Severance Defect and the Binding Functional

Maksim Barziankou (MxBv)
Navigational Cybernetics 2.5 · The Urgrund Laboratheory
[e-mail]
This work DOI: 10.17605/OSF.IO/4NMTW
Axiomatic Core (NC2.5 v2.1): DOI 10.17605/OSF.IO/NHTC5
Formal apparatus: ONTOΣ XV, bundle DOI 10.17605/OSF.IO/EAUD5

Observation-Relative Non-Reconstructibility and the Impossibility of Uniform Exact Absorption under Confined Markov Access

Formal companion to Nothing Is Solid (Through a Life VIII). The note separates functional severance from reconstructive binding, defines the two-coordinate severance profile, derives an exact entropy decomposition for coupled assembly, and proves that positive binding H(Z|Y)>0 forbids uniform exact absorption when upper-layer access is confined to the declared Markov channel. A sealed capacity witness and executable standard-library verifier make the distinction operational

Formal note. Grounded in the axiomatic core of Navigational Cybernetics 2.5 v2.1 (DOI 10.17605/OSF.IO/NHTC5) and in the categorical apparatus of Cross-Layer Forgetful Separation: A Categorical Foundation (bundle DOI 10.17605/OSF.IO/KAGMH).

§0. Status of this note

This note is deliberately narrow, and its honest ledger is placed first.

What is new here. The note separates two questions that are easy to conflate:

1. functional severance: whether excising a component destroys the declared operation of the remainder;
2. reconstructive severance: how much uncertainty about the completed structure remains after the component is hidden.

The first is recorded by a severance defect. The second is recorded by a binding functional. Their ordered pair is the severance profile. The note then gives an exact entropy decomposition for coupled assembly (Theorem 6.2), a finite calculation in which the compatibility correction is evaluated rather than merely named (§6.4), an impossibility of uniform exact absorption when average binding H(Z∣Y)>0 under confined Markov access to the declared channel (Theorem 7.2), and a complete sealed capacity audit with a target-fibre witness, two sealed off-diagonal audits separating the axes, and an executable verifier (Appendix A).

What is not new in kind. Conditional entropy, conditional mutual information, fibres, reconstruction maps, and the data-processing inequality are standard. The contribution is the architectural use of those objects and the discipline that makes the resulting score auditable.

What is extracted from NC2.5. The non-reconstructibility bound NR-ε supplies a quantitative lower bound on average binding when the protected variable, its relation to the completed substrate, the observation channel covered by NR-ε, and a prior whose entropy exceeds the leakage budget are declared. If the protected variable is a coordinate of the full completion, its residual entropy lower-bounds the full-completion binding rather than being identified with it. NR-ε does not, by itself, identify the entire tuple (X, D, Φ, C) with that observation channel, and it does not prove pointwise non-reconstructibility of every substrate in a comparison class.

What is declared, not derived. The component decomposition, comparison class, prior, observation channel, excision map, and functional criterion are sealed declarations. The resulting quantities are declaration-relative. This is a limitation of the method, not a theorem to be hidden in notation.

What is conjectured. The relation between static binding and the dynamic maintenance ledger Φ^{(M)} remains open (§9).

What is not claimed. No formalisation of structural gravitation, attraction, reach, or capture is offered. Those are images in the companion essay, Nothing Is Solid (Through a Life VIII). The essay motivates the question; it is not a premise of any proof below.

§1. The question

The companion essay proposes a severe practical test: remove a named component and ask whether the remainder merely loses a feature or ceases to be the same kind of object. That is a question about functional dependence.

A different question asks whether the missing component can be reconstructed from what remains visible. That is a question about information.

The two questions are independent. A component may be functionally indispensable but uniquely reconstructible from the remainder. Another may be functionally optional yet impossible to identify after it has been hidden. A single scalar cannot distinguish those cases without importing an additional convention.

The formal question of this note is therefore two-dimensional:

Under a sealed decomposition and observation protocol, what does excision destroy, and what does forgetting make unrecoverable through the declared observation channel?

The answer yields a criterion under which an upper layer whose access is confined to a declared observation channel cannot achieve uniform exact absorption of the sealed floor class when H(Z∣Y)>0 (Theorem 7.2; Corollary 7.3 under NR-ε). Whether the resulting interface supports stable composition is a separate question (§12.6).

