Skip to content

Latest commit

 

History

33 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Marginalia

mar·gin·a·li·a (n.) — notes scribbled in the margins of a book; the traces a reader leaves behind.

Selective research & reading notes by Yunqi Chen — distilled musings from AI-assisted analyses, published bilingually (English / 中文).

🌐 Website: https://uniqueclouds.github.io/marginalia/ — 全部内容物在线阅读(MkDocs 构建,push 自动部署)

This repo is deliberately sparse. The workspace behind it holds raw corpora, private drafts and unfinished work; by default everything is git-ignored, and only explicitly whitelisted, curated notes are ever committed. Nothing lands here casually — each entry is published on purpose, as one issue → one pull request → one squashed commit.

How an entry is born

  1. Issue — the musing itself: what triggered it, where the data came from, what the original prompt/idea was.
  2. Pull request — the distilled bilingual note (note.en.md + note.zh.md), cross-linked to the issue, plus one new row in the index below.
  3. Commit — exactly one squashed commit on main per entry, so the history reads like a table of contents.

Provenance format

Every note opens with full provenance — no orphan musings:

---
id:              marginalia-NNN
title:           ...
date:            when the analysis was done
published:       when this note was committed
kind:            musing | analysis | proposal | survey
sources:         local corpora / files / tools the musing is grounded in
initial-prompt:  the request (idea) that started the conversation
agent:           ZCode CLI
model:           the LLM behind the conversation
issue:           N
---

Index

# Entry Date Issue → PR
001 Do coding agents have taste? What 29,787 Discord messages say about beautiful vs. ugly code · 中文 2026-08-15 #1
002 Writing like Dourish: a 21-text corpus analysis of a critical HCI voice (2004–2026) · 中文 2026-08-15 #3
003 NOTUGLY-S: learning "not ugly" instead of "beautiful" — an NLP × program-analysis proposal · 中文 2026-08-15 #2
004 Is CHI/ACL a storytelling festival? From community gripe to measurable construct · 中文 2026-08-15 #7
005 Four academic voices, measured: language styles of BDS/HCI/Sociology/SE classics (314 papers, 3.3M words) · 中文 2026-08-15 #11
006 Spotify Podcast Guide 2026 · 英文播客推荐清单(47 节目 / 77 集精选,全部官方链接) · English 2026-08-17 #13
007 Nuance rising and falling in scientific writing: identification, measurement, and one local-corpus test (314 papers / 3.3M words) · 中文 2026-08-17 #14
008 How to Scale Your Model — reading note: LLM training taken from alchemy to a roofline accounting (JAX-ML scaling book) · 中文 2026-08-17 #15
009 Homologies in Fields of Cultural Production. Evidence from the European Scientific Field — borrowed the boundary, not the fractal · 中文 2026-08-17 #17
010 组织惰性:成功的副产品,抑或组织病症?——基于系统性审查方法的述评与展望 — a reading note: the byproduct–symptom dichotomy, left standing · 中文 2026-08-18 #22
011 Awesome Auditable AI — reading note: 188 entries / 9 sections / 132 arXiv papers; how a curated list turns AI agent auditability from a slogan into reliability engineering · 中文 2026-08-17 #26
012 Large-Scale Temporal Analysis of Wikipedia Edit History and Talk Pages — survey: the pieces are ready, the joint study is missing · 中文 2026-08-18 #28
013 Ethnomethodology — survey: origins, its relation to ethnography, and development · 中文 2026-08-21 #30
014 Sandwich OCR for scanned books — making Putnam and Rorty pixel-identical and fully searchable · 中文 2026-08-22 #37

Artifacts

Companions to the entries above, published verbatim — live skills, original source documents, full reports — in whatever language they were born in. Each lives inside its entry's directory but arrives in its own PR; artifacts need no issue.

Artifact Entry PR
dourish-style polishing skill (since superseded by academic-voices) 002 #9
Storytelling originals — survey + proposal v2 (中文) 004 #10
Five full discipline-style reports (中文) 005 #12
academic-voices skill — five measured voices, density baselines + voice classifier 005 #13
ocr_pipeline — sandwich OCR for scanned books: byte-identical page images + per-character invisible text layer 014 #36

Languages

Every note exists twice — *.en.md and *.zh.md, same content, neither an afterthought. Verbatim artifacts (skills, original source documents) are published as-is, in whatever language they were born in: switch to 中文版 README.

License

The contents of this repository and the website — the entry notes, reports, the podcast guide — are licensed under Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0). You're free to share and adapt for any non-commercial purpose, as long as you credit the source; commercial use requires prior permission. The underlying data corpora are not redistributed here and remain governed by their original sources (each entry's sources field cites them).

About

Research & reading marginalia — selective bilingual (EN/中文) AI-assisted research notes, each with full provenance: data source · initial prompt · date · model.

Topics

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages