ULR Claim Ledger — Canon 24's 54 Active Rows

Distinguishing the counters#

Canon 24 contains two different counts.

54 is the number of rows in the current active claim table. 50 is the cumulative counter in the change history of a particular carrier/gauge programme. They differ because several early empirical atoms were not included in that change-history counter.

The authoritative LEDGER owns the detailed English status labels and refutation conditions. This page enumerates every row active in Canon 24 and explains what each cluster does—and does not—establish.

E0–E5 — Six empirical atoms#

E0 ULR-CONV-EXIST#

For a limited set of concept/model pairs, relational geometry exceeds a shuffled null. Shared alignment is supported for those pairs; an observer-independent common space is not.

E1 ULR-CONV-INDEP#

Independent vision-only × language-only pairs also exceed chance. Joint multimodal training therefore cannot explain the entire phenomenon, but this does not imply that every modality converges into one space.

E2 ULR-CONV-DIM#

20 of 25 CIFAR dimensions and 24 of 25 Caltech dimensions exceeded the held-out null. The result supports a multidimensional phenomenon, but it does not identify 17, 20, or 24 as an exact intrinsic dimension.

E3 ULR-CONV-SEM#

Alignment has a graded association with the selected semantic/language proxies. Proxy correlation must not be promoted to identity with meaning.

E4 ULR-CONV-ORIGIN#

The statistic strengthens during training from a near-chance initialisation, with a typed trend of $\rho=0.881$. This is evidence of a learning trend, not of an exact formation time.

E5 ULR-EMP-GAUGE-ROBUST#

The T1/T2 exact results and T3 trend are preserved under the specified permutation × positive-diagonal typing. This excludes an explanation based solely on raw-coordinate artefacts, but does not establish robustness under general GL, nonlinear, or full-model gauges.

G0 — Four architecture and observable-typing claims#

G0 excludes identity at the level of raw coordinates or raw geometry. The separate claim that meaning exists after quotienting does not follow.

G1 — 12 node/local-gauge claims#

Transformations permitted by local equations do not yet constitute a global function gauge. Lifting and gluing are tested separately in the next layer.

G2 — Seven assembly, lifting, and gluing claims#

The central point of G2 is

These atoms make no direct claim about semantic organisation.

G3 — Four global-completeness and degeneracy claims#

G3 establishes conditional global completeness and generic infinitesimal completeness for the specified minimal block. It does not classify the maximal gauge of every Transformer.

G4 — 10 function-fibre and quotient-rigidity claims#

G4 shows that a same-function fibre can be decomposed into gauge and residual components. It does not include a bridge assigning semantic or circuit identity to the residual.

G5 — Three generic/assembly-evidence claims#

A generic open set, a pilot case, and a real-weight battery occupy different evidence tiers. One pilot certificate must not be promoted to full neural universality.

G6 — Four typed-UAR-complex claims#

These four atoms do not mean that “the UAR monomorphism is complete.” The channel map and assembly map are different maps.

O0–O2 — Three observer-relative role/port claims#

O0 ULR-ROLE-PORT-SEPARATION#

A persistent writer and a cache writer with the same passive trajectory and intact output separate under the declared reset/clone response $HRb$. The response is preserved under GL co-transformation.

The permitted conclusion is that a role unidentifiable through passive observation can become identifiable through a richer physical interface. This is not a universal role ontology for real Transformers.

O1 ULR-PERSISTENT-SUBSPACE-IDENT-BOUNDARY#

For reset response $Y_R=HRB$, a full-column-rank read port $H$ and a full-row-rank preparation port $B$ recover $R$, whereas partial ports determine only $HRB$. Even matrices $R$ of different rank can produce the same partial response.

The identified object is not $R$ itself, but $[R]_{H,B}$.

O2 ULR-OBSERVER-ROLE-DISCRIMINATION#

Once a role $C\in\{L,I\}$, observer filtration $\mathcal F^O$, and transcript laws $P_L,P_I$ are declared, exact deterministic discrimination is possible if and only if $C$ is $\mathcal F^O$-measurable. For equal priors, Bayes error is characterised by total variation:

Adaptive distinguishability over admissible policies $\Pi$ is characterised by $\sup_{\pi\in\Pi}\lVert P_L^\pi-P_I^\pi\rVert_{\mathrm{TV}}$. Under a fixed infinite policy, finite-prefix error tends to zero if and only if the induced full path laws are mutually singular.

N0 — One small-neural application#

ULR-NEURAL-ROLE-PORT-ABSORPTION#

In a 32-seed engineered family, the same learned neural core is connected either to recurrent hidden state or to an attention cache. Passive outputs and activations are pairwise exact, while reset/clone responses separate for every seed. The typed response $Q$ predicts held-out partial-clone transfer better than the passive predictive baseline.

However, raw port-response B3 and declared-routing B4 make nearly identical predictions, so the preregistered nonabsorption gate is not passed. The neural realisation of the exact toy remains an active result, but it does not establish a neural-specific object, learned organisation, a large-model law, or a ULR ontology.

Retracted claim excluded from the active 54#

ULR-ASSEMBLY-UAR-MONO is RETRACTED-MALFORMED-COMPOSITE and is counted neither among the 44 gauge/assembly atoms nor among the active 54.

It was retracted because injectivity of the full channel incidence and unramifiedness of the joint assembly quotient were combined into one monomorphism even though they have different sources and maps.

This provenance remains visible in Canon Evolution.

How to read the 54 in one statement#

The ledger's final conclusion is: preserve the strong phenomena and theorems, but do not introduce a distinct neural-specific ontology on the present evidence.