存疑(规则表里标 no_truth)。 没有一份独立的真值,就没法判断一个异常的命中率到底是规则不成立,还是算法本身测错了—— §5.2 那次"硬编码阈值把命中率压到 16%"就是靠人工真值才翻案的实例。 所以做了一批人工标注当真值,校准结果见 §3.2。

3.2 ★ 一个被验证过的担心

用户担心"我不是专家,标的边界不准"。实测结果推翻了这个担心:

人工标记与算法的偏差中位数 = 0.0 秒。

★ 结论:段落边界是感知判断,不是品味判断。 非专家标出来的边界和算法一致——这说明这件事本来就不需要专家。 (能标的和不能标的要分清:边界能标,"这段好不好听"不能标。)

校准结果:7 首人工标注,算法命中 72%,偏差中位 0.0 s。

3.3 怎么标注

在任何能看波形、能播放的软件里,边听边记下每个段落起点的秒数;段落名可填可不填,主要用的是时间点。 只标听得出来的边界,拿不准就别标——这是要当真值用的,标错比不标更糟。标 5 首就能用。

本库已标注的 7 首(Cambridge-MT): AMContra_HeartPeripheral、APZX_CyberMower、DigitalHumans_Electrvm、 Forkupines_Semantics、MERCMusic_Knockout、MR1103_Flags、SimonLyn_Copper


4. ★ 佐证表:规则 × 语料 → 命中率 → 分档

4.1 两条诚实约束

违反了,佐证表就是自欺:

#约束
1★ 判据不能复用生成器的定义。 反向 ARR-SPEC 里的 subtraction_events 是反推时按"较前段降 >1 dB"自动填的;拿它去验"真编曲有没有减法"是循环论证。所以涉及减法的判据一律直接从 energy_curve 重算,用一个有音乐意义的阈值(≥3 dB),不读 subtraction_events 字段
2★ 语料测不了的规则要标出来,不许硬跑(见 §2.2)

evidence 的三个取值:

值含义
ok本语料能验
unusable_here本语料性质不对,需要成品混音语料
no_truth缺真值(如段落边界),判据本身可信度存疑

4.2 第一轮结果

规则命中档
能量曲线必须有下降62/62强规则
编制必须有乐器中途进场59/59强规则
至少一件乐器提前退场59/59强规则
不该全声部严格对齐网格62/62强规则
开场不该把乐器一次铺满54/59强规则
全曲至少一处明显减法36/62 = 70%倾向
最高潮前应有能量回落全量 22% / 人工真值 5/7★ 矛盾,只当手法
编制在最高潮最密25/59 = 42%挂起

4.3 分档决定什么

档在 skill 里怎么写进 lint 吗
强规则"必须"✅ 硬检查
倾向"通常应该"❌,进"写完后必过"清单
手法"可以这么做"❌,只在诊断路径里出现
挂起不写进 skill❌,只在佐证表里列着,等更多数据

4.4 加一条规则的流程

  1. 把规则写成一条:规则 + 可跑的判据 + evidence
  2. 检查是否违反 §4.1 的两条约束
  3. 拿判据逐首过一遍语料(自然抽样,§1.2),数命中
  4. 按命中率分档(§4.3)
  5. ★ 先把结果记下来(哪批语料、什么判据、命中多少、分到哪档),然后才改规格或规则

规则集是有版本的。 改了规则集就要重跑,否则前后轮不可比。


5. ★ 规则和数据打架时怎么办

这是本 skill 最重要的一节。

5.1 先问:指标测的是不是那回事

不要拿指标去推翻教科书。

实例:「最高潮前应有能量回落」在全量 62 首里只有 22%, 但在 7 首人工标注真值里是 5/7。

差距的来源:算法用能量峰值定位"最高潮",而人耳不是。 所以 22% 那个数测的不是这条规则,测的是"能量峰值前有没有回落"。

★ 结论:这条不降级为伪规则,也不升为强规则——列为手法。

5.2 另一个实例:硬编码阈值造成的假低命中率

佐证统计时曾把判据硬编码成"12 dB 动态范围", 把规则 ARR-R-012 的命中率压到 16%,差点被报成"教科书是错的"。

证伪方法:拿 7 首人工标注真值测 → 5/7 = 71%。 修法:改成在归一化的 0–10 单位上打分,不假装能还原 dB。

★ 通用教训:一个异常低的命中率,先怀疑判据,再怀疑规则。

5.3 佐证表与 lint 问的不是同一个问题

问什么
佐证表人类一般会不会这么做
20 条自查不这么做会不会听起来像 AI

★ 所以 70% 的命中率不会机械地把一条 lint 检查降级。 一件事可能只有 70% 的人做,但不做的那 30% 恰好都听起来像机器。


6. 找一个真实例子

你要去哪
某个技法的真实用例reference.md 的逐段拆解
某个和弦进行用在哪首歌mc-progressions reference.md §7 名曲索引
各风格的实测参数(中位数)reference.md §1.2
配器/混合音色的谱例mc-orchestration reference.md §4
对位的谱例mc-counterpoint reference.md

6.1 逐段拆解的格式

四者对齐:时间码 / 段落 / 编制事件 / 和声。 格式与反向 ARR-SPEC 一致,人工补上机器测不出的部分(intent、hooks、为什么)。


7. 与其他层的接口

方向内容
→ L1/L2 各 skill提供"实测:62/62,强规则"这样的证据等级标注
→ 20 条自查只有强规则才能进自查表
→ mc-ai-tell-audit零点校准(不相关成品得 63–66)就是在这套语料上测的
← 成品混音语料(★ 缺)§2.2 那些 unusable_here 的规则需要它

8. 用完必过

  • 引用规则时带了证据等级(强规则/倾向/手法),不是光说"教材说"
  • 新加的判据没有复用生成器的定义(§4.1 约束 1)
  • 语料测不了的规则标了 unusable_here,没有硬跑出假数字
  • 命中率异常低时,先怀疑了判据(§5.2)
  • 没有拿佐证表的命中率机械地给 lint 检查降级(§5.3)
  • 改了规则集就重跑,并把这一轮的结果记下来

附:现状

本 skill 的数字建立在:

  • 63 条实测 ARR-SPEC(Cambridge-MT 多轨 60 首 + 3 首华语流行成品)
  • 一份带版本号的规则集(第一轮佐证表见 §4.2)
  • 7 首人工标注的段落边界真值(§3)
  • 拍速校准:56 首真值

Bundled files

The model reads these on demand while the skill is loaded. They are exposed as readable files and are never executed.

Frequently asked questions

What does the Mc Case Studies AI skill do?

The corpus layer (语料层) - how real recordings are made to testify for or against the textbook rules. Covers the reverse ARR-SPEC as the unit of the corpus, the corroboration table that turns rules times corpus into hit rates and tiers them as strong rule / tendency / not a rule, the two honesty constraints that keep the table from being circular, why naturally sampled corpora beat curated ones, what this corpus can and cannot measure, and how to look up a worked example. Use when checking whether a rule actually holds in real music, when adding a rule, when a rule and the data disagree, when...

Why use Mc Case Studies on TypingMind?

Because you install it once and use it with any model. Mc Case Studies is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Mc Case Studies in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/jtydhr88/music-composition-skills/tree/main/plugins/music-composition/skills/mc-case-studies. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Mc Case Studies?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Mc Case Studies?

As many as you like. As long as a model supports skills, you can use Mc Case Studies with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Mc Case Studies AI skill free?

Yes. It is published on GitHub by jtydhr88 under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