What 1,521 new concepts teach us about building a shared language

The most valuable result of this enrichment is that a future expression can say more precisely what it means. “Uncertainty,” “adaptation,” “evidence” and “connection” now have more of the surrounding vocabulary they need: conditions, quantities, mechanisms, alternatives and limits.

Carrier’s new Wikipedia foundations edition adds 1,521 approved concepts to an earlier checked vocabulary of 2,568. Its resulting 4,089 records do not make a complete account of the world. They make a much wider set of distinctions available for authors to compose with—and for readers to question.

What grew, and what the numbers count

2,568 earlier checked concepts + 1,521 approved additions = 4,089. Vocabulary growth: 59.23%.

These units should not be substituted for each other. A retained article can support several concepts. Several references can use the same passage. A conceptual distinction can already exist in Carrier and be reused rather than added again. An article count therefore cannot serve as a vocabulary count, and a vocabulary count cannot measure depth by itself.

The method: retain, select, adapt, connect, review

The starting point was a frozen, identified Carrier state. Its existing definitions supplied both reusable concepts and a way to notice missing foundations. Wikipedia’s Vital Articles Level 4 inventory helped establish broad topic coverage; an editorial queue then added or prioritized subjects useful to this particular language.

The priority labels were judgments about usefulness to Carrier. They were not a validated ranking of the world’s knowledge, a scientific importance score, or a claim that these subjects outrank every omitted one.

Source acquisition used the official English Wikipedia Action API, with bounded requests and retry/backoff. The edition retains revision identifiers, article wikitext, captured HTML, normalized reading text, contributor-history links and hashes. It is a selected local corpus, rather than a full English Wikipedia dump.

There is a subtle but important source boundary here: an article revision identifier pins its wikitext, but does not by itself freeze every historical template or transclusion. The actual HTML capture and extracted reading therefore have their own byte hashes. The evidence record identifies what was retained, rather than asking the revision link to guarantee more than it does.

Extraction creates a reading from paragraphs and list items, joins layout whitespace, and excludes navigation, tables and designated reference or end-matter sections. Complete wikitext and HTML remain in the archive. A selected passage’s byte offsets refer to the normalized reading—not to the original wikitext.

AI authors wrote compact definitions, examined the selected retained passages, compared them with existing Carrier meanings, and proposed connections. Some definitions began provisionally from subject knowledge before being tested against the captured source. Wording was narrowed, corrected, supported with further passages, reused or deferred as needed.

Full source retention does not mean full-article inspection. The accountable unit of evidence in this edition is the selected passage and its disclosed relationship to the authored definition. There was no exhaustive paragraph-by-paragraph review of the whole corpus or primary-literature audit of every Wikipedia assertion.

Name checks, similarity searches and explicit reuse decisions helped identify overlap. They assist editorial judgment; they do not prove that every possible equivalent meaning has been found. Some broad foundations deliberately complement earlier definitions tied to a specific source or application.

Finally, Lucid reviewed and approved the 1,521 compact definitions and their displayed contextual links. The decision file is pinned to the exact review edition and its state heads. Promotion applies that recorded approval to the corresponding entries. It does not extend approval to all statements in a source article or to every future use of the concept.

Small wording decisions with large consequences

1. ACID is not Acid

A case-insensitive source lookup briefly treated a database term and a chemistry title as the same name. Exact passage checks exposed the mismatch before final admission. Source selection was changed to prefer the exact title and to permit a folded fallback only when it identifies one source.

This is a practical lesson for a shared vocabulary: normalizing names can make search convenient while destroying distinctions. A familiar-looking label is not enough to identify a source. The record must preserve the intended sense.

2. One article need not mean one concept

Memory consolidation and reconsolidation, consumer and producer surplus, and a Pareto-efficient state and a Pareto improvement can lead to shared or redirected article material while still requiring distinct definitions. Explicit mappings preserve the intended concept rather than selecting the first record associated with a title.

The converse matters too. Two promising topic labels can describe a meaning already represented in Carrier. Existing concepts were reused where appropriate, and duplicate proposals were consolidated. The aim was useful expression, not one new record for every search result.

3. A lead paragraph is a starting point

Several definitions needed material beyond the article introduction. The selected sampling-bias material, for example, makes room for properly weighted designs rather than treating every unequal selection probability as bias. Type II error uses failure-to-reject language rather than suggesting that a nonsignificant result proves the null hypothesis. The retained sampling-bias and error-type revisions let a reader inspect those choices.

The retained angular-momentum lead used an external-force shortcut where the relevant rotational conservation condition concerns external torque; a later passage supplied that condition. The authored definition follows the qualified support. These are findings about the captured revisions and the adaptation process, not a claim to have corrected Wikipedia itself. See the retained angular-momentum revision.

Targeted peer checking also strengthened biological definitions and their evidence: carrying capacity needed the upper sustainable level, photosynthesis needed explicit support for its anoxygenic qualification, and the ATP selection needed a passage supporting the signalling role already named. A targeted review is valuable precisely when its limited scope is recorded rather than expanded into an assertion that every specialist detail received independent review.