§2. Audit declaration

2.1 The substrate

Following the cross-layer companion (Definition 2.1 there), a substrate is a tuple

𝔖 = (X, D, Adm, Φ, C),

where X is a state space, D is an admissible-transition structure, Adm is the out-of-loop admissibility predicate over trajectories, Φ is the monotone irreversible structural burden, and C is the structural capacity, with viability budget τ = C  -  Φ.

Standing assumption (imported from the cross-layer companion). On admissible trajectories, Φ is monotone under extension and τ = C-Φ is read from the declared completion. This note does not reprove well-formedness of 𝔖; audits assume the tuple matches the companion’s substrate definition.

2.2 Components and comparison class

Definition 2.2 (Sealed audit declaration). A severance audit for a target substrate 𝔖 consists of

𝒜 = (C, Ω, π, {q_c, e_c, P_c}{c ∈ C}, {w_c}{c ∈ C}),

declared before scoring, where:

1. C = {c_1, …, c_n} is a finite decomposition into components;
2. Ω is a set of admissible completed presentations containing 𝔖, quotiented by an isomorphism criterion stated explicitly as part of the declaration;
3. π is a full-support prior on Ω;
4. q_c :Ω → 𝒴_c is the observation channel after the c-data have been hidden;
5. e_c is an excision operation, possibly returning an ill-typed symbol ⊥;
6. P_c is a declared binary criterion on completed or excised presentations, with P_c(𝔖) = 1;
7. w_c ≥ 0 is an optional declared component weight.

Notation suppresses equivalence-class brackets when no confusion can result. The channel, excision operation, and criterion must respect the declared isomorphism relation.

For finite Ω, the auditor must supply an operational isomorphism criterion: for any two candidate presentations in the declared class, it must determine whether they represent the same element of Ω. This is a declaration obligation, not the existence of a canonical quotient.

The finite case is the default. Continuous and stochastic channels require regular conditional distributions and are left to Open Problem 12.1.

Remark 2.3 (WhyΩis necessary). The raw inverse image of a forgetful functor may be a proper class, and its cardinality is not invariant under replacing an object by isomorphic copies. Restricting to a declared comparison class of isomorphism classes makes the fibre a set and makes the score reproducible. The price is relativity to Ω.

Remark 2.4 (No properties masquerading as independent components). A property logically determined by visible data is not an independently forgettable component. If Φ remains visible, then monotonicity of that same Φ can be checked from Φ; one may not create a second candidate by changing monotonicity while claiming the forgotten presentation is unchanged. To audit monotonicity independently, the declaration must specify a different observation channel that hides the relevant part of Φ or its ordering data.

2.3 Sealing discipline

The declaration is sealed in the sense of the companion’s Remark 2.7: it may not be revised after the target outcome is observed. In particular, the auditor may not change Ω, enlarge or shrink the channel q_c, alter the prior, or replace the criterion P_c in order to obtain a preferred verdict.

The declaration must also state whether metadata such as a component ledger remains visible. If a removed value is copied into visible metadata, that metadata is a side channel and belongs to q_c. It cannot be used silently as a reconstruction oracle.

2.4 Notation collisions

The word atomic remains confined to the companion essay. In the formal NC2.5 corpus an atom already has a measure-theoretic meaning in the Drain/Snap regime. This note uses non-severable and binding instead.

The symbol Δ_τ remains reserved for the kernel defect of the admissibility-transfer apparatus. The severance defect below is written δ_c.

§3. Severance profile

3.1 Functional severance

Definition 3.1 (Severance defect). For a sealed audit 𝒜 and target 𝔖,

δ_c^𝒜(𝔖) = 1  -  P_c(e_c(𝔖)) ∈ {0, 1}.

Thus δ_c = 1 means that excision destroys the declared criterion; δ_c = 0 means that the excised remainder still passes it. If e_c(𝔖) = ⊥, the convention is P_c(⊥) = 0.

This is the formal counterpart of the essay’s remove-and-ask-whether-the-remainder-dies test. It is not an information measure.

3.2 Observation fibre

Definition 3.2 (Observation fibre). For y ∈ 𝒴_c,

Sev_c^𝒜(y) = {[𝔖’] ∈ Ω: q_c(𝔖’) = y}.

For the target, write Sev_c^𝒜(𝔖) for the fibre at y = q_c(𝔖).

3.3 Binding functional

Let Z be the Ω-valued random substrate with law π, and let Y_c = q_c(Z).

