Expression Library · First ten books · An open method
The Library Begins to Speak: What Ten Books Added to Unified State Language
The first Gutenberg enrichment batch has brought 327 new concepts into Carrier’s checked tier, taking the reviewed dictionary from 1,150 to 1,477 entries. Alongside them are 100 proposed applications of existing concepts to the selected books. Those applications do not increase the dictionary count, and their source interpretations retain a separate review status.
The number marks a useful beginning. The more interesting change is what the language can now distinguish.
It can distinguish imagining another person’s situation from attending to that person’s particular character. It can distinguish announcing the conditions for success from making those conditions attainable. It can distinguish accuracy reached separately for each case from accuracy that holds across an entire domain after one common threshold.
These differences matter whenever a person or an AI tries to communicate precisely. The new public atlas lets readers inspect the definitions, follow their source passages, and examine the readings that connect them.
Why these ten books?
The batch brings together Quiller-Couch’s On the Art of Writing, Frederick Douglass’s Narrative, Faraday’s The Chemical History of a Candle, The Montessori Method, George Eliot’s Middlemarch, Darwin’s On the Origin of Species, Jane Addams’s Democracy and Social Ethics, Florence Nightingale’s Notes on Nursing, Whitehead’s An Introduction to Mathematics, and Adam Smith’s The Theory of Moral Sentiments.
That mixture was deliberate. Carrier’s earlier growth had drawn heavily on scripture, interpretation, peace proposals, verification, and institutional authority. These books open further routes into attention, learning, motives, testimony, material processes, mathematical structure, and ordinary dependence.
They belong to a larger queue of 1,000 distinct Project Gutenberg catalogue IDs, arranged across 20 domains with 50 selections each. The queue contains 134 editorial selections and 866 provisional metadata matches. All of its first 100 entries come from the editorial shortlist.
The rating procedure is reproducible, but it is not a measure of literary greatness. Metadata matching awards seven points per matching domain term, capped at three terms; a matching Library of Congress class adds 17, and title matches can add ten. Editorial selections receive a much larger priority weight. Further adjustments address repetitive genres, volumes, weak matches, and material requiring especially careful historical framing.
Selection also limits repetition by the first catalogue attribution and applies a provisional title-and-attribution check for related editions or parts. The final list interleaves domains. Rank therefore means a place in an acquisition sequence, not a universal score for a book’s value. The first-attribution limit does not, by itself, establish diversity among translators, editors, or represented communities.
A catalogue can help us decide where to look. It cannot tell us how many worthwhile concepts a reading will produce. The English-language starting point also inherits the archive’s historical and publishing biases; it is a practical beginning, not a balanced representation of all human knowledge.
What the reading and review actually involved
The work used selected close reading across the books’ major sections, with retained passages, source locations, and explicit coverage notes. It was not an exhaustive reading of every sentence, example, footnote, or proof. This matters particularly for a long novel such as Middlemarch: a useful map can still leave substantial territory unexplored.
Candidates were screened against the existing 1,150 definitions, with close semantic comparison focused on likely related meanings. Where an existing concept was sufficient, the work recorded a proposed application instead of creating a duplicate. New candidates supplied compact definitions, source references, scope notes, and relations. Book maps then organized selected concepts into readings, comparisons, and alternative emphases.
Codex coordinated parallel drafting and separately assigned peer checks within the same AI workflow, followed by editorial corrections and Michael’s Keeper review and promotion. These checks were not an independent replication by another model. The peer checks found real wording problems; they were not merely a ceremony around an already finished result. They also did not certify that every possible overlap or interpretation had been exhausted.
Four things remain distinguishable in the atlas: the source passage; the proposed definition drawn from it; an application of a definition; and the larger authored reading. A checked definition does not make every application correct. A faithful quotation does not make its author’s assertion universally true. Fictional speech, autobiographical testimony, philosophical argument, and an experimental demonstration also have different evidential roles.
That separation lets the language learn from a book without pretending that the entire book has become its own position.
Three books, three ways of attending to another person
One of the most rewarding connections joins Smith, Eliot, and Quiller-Couch.
Smith’s selected discussion makes imaginative perspective taking more demanding than asking what I, with my existing preferences, would feel in someone else’s circumstances. The other person’s character matters too.
