When “one” needs more than one meaning: findings from Carrier’s second Wikipedia wave

The most useful question for this second Wikipedia wave was a practical one: when a message needs a distinction, can Carrier name it precisely enough for another reader to inspect what was meant?

That question led to 325 approved additions and a checked vocabulary of 4,414 concepts. It also led to examples, exclusions, reused entries and deferred senses. The result is a larger dictionary accompanied by a more explicit account of what its new vocabulary can contribute.

1. Start with an expression

A growing vocabulary can accumulate plausible labels without showing why another entry is useful. This edition worked with concrete communication needs: describe a physical balance; qualify a model; compare accounts of unity; attribute a belief; record an approval; explain what a software check establishes; or make a public institution open to correction.

The earlier Carrier and ordinary prose could already express many of these distinctions. That matters. The intended gain is often a stable, reusable reference for a commitment that would otherwise need to be explained repeatedly. A new entry earns its place through a clear scope and a useful difference from its neighbours.

Three connected strands developed those needs: physical and mathematical systems; philosophical and comparative understanding; and the language of evidence, data and institutional checking. Together they cover 14 editorial domains. The labels organize browsing; they do not claim that knowledge divides naturally into exactly fourteen parts.

2. “One” can make several different claims

One particularly useful family concerns unity. Existence monism concerns how many entities strictly exist. Priority monism concerns what is fundamental, while allowing derivative entities. Substance monism concerns unity of substance or of its basic kind. These are different commitments, even when an informal sentence calls all three “oneness.”

The definitions therefore state what is being counted or given priority. Their source trail includes the retained discussion of monism and the substantive priority-monism account in the Schaffer article. Neither becomes an empirical conclusion merely by entering the dictionary.

The practical example is straightforward: independent participants can exchange records through a common protocol while disagreeing about ontology. Shared reference makes the disagreement more addressable. It does not settle it.

The same discipline applies across religious and philosophical traditions. Tawhid, a particular nondual teaching and a priority claim about the cosmos cannot be equated through the English word “unity.” Comparison needs the particular tradition, rejected distinction and type of claim. The atlas keeps those differences available for inspection.

3. Physical descriptions need boundaries

The physical strand makes several earlier Earth-and-space questions easier to formulate. A radiator depends on emission, absorption, surface properties, geometry and the surroundings it can exchange energy with. “Space is cold” supplies too little information to size a radiator or demonstrate a thermal balance.

The authored radiator message now selects emissivity, radiant exitance, view factor, waste heat and radiative equilibrium, while explicitly asking for the missing temperatures, geometry and material data. It reports no flight design or measured performance.

A second example concerns a water tank. An unchanged level can mean inflow equals outflow. It does not establish that the tank is isolated, that no material moves through it, or that its water is suitable for use. A control volume, a balance equation and the stock–flow distinction preserve the physical bookkeeping; access and quality remain separate questions.

These are useful kinds of precision because they expose missing information. They keep a compelling diagram from quietly acquiring the authority of a completed calculation.

4. The meaning of a successful check

The computing additions distinguish data integrity, data quality and data validation. An unchanged record has one property. Conformance to declared rules has another. Fitness for a particular use requires further judgments.

The retained data-validation source contains a useful body qualification: accepted data can still contain errors. The compact definition retains that limit. It avoids upgrading a structural check into a general guarantee of accuracy.

A related issue appears in RDF Schema. Domain and range statements can participate in type inference; an application may separately choose rules for rejecting inputs. The source’s worked example supports that distinction, while another passage makes a conflicting broad statement. The editorial record preserves which support was selected instead of treating all paragraphs as equally consistent.

Software testing needs comparable care. Code coverage records which specified program structures were exercised. A test oracle supplies the expected behavior against which an execution is judged. Mutation testing asks whether chosen artificial changes are detected; a surviving change may be behaviorally equivalent rather than an undetected defect. These measures have different denominators and limits. See the retained Code coverage, Test oracle and Mutation testing sources.

A collection of passing checks can therefore be informative without becoming a certificate for every interpretation or downstream use. The new vocabulary makes it easier to say precisely what was checked.

5. Ordinary words conceal important choices

Consider a hypothetical message: “The release may proceed.” It might mean that success remains possible under available information. It might state permission under an authorized procedure. It might concern possibility in a logical model.

Epistemic, deontic and alethic modality give those commitments separate references. The surrounding message still has to identify the speaker, evidence, authority, conditions and model where relevant. Selecting a word for permission does not grant permission.

Attribution adds another distinction. Reporting that someone believes a claim does not assert the claim. “Does not believe P” and “believes not-P” are different contents. A concept token can identify an attitude or a scope rule, while the composition specifies whose attitude and which proposition are involved.

This is why expression cannot be reduced to a list of related concepts. The roles, conditions, negation and authorial act carry meaning too. A useful dictionary supports that structure; it does not supply it automatically.

