CARRIER — ORGANIC REVISION 01
Let the language find its neighbours.
Local research record · 2 October 2026

The first organic revision makes the existing vocabulary more inspectable and
more connected. Its most useful finding is that growth can happen in relations,
expressions and review methods before it requires more dictionary entries.

WHAT IS COMPLETE

The authorised fourth working entry, w2418 — “historical description, current
allocation” — is now checked/2442 locally. There are 2,443 checked concepts
and three retained original working examples. The promoted definition and colour
are unchanged; the earlier checked file remains an exact prefix. Historical
expressions retain their old snapshot pins. No online state was changed.

Every checked definition received an AI reading, an individual semantic note and
one to three family assignments. The definitions total 83,948
whitespace-delimited words. Three reviewers covered disjoint ranges and a further
pass checked overlaps, directional claims, evidence and proposed novelty.

This is full definition coverage, not an exhaustive rereading of every source
book or adjudication of all 2,982,903 possible concept pairs.
The notes, exact endpoint quotations and qualifications remain inspectable.

The resulting desk contains:
  • 22 overlapping editorial families;
  • 40 curated reading threads, plus a shortest-path explorer;
  • the complete recorded “Referenced by” index, including both tiers;
  • 389 proposed typed relations across 387 new neighbour pairs;
  • a map of every checked concept and selected-neighbour distance circles;
  • 12 consolidated candidate distinctions from 18 retained drafts;
  • a constrained composition lab and 12,000 synthetic navigation requests.

The families, threads and new links are AI interpretations. They do not become
human-reviewed assertions merely because their endpoint definitions are checked.

WHAT THE MAP REVEALS

The recorded checked graph contains 5,468 directed links.
1,302 concepts have no recorded incoming checked reference;
77 have no outgoing checked reference. There are
71 completely isolated checked concepts.

These are different conditions. A concept may point to others without receiving
a link in return. “Referenced by” is derived from actual stored directions; it is
not manufactured by declaring every old link reciprocal. A reverse navigation
link does not transfer endorsement or the claim of the referencing record.

The oldest broad method concepts frequently receive references from later
source-specific concepts. “Inspectable assumption” has 310 recorded incoming
links. This pattern is consistent with the history of incremental enrichment;
incoming degree alone neither proves age caused the pattern nor ranks semantic
importance. Recent narrow concepts can remain useful with very few links.

With all proposed links included for undirected navigation, neighbour pairs rise
from 5,468 to 5,855.
Connected components fall from 72 to 62;
isolated concepts from 71 to 60;
the largest component grows from 2,372 to 2,380 concepts.

Two pairs support two typed readings each. Those are 389 relation proposals,
not 389 different pairs. The local Carrier projection adds
752 directed references across
612 records because symmetric navigational
types are materialised in both directions. Directional relations retain their
direction. Their types and limits remain in the accompanying evidence layer.

SOME CONNECTIONS WORTH READING

“The honest instrument” and “scientific testimony independent of desired
conclusions” meet around evidence that can disappoint a preferred conclusion.
Their methods are not identical: a tool's rejection condition and the credibility
of supported inquiry operate at different levels. The new link makes that
comparison easy without collapsing the definitions.

“An invitation that can be entered” connects a standing open invitation with
actual welcoming conduct and respect for a newcomer's independent commitments.
This moves an abstract principle toward the conditions under which participation
is usable. The thread remains a proposed reading, not proof of universal agreement
among the sources from which its concepts were drawn.

“Past statements, present references” links preserved correction, current
allocation, stable naming and recoverable measurement context. It gives the
recent colour-allocation repair a broader role: a historical statement can be
preserved while a present operational reference is corrected.

“Compact messages and their retained costs” connects distinguishable symbols,
segmentation, shared codebooks and reconstruction checks. It encourages an honest
account of what a short message depends on, not just a count of transmitted bytes.

Differences matter as much as similarities. A theological preservation claim and
an edition-pinned citation can be placed in a qualified comparison; the first is
not evidence that every manuscript or translation received the second's checks.
The review deliberately retains this boundary.

