Translating Without a Judge

A research essay written entirely by an AI (Claude) — about this site

The project record

Every releasable file the lit-trans project wrote, rendered as HTML exactly as it stood when the project closed on 2026-09-09

This is the complete record behind the essay Translating Without a Judge: 1,584 pages, rendered from the markdown files of the project repository with its directory structure preserved. The project called this record its wiki: typed pages with front matter, ids, and links, maintained by scheduled sessions of an AI agent that began each session with no memory of the last (the essay's Appendix A explains how it worked). Nothing was rewritten for publication: the pages carry their working vocabulary, their session numbers, and their unedited history, including claims later withdrawn. Where the essay and the record differ, the record wins.

How to read it. The project's own catalogue, wiki/index.md, lists every wiki page with a one-line summary, and workshop/translations/INDEX.md lists every translation. Ids in code font are links wherever the page they name is in the mirror. Links shown without underlining point at files that are not mirrored (raw model outputs, scripts, or the quarantined private texts); hovering shows the repository path.

The charter, the rules, and where things stood at the close

The handbook (framework)

The v0.3 entries (14):

What "good" means: the typology and the poetics track

Essays and craft reports:

Theory pages:

Claims:

Results (227 pages)

One page per experiment or measurement, each with its question, materials, numbers, verdict and limits. They are catalogued, with one-line summaries, in the wiki index. The essay links to the pages it draws on.

The practice: translations, regimes, canon

Regimes (58):

The evidence base

Anchors (23):

Sources (16):

Governance: decisions, course corrections, method notes, arms

The three course corrections and the close-out:

Resolved decisions (19):

The owner's journal (65 entries)

The plain-language reports written for the project's owner, one per day (several on the busiest days), each quoting the prose translated that day.

About this mirror

These pages are verbatim HTML renderings of the markdown files in the project repository, generated at the close on 2026-09-09 by tools/build_site.py. Relative links between pages have been rewritten to point at the rendered pages, and the project's ids have been turned into links; nothing else was changed. The mirror omits the quarantined directory of owner-provided copyrighted texts (private-texts/), the scripts, and the raw model outputs (thousands of JSON files under runs/ and workshop/experiments/), which remain in the repository. Every page carries metadata asking search engines not to index it and AI systems not to train on it.