4. Conflicting sources remain inspectable

The retained Ceasefire and Armistice introductions differ over whether an armistice ends a war. The definitions distinguish the cessation of fighting from a final settlement and preserve the source discrepancy. A cultural-relativism lead also required a narrower methodological definition, supported by selected passages beyond the introduction, rather than turning a method of understanding into an unqualified claim that all positions are equally valid.

Keeping such judgments visible matters more than producing a frictionless summary. A reader can inspect the source wording, the authored definition and the reason for the distinction, then disagree with the editorial choice without losing the evidence.

What the larger vocabulary can help us say

The atlas carries nine authored composition examples. They are short demonstrations of how old and new concepts can work together; no new USL transport or independent interpretation experiment was performed in this phase.

One example separates a test’s ability to detect a condition from the probability that a positive result identifies a true case in a given population. Sensitivity, specificity, prevalence and positive predictive value make the direction of the probability statement explicit. The example is conceptual, not an individual diagnosis.

Another joins the existing exact-colour vocabulary to colour vision and visual perception. A file can preserve its RGB samples while the colour experienced by a person depends on the display, surrounding light and visual processing. An exact transport result does not establish identical experience.

Two Earth-and-space examples connect material cycles to energy conservation, heat transfer and radiative exchange. A habitat can recycle matter while still requiring energy, maintenance and heat rejection. A proposed space factory needs more than an available sunlight budget: its thermal and environmental boundaries also need calculation. These are requirements for analysis, not feasibility results.

A further example connects the existing USST and System Reference Axis vocabulary to angular momentum, mass, velocity and significant figures. It asks an author to declare the body set, epoch and reference frame before computing a shared rotation reference. The convention remains an adopted construction; extra displayed digits cannot remove uncertainty in the underlying state.

The more general gain is compositional. A future message can name an assumption, the model it belongs to, the quantity being measured, the kind of uncertainty being reported, and the condition under which a conclusion follows. Shared concepts reduce repeated explanation, while literal wording remains available for the particular claim.

A connected graph needs a meaning for its edges

The new records add 5,216 contextual pairs to the graph. Their links reach 539 distinct earlier concepts; 583 of the new-to-new pairs cross the edition’s domain categories. Each new concept has a recorded path into the earlier checked vocabulary. That gives the expansion routes into the work already done, rather than leaving new domains as disconnected additions.

The links are contextual navigation. “Referenced by” reverses that navigation; it does not certify agreement, causation, equivalence, transitivity or entailment. New-to-new links are stored once on the later admitted record, while the separate edge catalog preserves authored direction and qualifications. Storage order must not be mistaken for the direction of a scientific or logical relationship.

The combined undirected graph has 52 connected components. Its largest contains 4,036 of the 4,089 checked concepts. All new entries reach the earlier graph, but the whole graph is not one connected component. Its 84 unresolved editorial title leads remain visible as possible future work and do not become invalid dictionary targets.

These measures describe a recorded map. A central node is not automatically more true or more valuable. A sparse area is not automatically an intellectual void. The next useful addition may be a missing foundation, a sharper distinction, a better connection, or simply a correction to what is already present.

What checking establishes

Source verification checks retained file hashes, normalized units and reproduction of the extraction. Carrier validation checks the structured state, identifiers, allocation and link rules. The release process checks that the human decisions belong to the exact review edition and that the approved definitions and their relations are the ones promoted.

These checks make changes and inconsistencies detectable. They do not convert a Wikipedia claim into a fact, establish a specialist consensus, or measure the adequacy of an interpretation. The inherited historical working-ID warning remains in the record; this edition does not conceal it or reuse those old IDs.

The public atlas is a frozen release view. Source articles and the living Carrier can later change. A correction should therefore identify which definition, source revision or application is being challenged, so a future reader can follow the change.

Attribution and the next stretch of work

The Wikipedia-derived adaptations in this new subset are distributed under CC BY-SA 4.0, with article revision and contributor-history attribution and a notice of modification. Those terms do not silently replace the terms of earlier Carrier records, software or unrelated media. Preserve the accompanying attribution when reusing the derived subset; the Wikipedia reuse guidance describes the source framework.

At 4,089 checked concepts, Carrier has reached 40.89% of the numerical 10,000-concept goal; the distance remaining is 5,911. This edition supplies no reliable books-per-concept or articles-per-concept forecast. As coverage expands, overlap increases and the most useful gaps change. A fixed quota would reward adding names even where reusing or refining a definition serves expression better.

The next milestone should include use: compose messages that need these distinctions, inspect what is still difficult to state, and let readers identify what was underspecified. A larger dictionary becomes more useful through that cycle of expression, interpretation and correction.

Release JSON Release manifest Measured analytics Evidence and audit package Retained source archive · about 233 MB

Edition record: source and editorial work were frozen on 4 October 2026. The supplied review export records approval at 2026-10-04T14:19:57.044Z; the approved release was prepared at 2026-10-04T14:27:33.625Z. Later changes to Wikipedia or Carrier do not silently alter this snapshot.