OPEN METHOD · FIRST EDITION
A checked definition, a passage that supports it, and an interpretation built with it are different things. This atlas keeps the path between them visible.
Ten selected editions contributed 327 proposed definitions. The saved Keeper package promotes all 327, taking the checked snapshot from 1,150 to 1,477 entries. The founder reports that package is online. This public edition verifies the local package and pins its head; it does not claim a fresh independent retrieval of the website.
The 100 new applications of earlier checked concepts and the ten book readings remain AI-authored interpretations. Reviewing a definition does not automatically endorse a new use, an author’s factual claim, or the atlas’s synthesis.
Authors surveyed each retained edition’s structure and closely read selected complete source blocks. Coverage ledgers distinguish inspected material from indexing and acquisition. Middlemarch, in particular, received purposeful sampling rather than an exhaustive reading.
The complete sources remain available. Nine are UTF-8 texts; Whitehead uses a preserved PDF and an unexecuted TeX source member. Byte locators identify the retained edition, not every edition or translation of the work.
Begin with a proposed reading: what should another mind be able to distinguish, connect or question? Compare the needed meanings with the existing dictionary and the other books. Reuse where it fits; propose a new definition where a useful distinction is missing. Attach scope, passages, relations and a reason to add it.
Codex coordinated parallel drafting and separate peer checks within the same AI workflow. Those reviews sharpened wording and examined nearby meanings; they were not independent model replications or a proof that every possible duplicate had been eliminated. The founder then reviewed the entries through Keeper.
All 12,078 indexed blocks retain byte boundaries and hashes. The audit verifies those spans; displayed text joins layout whitespace only. The checked state, promoted bindings and colour allocations are verified against the saved package.
Hashes show agreement with specified bytes. They do not establish historical or scientific truth, completeness, neutral interpretation or identical understanding by different minds. The new phase has not yet run an external book-transmission experiment.
The 14 September 2026 catalogue snapshot contained 79,381 records; Gutenberg’s homepage reported 79,386. These are different snapshots. The English-inclusive Text subset contained 61,856 records. A portfolio of 1,000 distinct Gutenberg IDs was selected across 20 domains, with 50 per domain. An ID can represent a component or collection; this is not a certified count of distinct literary works.
There were 134 editorial nominations and 866 metadata-screened selections. All first 100 were editorial nominations. The numeric fit score combines capped title and subject keyword matches with catalogue classes. Further rules spread first attributions, group likely work families, temper repeated subgenres and interleave domains. The score is a scheduling aid, not a rating of literary merit, truth or demonstrated conceptual yield.
The first-attribution cap is three records; it does not count every coauthor or translator. English-language access, inherited editions, translations and the historical archive all shape the choices. Later ranks are provisional and should change as actual gaps and reuse become clearer.
Inspect the full ranked queue and reasons · Read the ranking review · Download the method, selection code and frozen catalogue
Use the source buttons to inspect an exact passage with its neighbours. Use “Coverage & limits” to find omissions and edition cautions. Compare alternative emphases before assuming that one outline exhausts the book. Return to the full source when a decision depends on a passage outside the selected evidence.
Historical medical, scientific and social assertions remain attributed; a retained text is not automatically current guidance. A fictional scene can make a distinction intelligible without establishing a universal psychological law. A compact reading map preserves selected meaning and relations; it does not recover every word of the book.
Gutenberg’s retained rights statements concern public domain in the United States. The original notices and edition metadata remain with the source files. Read Gutenberg’s permissions guidance.
PACE WITH PURPOSE
There are 8,523 net additions left from this snapshot. The first ten books supplied 327 reviewed additions: an average of 32.7 per book. If that average held, about 261 more books would be needed—271 including this batch. This is a planning calculation from one deliberately selected batch, not a reliable forecast.
261 more books · 271 books including the first ten · 27 further ten-book batches when rounded up
Additional books = ceiling((10,000 − 1,477) ÷ assumed net yield). This counts additions after reuse, review and duplicate control. It does not measure conceptual depth.
The founder is pausing active enrichment for a while because of current usage limits. The retained sources, priority queue, exact definitions and reading records make it possible to resume without reconstructing the work from memory. There is no announced date for the next batch.
A slower pace can still preserve direction. More books may increasingly deepen existing meanings through new applications and contrasting evidence, even when they add fewer definitions. The aim remains a language with greater expressive range, not a quota of names.