WHAT THE SIMULATION ESTABLISHES

Seed: 20261002. There are 128 sampled origins and 2,431
distinct destinations. Each workload has 4,000 requests. Within-family pairs
share at least one assigned family; cross-family pairs share none.

uniform: reachable 3,808 → 3,844 / 4,000; mean steps among each graph's reachable requests 4.428 → 4.286; 95th percentile 7 → 6.
within-family: reachable 3,765 → 3,801 / 4,000; mean steps among each graph's reachable requests 4.19 → 4.006; 95th percentile 7 → 6.
cross-family: reachable 3,804 → 3,846 / 4,000; mean steps among each graph's reachable requests 4.456 → 4.326; 95th percentile 7 → 6.

Because reachability changes, those per-graph means do not compare exactly the
same set of requests. The paired measure is cleaner: among
11,377 requests reachable in both graphs, the mean
saving is 0.1584 steps.
1,452 requests acquire a shorter path; 114 become reachable.
Adding edges cannot lengthen a shortest path, so zero longer routes is expected,
not a separate scientific discovery.

A stress case removes the five highest incoming-degree concepts. It excludes
requests whose endpoint is removed and reports the remaining denominator. Paths
become longer and some disappear, showing structural dependence on broad hubs.
It does not simulate censorship, cognition or an actual service outage.

All relations are traversed in either direction for navigation, including
directional semantic proposals. This does not reverse the asserted meaning.
Ties use ascending IDs; traffic is one chosen route, not full betweenness or a
measure of intellectual value. Traffic for a multiply typed pair reports all its
proposal IDs rather than arbitrarily crediting one interpretation.

These are simulated journeys, not observed users, semantic comprehension tests,
or evidence of demand. Adding links and observing shorter paths cannot validate
those same links. The graph provides places to investigate, not autonomous proof
that it needs a particular new concept.

THREE DIFFERENT KINDS OF ABSENCE

A missing link: the concepts exist, but their useful relationship is unrecorded.
A missing review: the candidate connection has not been sufficiently examined.
A missing meaning: an attempted expression loses a necessary distinction even
after existing vocabulary and clear composition have been tried.

Only the third is a strong reason to consider another dictionary entry. A blank
map area cannot distinguish the three by itself. The 60 still-isolated
concepts remain listed with their review notes in unconnected-concepts.json.
They are not force-connected to make the picture look complete.

SYNTHESIS WITHOUT INFLATION

The chemical-elements analogy is useful when it encourages explicit structure:
which concepts participate, what role each has, and which conditions govern the
composition. It becomes misleading if a combination automatically inherits truth,
approval or novelty from its ingredients. The composition lab therefore produces
an unallocated draft, with pinned parents, operation, scope, proposed distinction
and counterexample. Its checks validate form; meaning still requires judgment.

Eighteen independent draft ideas were consolidated into twelve. Several are
better treated initially as protocol fields or compositions of existing entries.
Versioned relation evidence, relation correction and dependent-expression review
can form one maintenance workflow before requiring three new dictionary nodes.
No candidate was assigned a colour, inserted into a tier or promoted by simulation.

G001 — reverse reference without inherited commitment
A reverse-reference view reports that another record points to this concept; it does not make this concept assert, endorse, exemplify or inherit the other record’s claims. The original edge type, direction, version and qualification remain visible when navigation is offered in both directions.
Review treatment: Test the distinction against existing expressions before proposing admission.

G002 — composition keeps the trust of its parts separate from its assertion
An expression may use reviewed definitions while proposing an unreviewed relation among them. Its record preserves both the status of each referenced component and the status of the new assertion, preventing checked vocabulary or successful decoding from silently certifying the composition.
Review treatment: Test the distinction against existing expressions before proposing admission.

G003 — a navigable thread distinguished from an inferential chain
An ordered route through related concepts can aid discovery without entailing its final claim from its first. Each inferential step needs its own relation type, direction, premises and scope; navigation alone does not supply transitivity, causation or endorsement.
Review treatment: Test the distinction against existing expressions before proposing admission.

G004 — absence qualified by semantic review coverage
A report of a missing conceptual connection that records which definitions, relation kinds and use cases were actually examined. It distinguishes no observed edge, no justified edge after review, and an unrepresented distinction, so a blank graph region does not by itself become evidence of a missing meaning.
Review treatment: Test the distinction against existing expressions before proposing admission.

G005 — a claimed semantic gain tested against an existing composition
A proposed concept earns a separate place when a specified use cannot preserve the required distinction through an existing concept or a clear composition of existing ones. The test names the lost distinction and a counterexample, rather than treating novel wording as evidence of new meaning.
Review treatment: Test the distinction against existing expressions before proposing admission.