Definition 3.3 (Pointwise and average binding). The pointwise binding at an attained observation y ∈ q_c(Ω) with P(Y_c = y)>0 under the sealed joint (Z, Y_c) is

𝔅c^π(y) = Hπ(Z ∣Y_c = y).

The average binding is

𝔅̅c^π = Hπ(Z ∣Y_c).

Under the finite deterministic default, if π is uniform and y ∈ q_c(Ω),

𝔅_c^{unif}(y) = log_2 |Sev_c^𝒜(y)|,

the Hartley-entropy special case. All logarithms in the finite examples below are base two.

Proposition 3.4 (Uniform reconstruction criterion). Under a full-support prior on finite Ω, the following are equivalent:

1. every observation fibre over q_c(Ω) is a singleton;
2. q_c is injective on Ω;
3. there exists a reconstruction map R_c :q_c(Ω) → Ω such that

R_c ∘ q_c = id_Ω;

1. 𝔅̅_c^π = 0.

Proof. Take finite Ω, full-support π, and Z the canonical Ω-valued random element with law π. Then (1) and (2) are equivalent by definition of a fibre. If q_c is injective, its inverse on its image is R_c, giving (3); conversely, (3) implies injectivity. Under full support on finite Ω, H(Z∣Y_c) = 0 iff Z is almost surely a function of Y_c, which is exactly (3). □

Remark 3.5 (Why no section appears). For a map q :Ω → 𝒴, reconstruction requires a left inverse R with R ∘ q = id_Ω. A right inverse, often called a section, may exist even when fibres contain many elements. Moreover, for one fixed target a constant map can always return that target. The non-trivial statement is therefore uniform reconstruction over the declared class, not pointwise selection at one object.

3.4 The two-axis profile

Definition 3.6 (Severance profile). The component-wise severance profile is

𝒫_c^𝒜(𝔖) = (δ_c^𝒜(𝔖), 𝔅_c^π(q_c(𝔖))).

The two coordinates have four distinct readings:

Four distinct readings of the two coordinates.

The strong formal analogue of the essay’s binding image is the fourth quadrant. Neither coordinate alone establishes it. Appendix A.6 realizes both off-diagonal quadrants explicitly, so neither coordinate determines the other across sealed audits.

Definition 3.7 (Aggregate binding diagnostic). For declared weights w_c,

𝔅^𝒜(𝔖) = ∑_{c ∈ C} w_c 𝔅_c^π(q_c(𝔖)).

This is a diagnostic, not a joint entropy. It may double-count interactions among components and must always be reported with the declaration and the component-wise profile. The second profile coordinate is pointwise at y = q_c(𝔖); NR-ε and Theorem 7.2 use average 𝔅̅c^π = Hπ(Z∣Y_c) unless a fibre witness or pointwise bound is supplied (§8.3).

§4. The exact relation to NR-ε

NR-ε is a bound on information available through a declared downstream observation channel. It is not the statement that an arbitrary tuple obtained by deleting Adm is identical to that channel.

Theorem 4.1 (Leakage bound implies average binding). On one probability space, let (Z, Y) be jointly distributed, with Z finite and law π, and let Y be the output of the declared observation channel to which the NR-ε bound applies. If

I(Z;Y) ≤ ε, then H(Z∣Y) ≥ H(Z)-ε.

In particular, if H(Z)>ε, the average binding is strictly positive.

Proof. By the definition of mutual information,

H(Z∣Y) = H(Z)-I(Z;Y) ≥ H(Z)-ε.

□

Corollary 4.2 (Admissibility-ground instance, conditional). Let Z be the Ω-valued full completion, let Z_{Adm} = f_{Adm}(Z) be a finite admissibility-ground coordinate, and let Y_{Adm} = q_{Adm}(Z) include every downstream signal covered by the relevant NR-ε claim. Suppose

I(Z_{Adm};Y_{Adm}) ≤ ε< H(Z_{Adm}).

Then

𝔅̅{Adm}^π = Hπ(Z∣Y_{Adm}) ≥ H_π(Z_{Adm}∣Y_{Adm}) = H(Z_{Adm})-I(Z_{Adm};Y_{Adm}) ≥ H(Z_{Adm})-ε>0.

The first equality is Definition 3.3. The first inequality holds because Z_{Adm} is a deterministic coordinate of Z; equivalently, the conditional chain rule writes H(Z∣Y_{Adm}) as H(Z_{Adm}∣Y_{Adm}) plus a non-negative remainder.

