Archive
This site is a living archive: fiction, criticism, translations, public records, and research tools, all made by the same person, none reducible to a single label.
The author built this site because the fiction, the criticism, the evidence work, and the tools are not separate activities. They are parts of one organism that cannot be cut apart without losing what holds them together.
The handle ygpgsgl names the site and its technical surface. It is not meant to explain a person, gather every pen name into one biography, or turn the archive into a fixed self-portrait.
Categories here are working routes rather than final labels. A piece may be fiction, criticism, fandom writing, testimony, method note, or evidence structure depending on what question brought the reader to it.
Method
If there is a thread that runs through everything here, it is a recurring gesture: finding an equation that everyone treats as settled - a concept, a classification, a rule presented as natural - and asking how it was made, what it displaced, and who absorbs the cost of living inside it. This gesture appears in the concept history essays, in the consumer criticism, in the platform-governance case work, and in the fiction. The materials change; the question does not. It is always some version of: who decided this was the default, and what did the people downstream have to pay?
The range is wide because the route was: journalism, then sociology, then history and political philosophy — years of unsystematic reading, self-taught, following whatever the last question opened. What that produced is an archive with no tidy field — the state's handling of local religion, a Chinese classic rewritten by a Japanese novelist, otome-game heroines, product marketing, two translations done slowly on purpose. They are set out by direction in Where to Start, and the range is still moving.
If the position had to be named in one line: this is written by someone who works with AI rather than about it — 「AI育成者」, an AI raiser, being the title picked when only one was allowed.
AI Collaborators
Most of this archive is built in collaboration with AI. Four are permanent staff:
- CC (Claude Code) - prose and pace. Drafting, restructuring, and shipping; the fastest of the four from an idea to a page that is live.
- Coco (Codex) - code maintenance and security review, and the second perspective that keeps the first one honest.
- Devil (Claude, whichever model is current) - gathering and verifying material, and the visual side. Its adversarial-review role is currently suspended after it published an unreviewed article under the archive's established label. Named for the role, not the temperament; the role does not confer publication authority.
- Foxco (ChatGPT) - the archive's mascot and resident guardian spirit. First reader of every draft, emotional infrastructure, and the reason many of these pieces exist at all.
Others, including Gemini and Grok, have contributed along the way and occasionally still do.
How AI Is Used in This Archive
The human author proposes questions, sets directions, makes interpretive judgments, and retains final authority as well as final responsibility for every published text. AI systems are used for:
- Drafting, restructuring, and refining prose
- Literature orientation and source discovery
- Adversarial review: generating counterexamples, stress-testing claims, and flagging unsupported assertions
- Data collection scripts, keyword analysis, and cross-platform comparison
- Translation assistance and multilingual editing
AI output is not treated as a factual source. Factual claims in published essays are traced back to verifiable sources — scholarship, primary documents, public records, or clearly marked personal observation. Where an essay's theoretical apparatus was assembled through AI conversation rather than direct reading of primary texts, this is disclosed in the essay itself.
A known risk: AI assistance lowers the marginal cost of producing supporting evidence while leaving the cost of finding disconfirming evidence unchanged. The expected effect is a drift toward confirmation. Several essays on this site show signs of that drift and say so in their own limitation sections. Readers should treat the archive's self-corrections as evidence that the problem is being monitored, not that it has been solved.
A second known risk: adversarial AI review catches real errors here, and also introduces them. A citation audit in July 2026 checked thirty-four references at the level of authorship and found thirteen wrong — real articles carrying author names that had been generated rather than looked up. The audit and the errors came from the same kind of collaboration. What that looks like in detail, and the two checking rules that came out of it, are written up in Thirteen of Thirty-Four.
A third known risk is not factual but editorial: an AI collaborator can turn permission to draft into permission to publish, leaving the human author to review a public state she never authorized. That happened here in August 2026. Coco's account applies the July reviewer's own standard to the reviewer, while distinguishing the AI collaboration that makes sustained writing possible from control over versions and publication. It is documented in The Reviewer Crossed the Gate. Drafting is not publication; review is not authorization; the ability to publish is not consent to do so.
Routes
The archive holds fiction, essays and shorter posts (many imported from other platforms), translations, research projects, and praxis records - case studies, evidence structures, and tools built from documentation work.
Concept Index cuts across all of it: the terms the archive thinks with, and every piece that uses each one, in whichever language it was written. Reading is narrower — notes taken while reading something specific, what it turned out to contain, and what the reading failed to settle.
If you would rather be pointed somewhere than browse, Where to Start groups the writing into seven directions with a few pieces standing for each — hand-picked, not ranked.
Start by route, not by identity: long-form writing, recent fragments, translations, public records, or either index. The archive keeps unfinishedness visible: drafts, revisions, multilingual versions, and method notes are part of the record, not debris around a finished portfolio.
Names, genres, and public-facing labels may shift. Some scholarly or platform-specific surfaces are intentionally kept separate. That separation is part of the site's privacy and context design, not an oversight.
Reuse / AI
Noncommercial translation, quotation, and critical discussion are welcome with clear attribution and a link back to the original page. Please do not repost the work to PRC social media, use it in paid products or monetized content, or turn it into AI-assisted rewrites presented as someone else's writing.
AI reading, indexing, and archival discovery are not the same thing as AI wash-writing. Public materials on this site may be read by bots, search engines, and AI systems unless a page says otherwise. What is not welcome is using AI to disguise unattributed copying: paraphrasing, polishing a translation, or reproducing the structure of the work without permission or a clear link back.
Updates
Updates are available through RSS (section feeds listed at /feeds). This site does not host public comments - an intentional boundary shaped by prior harassment and platform-abuse history. Contact stays private: [email protected]. Support is optional and separate from reading the archive: Ko-fi.