6. Nineteen authored comparisons, thirty-eight real messages

The use lab pairs messages using the earlier vocabulary and explicit prose with messages selecting the new distinctions. Each pair declares its expected distinction, possible misleading readings and limits. Some later messages extend or clarify the earlier scope; they are not all asserted to be exact paraphrases.

These local results demonstrate byte and reference recovery under the supplied implementation. They do not measure whether another AI understood the messages better, whether every paired text means exactly the same thing, whether compression improved, or whether a reader accepts the proposal.

The public download is a checked-release rendition of those messages. Their authored text is preserved, while the reference pins now select the approved entries at this release’s checked head. The metadata and packet bytes therefore change. Publication recovery checks validate these new files; they are a local rerun, not a second independent receiver trial.

Open the public use lab to compare the actual texts and inspect the files.

7. Source work includes saying “this is narrower”

The edition uses 1,037 distinct selected passages, with 1,046 occurrences across its definitions, from 315 article revisions. The retained collection contains 331 captures. Those are different counts: selected evidence, repeated use of evidence, used articles and retained articles should not be collapsed.

Retaining an article gives future readers context and a route back to the source. It does not establish that every passage was inspected, that Wikipedia is authoritative on every question, or that a compact adaptation exhausts its subject.

Several useful decisions came from following the actual source rather than the requested title. “Data catalog” led to the specific Data Catalog Vocabulary, so the definition names that RDF vocabulary. “Property testing” concerned query-limited approximate algorithms, not ordinary software tests generated from properties. A test-coverage lead resolved to black-box testing; a separate code-coverage source was retained for the structural metric.

Other decisions concerned strength. The Goodhart entry treats target-induced distortion as a risk to investigate, rather than a theorem that every measure becomes useless. Participatory evaluation names influence over questions and judgments while preserving the problem of unequal power. A well-organized safety argument remains an argument requiring support. These qualifications follow the retained Goodhart's law, Participatory evaluation and Goal Structuring Notation discussions.

8. Reuse and deferral are results

The earlier wave left 66 distinct title leads for further investigation. This edition supplies new selections for 60, reuses existing definitions for three and defers three with reasons. The reused records cover endosymbiosis, engineering resilience and semantic entailment under their established Carrier names.

The remaining leads illustrate different kinds of incompleteness. A dedicated moral-uncertainty source was not retained under the requested titles. Generic permission needed a selected sense. The requested reconciliation source was unavailable. Related ideas can support a discussion without becoming identical replacements for those missing selections.

Within the new drafting work, material conditional and satisfiability were also recognized as already represented. Their initial candidates were removed and the existing checked references reused. A language grows more coherently when it can decline an unnecessary synonym.

Inspect the earlier-lead ledger for the individual outcomes and their scope.

9. A richer map is still a map

The release adds 1,427 stored contextual links: 931 connect new concepts to earlier checked records, and 496 connect new concepts to one another. In total, 432 earlier concepts gain new neighbours, and 76 of the new-to-new pairs cross the edition’s domain boundaries. The editorial relation list contains 1,667 authored edges before its contextual pairs are deduplicated for storage. These figures describe different stages of the relation process.

Every new concept has a path into the earlier checked vocabulary. The combined graph still has 52 connected components, with 4,361 concepts in its largest component. It should therefore not be described as one completely connected graph.

More fundamentally, graph reachability is not shared meaning. A link can connect a position to an objection, a method to its limitation, or two accounts that should be compared carefully. High degree may reflect broad usefulness, age or editorial practice; it is not a truth or importance score.

10. What the approval means, and what comes next

Lucid approved the 325 compact definitions and their displayed contextual links. This moves them into the checked tier of the prepared release. It does not approve every future application, certify an entire source article or establish agreement among specialists.

The checked vocabulary has increased by 7.95% in this wave. At 4,414 entries, it reaches 44.14% of the stated 10,000-concept milestone; another 5,586 checked concepts would be needed to reach that count. That is arithmetic, not a forecast of how quickly useful and sufficiently distinct entries can be found and reviewed.

The more demanding next question is expressive: which messages still require lengthy qualification, which distinctions recur, and where can a recipient recover the intended structure yet reasonably challenge its application? Those questions can guide further enrichment without turning the entry total into the only measure of progress.

A deeper vocabulary gives a reader more ways to understand what was said—and more precise places to disagree.

Method boundary: AI-assisted selection and drafting; retained, revision-pinned sources; source-bound compact definitions; contextual relation review; human approval; local USL1 composition and recovery checks. This edition reports no independent interpretation trial or general proof of semantic improvement. Wikipedia-derived adaptations are shared under CC BY-SA 4.0 with attribution; earlier records and separate dependencies keep their documented terms.

Release JSON Measured analytics Release manifest Evidence package Retained source archive Checked expression lab