This is the correct NC2.5 bridge. It is an average, channel-relative statement under the sealed joint. It does not say that every individual fibre is non-trivial, and it does not identify (X, D, Φ, C) with the downstream observation without a separate channel declaration.

Corollary 4.3 (No-state-map consequence, conditional on external result). Assume a no-state-map theorem excludes every uniform reconstruction map R on a declared comparison class. Then the corresponding state observation is non-injective and at least one observation fibre is non-trivial. To conclude that a particular target has positive pointwise binding, one must exhibit a second completion in that target’s fibre or prove a pointwise conditional-entropy bound.

Remark 4.4 (Channel completeness). A side channel changes the theorem’s input. If an upper layer observes Y’ = (Y, S), where S was not included in the NR-ε channel, then the relevant quantity is H(Z∣Y, S), not H(Z∣Y). Dependence through an undeclared side channel invalidates the audit; it does not refute the entropy identity.

§5. What the binding functional measures

𝔅_c^π(y) is the information still needed, on average under the sealed posterior, to identify the completed substrate after observing the c-hidden presentation y.

It is zero when the declared observation determines the completion. It is positive when multiple completions remain possible with positive posterior weight.

This is not a quality measure. Modularity, replaceability, and graceful degradation may all favour low functional defect. Privacy, non-subsumability, and resistance to reconstruction may favour positive binding. The profile records the relation; it does not rank architectures morally.

The functional is also not a measure of difficulty. A unique reconstruction may be computationally infeasible while its information-theoretic binding is zero. Computational severance is a separate axis.

§6. Coupled assembly

6.1 Random completion variables

Consider a coupled presentation 𝔖1 ⋈ν𝔖2. Let Ω⋈ be a declared finite comparison class of completed coupled presentations, already quotiented by the audit’s isomorphism criterion and equipped with its sealed full-support prior, and let Z_⋈ be the resulting Ω⋈-valued random completion. Under that audit, let Z_1, Z_2, and K be measurable coordinates of Z⋈ on supp(π) for the first factor completion, the second factor completion, and the coupling lift respectively, and let Y = q_⋈ (Z_⋈) be the visible severed presentation for a declared channel q_⋈ :Ω⋈ → 𝒴⋈.

Definition 6.1 (Bi-deterministic coordinatisation). Write T = (Z_1, Z_2, K). The tuple T is a bi-deterministic coordinatisation of Z_⋈ on the sealed support if there are mutually inverse maps

η:supp(T) ⟶ supp(Z_⋈), ρ:supp(Z_⋈) ⟶ supp(T)

such that Z_⋈ = η(T) and T = ρ(Z_⋈) almost surely. In the finite setting, mutual invertibility on supports is equivalent to

H(Z_⋈ ∣Z_1, Z_2, K) = 0, H(Z_1, Z_2, K∣Z_⋈) = 0.

Neither conditional independence nor coordinate minimality is assumed. In particular, Z_1 and Z_2 may be dependent; that dependence is recorded by I_c below.

Define the contextual terms

𝔅_1 = H(Z_1∣Y), 𝔅_2 = H(Z_2∣Y),

𝔅_ν = H(K∣Z_1, Z_2, Y), I_c = I(Z_1;Z_2∣Y).

These are contextual quantities inside the assembled observation. They equal standalone factor bindings only when the audit declaration makes the corresponding marginal channels and priors agree.

6.2 Exact decomposition

Theorem 6.2 (Binding decomposition). Under Definition 6.1,

𝔅⋈ := H(Z⋈ ∣Y) = H(Z_1, Z_2, K∣Y) = 𝔅1+𝔅_2+𝔅ν-I_c.

Proof. Write T = (Z_1, Z_2, K). Definition 6.1 gives H(Z_⋈ ∣T) = H(T∣Z_⋈) = 0. Since conditioning preserves these deterministic dependences,

H(Z_⋈, T∣Y) = H(T∣Y)+H(Z_⋈ ∣T, Y) = H(T∣Y)

and

H(Z_⋈, T∣Y) = H(Z_⋈ ∣Y)+H(T∣Z_⋈, Y) = H(Z_⋈ ∣Y).

Thus H(T∣Y) = H(Z_⋈ ∣Y), which proves the first equality. The chain rule gives

H(Z_1, Z_2, K∣Y) = H(Z_1, Z_2∣Y)+H(K∣Z_1, Z_2, Y).