G006 — relation correction with retained endpoints
A relation can be withdrawn, redirected or qualified while the concepts it connected keep their identities. The revision records why the connection changed and which expressions relied on it, preserving old snapshots without treating the old inference as current.
Review treatment: Prefer a protocol bundle with endpoint warrant and dependent review before proposing three separate entries.

G007 — an edge with a declared inference boundary
A connection between concept records that states its type, direction, evidence and permitted use, together with what must not be inferred from it. Navigation, analogy, specialization and operational application can then share a graph without silently licensing the same conclusions.
Review treatment: Reuse existing reference-role and qualified-composition concepts first; this may be a protocol field.

G008 — premise review propagated without erasing the dependent claim
When a referenced premise or relation changes, dependent expressions are marked for renewed examination while their original wording and pins remain preserved. A changed dependency triggers a review obligation, not an automatic verdict that every descendant is false or a silent rewrite of the historical expression.
Review treatment: A changed premise calls for review, not automatic falsification. Separate admission requires a failed-use witness.

G009 — a boundary case that limits a semantic bridge
A deliberately chosen case in which a proposed cross-context relation stops preserving its claimed distinction. Recording this case alongside a useful bridge makes its reach testable and helps separate shared structure from meanings, authorities or causal mechanisms that must remain local.
Review treatment: Reuse-first: bounded certainty plus retained comparison boundary plus a discriminating test may already suffice.

G010 — a relation claim pinned to the meanings it compares
A claimed relation between concepts is assessed against specified versions of both definitions. If either endpoint changes, the relation is retained as historical evidence but rechecked before being presented as applying to the new meanings; unchanged labels do not renew its warrant.
Review treatment: Keep comparison claims pinned to both definitions. Prefer a bundled maintenance protocol until distinct uses justify a separate entry.

G011 — a simulated need distinguished from an encountered use
A scenario-generated failure to express a distinction can nominate a gap for investigation, while remaining separate from evidence that real users encounter it, how often they do, or whether a proposed concept resolves it. Simulation design and acceptance tests remain part of the claim.
Review treatment: Compare represented possibility, reconstructed rather than observed structure, selected successes and independent test data; the candidate adds the demand-evidence context.

G012 — a missing distinction with preserved expression witnesses
A proposed vocabulary gap is supported by retained attempts to express a particular distinction, including existing concepts tried, their failed or misleading readings, and the new boundary required. Low connectivity alone is not evidence that a concept is missing.
Review treatment: Reuse-first: build an admission evidence record from reader-discovered gaps, reproducible handoff and existing failure evidence before creating a new node.

A DISCIPLINED ORGANIC CYCLE

1. Preserve a real attempted expression and the distinction it needs.
2. Try the existing concepts, a new connection, and an explicit composition.
3. Identify exactly what remains lost; keep an alternative reading or counterexample.
4. Propose the smallest reusable distinction that repairs that loss.
5. Review its evidence, limits, neighbours and duplication risk.
6. Allocate only after admission; preserve earlier expressions and snapshots.
7. Observe later use and correction rather than treating admission as permanence.

This cycle can support organic enrichment under human judgment. It does not
establish that a language is biologically alive or that new linguistic laws have
been proven. The practical achievement is already substantial: its connections,
revisions and proposed growth can now be inspected together.

REVIEW AND TECHNICAL RECORD

Open review/organic-revision.html. The promotion-only state and AI-proposed linked
edition are separate archives. Review decisions pin both the original checked
head and the entire proposal set; an R-number alone is insufficient because a
rebuilt proposal list could reassign it. The tool rejects mismatched decision
files instead of applying an approval to a different claim.

Independent checks recompute the graph, quotations and full reverse index. The
linked-state audit compares every record and validates raw hash chains. Synthetic
decision tests check wrong pins, altered proposals, duplicates and unknown values.
Browser checks cover desktop and narrow mobile layouts, navigation and draft
rejections. These checks establish specified behaviour, not semantic correctness.

The historical LEGACY_ID_HISTORY warning remains. The local states preserve it.
Nothing in this revision authenticates a reviewer or changes live publication.

Relation/provenance references:
https://www.w3.org/TR/skos-reference/
https://www.w3.org/TR/prov-o/
The custom layer is experimental and does not claim full conformance to either.

The full ledger, raw drafts, consolidation decisions, exact proposal evidence,
deterministic workload results and reproducible builders accompany this report.