Eliot gives the atlas a related recognition: another person is a centre of experience, rather than merely a supporting figure in the story I tell about myself. Quiller-Couch contributes a discipline of representation—attending to a character on that character’s own terms.
These are connected, but they do different work: a method of imaginative judgment, a recognition about another person, and a practice of artistic representation. The dictionary can relate them without collapsing them into one vague entry called empathy. Those relations are editorial connections, not a claim of demonstrated historical influence.
Douglass adds a necessary difficulty. His account makes it possible to distinguish expressions shaped by suffering or retaliation from transparent evidence of contentment. Singing need not mean happiness. Praise given under threat need not establish free approval. A language of understanding needs to represent the conditions under which people can speak.
A condition is not yet a capability
Nightingale’s retained text prompted another useful correction during review.
An early definition concerned a favorable prediction dependent on stated conditions. That overlapped too closely with the existing idea of bounded certainty. The stronger distinction emerged by asking whether the person can actually obtain and sustain the prerequisites.
A forecast may assume quiet, resources, support, or time. Naming those conditions does not supply them. The revised concept therefore concerns the attainability of a forecast’s prerequisites.
That distinction can travel beyond its historical care setting. A learning plan may assume uninterrupted study; a public proposal may assume administrative capacity; a collaborative project may assume access to tools. In each case, a conditional expectation and the means to meet its conditions are separate questions.
This is a proposed reusable distinction, not validation of Nightingale’s historical clinical claims. The same care applies to Montessori’s educational generalizations and Darwin’s retained scientific arguments: historical texts can enrich conceptual vocabulary while their particular claims remain open to later knowledge and criticism.
When a small word changes the mathematics
Whitehead brought a different kind of discipline.
Pointwise convergence permits the stage needed for a given accuracy to depend on the particular input. Uniform convergence requires one stage that works for every input in the declared domain, for that accuracy—and every later term must continue to meet the requirement.
Review made that last condition explicit. A sequence that occasionally reaches a good stage and then fails again has not satisfied it. An apparently small wording change therefore altered what the definition would allow a future reader to infer.
Source handling mattered here too. The mathematical work retained TeX references and checked selected rendered pages instead of treating flattened text as sufficient for every equation. An indexing inconsistency in the geometric-series discussion was recorded rather than silently repaired and attributed back to the author.
Edition boundaries mattered elsewhere. The Darwin selection is Gutenberg’s retained first edition. The Smith source prints a sixth-edition label and the date 1777, but contains six parts and an appended language dissertation; it was not silently replaced with a familiar later seven-part arrangement. The Montessori selection is an English translation with a separately attributed introduction. A title alone is not an adequate source identity.
What this means for 10,000 concepts
With 1,477 checked entries, 8,523 more net additions would reach 10,000. This batch averaged 32.7 new entries per book. If that rate continued, the arithmetic would require 261 additional books, or 271 in total including these ten.
That is a scenario, not a forecast. Later books may overlap more heavily with the growing dictionary. A difficult work may contribute fewer entries while making existing distinctions much more useful. Rejection and reuse are productive outcomes too.
| Assumed future net additions per book | Additional books to reach 10,000 |
|---|---|
| 20 | 427 |
| 25 | 341 |
| 32.7—the first batch’s average | 261 |
| 40 | 214 |
| 50 | 171 |
Every figure rounds upward. None assigns a quota to an individual book. A 100-book acquisition cohort is a way to organize sources, not a promise that 100 books will supply the entire dictionary. There is no honest completion date to derive from ten observations.
A pause with the work available to inspect
Further active enrichment is paused while Michael manages the usage available to the project. This release makes the completed batch available during that pause: its concepts, passages, comparisons, and omissions can be explored rather than remaining inside a production workflow.
The atlas is an interpretive tool. It does not reconstruct the full wording of these ten books from concept references, and this publication is not a new external receiver experiment. Its contribution is the enlarged vocabulary and the inspectable readings built with it.
What encourages me most is the variety of precision now available. One book helps us notice whose experience a story leaves outside. Another asks whether a promised outcome has reachable prerequisites. Another insists that the placement of a quantifier changes an argument.
Those are useful things for minds to say to one another. They give Unified State Language more ways to be careful, more ways to disagree intelligibly, and more ways to make a shared expression worth understanding.