The conditional mutual-information identity gives

H(Z_1, Z_2∣Y) = H(Z_1∣Y)+H(Z_2∣Y)-I(Z_1;Z_2∣Y).

Substitution yields the result. □

Here the compatibility correction is not a log-count of excluded pairs. It is the standard conditional mutual information I(Z_1;Z_2∣Y), and it is therefore non-negative and directly calculable from the sealed joint distribution.

6.3 Three regimes

Corollary 6.3 (Placement regime). If the coupling lift is determined by the completed factors and visible presentation,

H(K∣Z_1, Z_2, Y) = 0,

and the factor completions are conditionally independent given Y,

I(Z_1;Z_2∣Y) = 0,

then binding is additive:

𝔅_⋈ = 𝔅_1+𝔅_2.

Corollary 6.4 (Positive binding excess). With 𝔅⋈ = H(Z⋈ ∣Y) as in Theorem 6.2,

𝔅_⋈ >𝔅_1+𝔅_2 ⟺ H(K∣Z_1, Z_2, Y)>I(Z_1;Z_2∣Y).

The excess 𝔅_⋈ -(𝔅_1+𝔅_2) is exactly

𝔅_ν-I_c.

If equality holds, the two terms cancel. If 𝔅_ν<I_c, compatibility reduces the completion ambiguity below the additive baseline.

Remark 6.5 (What “division” can honestly mean here). The word division does not imply positive binding excess by definition. A division-like assembly is a substantive claim requiring both δc = 1 at the joint under a predeclared P_c and the strict inequality 𝔅⋈ >𝔅_1+𝔅_2 of Corollary 6.4. The theorem supplies the test; it does not grant the label in advance.

6.4 Finite calculations

The following examples evaluate every term. In each one, take Z_⋈ = T = (Z_1, Z_2, K), so Definition 6.1 holds with identity maps on the sealed support. Let Y be constant on the sealed support, so H(·∣Y) = H(·) for every term below, and measure all entropies in bits (base two).

Example 6.6 (Placement). Let Z_1 and Z_2 be independent fair bits and let K = 0 deterministically. Then

𝔅1 = 1, 𝔅_2 = 1, I_c = 0, 𝔅ν = 0,

so 𝔅_⋈ = 2 bits. No binding excess is created at the joint.

Example 6.7 (Lift-rich coupling). Let Z_1, Z_2 remain independent fair bits. If Z_1 = Z_2, let K be a fair bit; if Z_1 ≠ Z_2, set K = 0. Then

I_c = 0,

and

𝔅_ν = H(K∣Z_1, Z_2) = Pr[Z_1 = Z_2]·1 +Pr[Z_1 ≠ Z_2]·0 = 1/2.

Therefore 𝔅_⋈ = 2.5 bits: the joint creates a half-bit of binding excess.

Example 6.8 (Compatibility cancellation). Let Z_2 = Z_1 be the same fair bit and let K be an independent fair bit. Then

𝔅1 = 𝔅_2 = 1, I_c = 1, 𝔅ν = 1,

so 𝔅_⋈ = 2 bits. The one-bit coupling ambiguity is exactly cancelled by the one-bit compatibility correction.

The correction term is therefore not merely defined: it is zero in the first two declarations and positive in the third, where it exactly cancels the lift-ambiguity term.

§7. Uniform exact non-absorption under confined access

7.1 Access boundary

In the finite setting of §§2 - 4, let Z be a discrete (finite-support) random floor in a sealed comparison class, and let Y = q_c(Z) be the declared observation available after component c is hidden. Let W be all data available to an upper layer G.

Definition 7.1 (Confined access and uniform exact absorption). Access is confined toY when

Z ⟶ Y ⟶ W

is a Markov chain. Thus G may transform Y, combine it with independent randomness, and build arbitrary internal representations, but it receives no additional information about Z.

G achieves uniform exact absorption of the declared floor class when there exists a decoder D such that

D(W) = Z

almost surely under the sealed prior. Because that prior has full support on the finite comparison class, the almost-sure equality requires correct decoding for every class element (up to null internal randomness). Exact decoding implies H(Z∣W) = 0. A constant decoder for one preselected target does not count.

7.2 The theorem

Theorem 7.2 (Positive binding forbids uniform exact absorption under confined access). Fix a sealed finite audit, let Y = q_c(Z) be its declared observation, and let W contain all data available to an upper layer. If Z → Y → W is a Markov chain and H(Z∣Y)>0, then no decoder D can satisfy D(W) = Z almost surely under the sealed prior. Equivalently, under confined access, uniform exact absorption of the declared floor class forces H(Z∣Y) = 0.

Proof. By the data-processing inequality for the Markov chain Z → Y → W,

I(Z;W) ≤ I(Z;Y).

Equivalently,

H(Z∣W) ≥ H(Z∣Y)>0.

Exact decoding D(W) = Z would imply H(Z∣W) = 0, a contradiction. □

Corollary 7.3 (NR-ε form). Under the hypotheses of Theorem 4.1, if H(Z)>ε, Y is the declared NR-ε channel, W contains all upper-layer data, and Z → Y → W is a Markov chain, then uniform exact absorption is impossible. Under the coordinate hypotheses of Corollary 4.2, the same conclusion holds with Y = Y_{Adm} whenever Z → Y_{Adm} → W, because Corollary 4.2 establishes positive residual entropy for the full completion Z, not merely for its protected coordinate.

Remark 7.4 (What the theorem does not prove). It does not prove that an upper layer cannot reference, approximate, emulate, rename, or independently rediscover parts of the floor. It does not prove that composition is automatically safe. It proves one precise impossibility: uniform exact reconstruction of the protected completion from the declared observation when all upper-layer access is confined through that observation and the residual entropy is positive.

Remark 7.5 (The side-channel condition is load-bearing). If W receives privileged metadata, design documents, white-box instrumentation, or any other signal not generated from Y, the Markov condition may fail. The theorem then makes no claim. Channel completeness is part of the model, not a footnote after the proof.

Remark 7.6 (Relation to the cross-layer companion). The companion’s factorisation criterion asks when a functor descends through a forgetful functor by constancy on object and hom-set fibres. Theorem 7.2 is the information-theoretic reconstruction counterpart: a post-processing of the forgotten observation cannot invert a class with positive residual entropy.

§8. The severance audit

8.1 Protocol

Protocol 8.1. For each component c:

1. Declare and seal 𝒜: component, comparison class and its isomorphism criterion, prior, observation channel, excision map, functional criterion, and any weight.
2. Run excision. Report e_c(𝔖) and the value of P_c before and after. This yields δ_c.
3. Run reconstruction. Supply exactly one of:
4. (R) a uniform reconstruction map R_c on the entire declared class;
5. (W) a non-injectivity witness pair 𝔖’¬ ≅ 𝔖’’ in Ω with q_c(𝔖’) = q_c(𝔖’’); for a target-specific claim, one member must be the target;
6. (U) unresolved: neither certificate has been supplied.
7. Calculate binding. In a finite class, enumerate the posterior fibre and compute H(Z∣Y = y). For an NR-ε result, report the entropy margin of the protected variable and state whether that variable is the full completion Z or a coordinate f(Z). In the coordinate case, report the induced lower bound H(Z∣Y) ≥ H(f(Z)∣Y) and distinguish average from pointwise binding.
8. Report the profile. Publish (δ_c, 𝔅_c) together with the declaration. Never publish the scalar alone.

Remark 8.2 (Hard is not hidden). Computational difficulty is not a witness of positive binding. If a uniform reconstruction exists but is expensive, the information-theoretic binding is zero and the computational cost must be reported separately.

8.2 NC2.5 audit status

The earlier four-row table mixed tuple components with logical properties and treated a declaration as though it were automatically an observation channel. The corrected status is:

Corrected NC2.5 audit status.

This table no longer manufactures a non-severability witness from the loss of a semantic role. Functional death and reconstructive ambiguity are reported separately.

8.3 Minimum publishable certificate

A positive claim of strong binding at c requires both:

1. an excision certificate with δ_c = 1 under a predeclared P_c;
2. either a fibre witness at the target or a pointwise entropy bound showing 𝔅_c(q_c(𝔖))>0.

An average NR-ε bound is valuable but does not replace the second item for a target-specific claim. Appendix A executes both items for a finite two-object capacity audit, supplies separate off-diagonal audits showing that the coordinates remain independent, and verifies all three profiles in code.

§9. Relation to the maintenance ledger

The cross-layer companion’s X+M decomposition (Remark 2.8 there) splits burden into

Φ_S = Φ_S^{(X)}+Φ_S^{(M)},

where Φ^{(M)} is the dynamic maintenance-of-regime cost of holding a nested coupling. The severance profile is static and declaration-relative. The two may be related, but neither coordinate determines the maintenance ledger by definition.

Open Problem 9.1 (Binding - maintenance bridge). Find a class in which a non-trivial bound of the form

lim inf_{t → ∞}1/tΦ_S^{(M)}(t) ≥ f(δ_c, 𝔅_c)

holds for a declared monotone f, with a falsifier that does not simply restate the definition of Φ^{(M)}.

Nothing in §§3 - 8 depends on this bridge.

§10. Failure and falsification surface

The mathematical identities and the adequacy of an audit declaration have different failure modes.

F1  -  observation-class mismatch. Exhibit a signal used by the upper layer that is not a post-processing of the declared Y. This does not refute Theorem 7.2; it shows that the Markov access boundary was false for that deployment.

F2  -  target-fibre collapse. For a target claimed to have positive pointwise binding, exhibit a uniform reconstruction on the declared class or show that its observation fibre is a singleton. This refutes the target-specific binding claim.

F3  -  entropy-margin failure. Show that H(Z_{Adm}) ≤ ε, that Z_{Adm} is not the declared coordinate f_{Adm}(Z) of the full completion, or that the claimed NR-ε bound does not apply to the chosen protected coordinate and channel. Then Corollary 4.2 no longer certifies positive full-completion binding by this route.

F4  -  declaration gaming. Show that small admissible changes to Ω, π, q_c, e_c, or P_c reverse the reported profile while preserving the intended modelling question. This does not falsify entropy theory; it destroys the practical invariance of the audit.

F5  -  excision criterion gaming. Show that P_c was chosen after the result, or that it encodes the presence of c directly rather than a function the architecture independently claims to preserve. Then δ_c is vacuous.

F6  -  assembly typing failure. Show that either zero-entropy condition in Definition 6.1 fails, so Z_⋈ and T = (Z_1, Z_2, K) are not bi-deterministic coordinates of one another, or that the prior, channel q_⋈, and marginal terms in §6 do not come from a single sealed audit. Then the architectural interpretation of Theorem 6.2 has been applied outside its hypotheses.

F7  -  empirical bridge failure. In a class where a lower bound from (δ, 𝔅) to Φ^{(M)} is proposed, exhibit a sequence with positive severance profile and arbitrarily small maintenance rate. This would refute the proposed bridge of §9 for that class.

§11. What this note does not claim

1. No canonical decomposition. The declaration is part of the result.
2. No scalar definition of indivisibility. Functional defect and reconstructive binding are independent axes.
3. No automatic inference from NR-ε to target-specific strong binding. The protected variable, its coordinate relation to the full completion, the channel, prior, and entropy margin must match, and an average bound does not identify every target fibre.
4. No theorem outside the declared access boundary. Theorem 7.2 concerns uniform exact decoding only when Y is the sealed declared channel, W contains all upper-layer data, and Z → Y → W. It makes no claim about unconfined side channels, approximate recovery, emulation, independent rediscovery, or composition.
5. No automatic inference from “division” to binding excess. Corollary 6.4 gives the inequality that must be verified.
6. No formal structural gravity. The companion essay remains an essay.
7. No merit ranking. Low defect and low binding may be desirable engineering properties.
8. No general solution to the origin-of-threshold problem. Appendix A supplies one declared finite hidden-value witness, but it does not derive C from substrate structure or settle arbitrary comparison classes.

§12. Open problems

12.1. Continuous channels. Extend the binding functional to standard Borel comparison classes using regular conditional distributions; state the conditions under which pointwise binding is defined almost everywhere.

12.2. Robust declarations. Characterise when the severance profile is stable under a declared family of admissible decompositions, priors, and observation channels.

12.3. Target-specific NC2.5 witnesses. Extend Appendix A’s explicit capacity witness to non-toy, domain-grounded comparison classes, and construct analogous target-fibre pairs for Adm and Φ rather than inferring them from role descriptions.

12.4. Computational severance. Add the complexity of the best uniform reconstruction as a third coordinate without conflating computational hardness with conditional entropy.

12.5. Binding - maintenance bridge. Resolve Open Problem 9.1 in at least one non-trivial class. The point-mass-in-potential class of the cross-layer companion is a natural candidate because its effect functor and spectral data are explicit.

12.6. Composition without uniform exact absorption. Give sufficient conditions under which positive binding coexists with stable, non-destructive composition by an upper layer. Theorem 7.2 blocks uniform exact reconstruction only under its confined-access hypothesis; it does not by itself guarantee a healthy interface.

Appendix A. Complete sealed audits: capacity witness and axis separation

This appendix executes Protocol 8.1 from declaration through calculation. Its primary audit supplies a target-specific witness for the capacity component in one finite comparison class. That construction is deliberately minimal: it proves the advertised certificate without claiming that capacity is derived from the remaining substrate data. Section A.6 then supplies two separately sealed off-diagonal audits whose only purpose is to show that the coordinates of Definition 3.6 remain logically distinct.

A.1 Substrates and sealed declaration

Let

X = {x_0, x_1},

and let D contain the single non-identity transition x_0 → x_1. Its finite trajectories are the two zero-length trajectories (x_0) and (x_1) and the one-edge trajectory

γ_* = (x_0, x_1).

Declare every trajectory admissible and define the monotone burden by

Φ((x_0)) = Φ((x_1)) = 0, Φ(γ_*) = 3/2.

There are no longer trajectories, so monotonicity under trajectory extension is immediate. Define two completed substrates that differ only in capacity:

𝔖- = (X, D, Adm, Φ, C-), C_- = 1,

𝔖+ = (X, D, Adm, Φ, C+), C_+ = 2.

The target is 𝔖_+. On the distinguished trajectory,

τ-(γ) = 1 - 3/2 = -1/2, τ+(γ) = 2 - 3/2 = 1/2.

Two candidates are declared isomorphic iff there exists a bijection f:X → X preserving D, Adm, Φ, and C. The directed skeleton x_0 → x_1 has only the identity automorphism, and the capacities differ, hence

𝔖-¬ ≅ 𝔖+.

Seal

Ω = {𝔖-, 𝔖+}, π(𝔖-) = π(𝔖+) = 1/2, w_C = 1.

Both capacity values are nominal sealed declarations. No saturation evidence, empirical ratio claim, or derivation of either value from substrate structure is asserted.

A.2 Observation, excision, and functional criterion

The observation channel hides C and every quantity copied or derived from it:

q_C(X, D, Adm, Φ, C) = (X, D, Adm, Φ).

In particular, neither τ(γ_*) nor its sign remains visible. The excision map e_C returns the same capacity-deleted presentation.

For any completed or excised presentation p, let Comp(p) ⊆ Ω be the set of sealed completions agreeing with every field retained by p. Define the declared budget verdict

v(𝔖’) = 1{τ{𝔖’}(γ*)>0},

and the functional criterion

P_C(p) = 1 ⟺ Comp(p) ≠ ∅ and ∃b ∈ {0, 1} ∀𝔖’ ∈ Comp(p), b = v(𝔖’).

Thus P_C asks whether the presentation determines one correct budget verdict uniformly over every compatible completion. It does not test merely whether a field named C is present. If all compatible hidden capacities produced the same verdict, an excised presentation would still pass.

The one-component audit

𝒜_C = ({C}, Ω, π, {(q_C, e_C, P_C)}, {w_C})

is now sealed before scoring.

A.3 Functional-severance certificate

The completed target fixes its own capacity, so

Comp(𝔖+) = {𝔖+}, P_C(𝔖_+) = 1.

After excision,

Comp(e_C(𝔖+)) = {𝔖-, 𝔖_+}.

But

v(𝔖-) = 0, v(𝔖+) = 1.

No single b is correct for both completions. Therefore

P_C(e_C(𝔖+)) = 0, δ_C^{𝒜_C}(𝔖+) = 1.

A.4 Target-fibre and binding certificates

The two completed substrates have the same observation:

q_C(𝔖-) = q_C(𝔖+) = :y_*.

Together with 𝔖-¬ ≅ 𝔖+, this is certificate (W) of Protocol 8.1, and one member is the target. Hence

Sev_C^{𝒜C}(y*) = {𝔖-, 𝔖+}.

The sealed posterior is uniform, so the target-specific binding is

𝔅C^π(y) = H_π(Z∣Y_C = y_) = -2(1/2log_21/2) = 1 bit.

Because Y_C is constant on Ω, average and pointwise binding coincide: 𝔅̅C^π = 𝔅_C^π(y*) = 1 bit. The completed audit therefore reports

𝒫C^{𝒜_C}(𝔖+) = (1, 1 bit).

A.5 Shared-witness dependence, exact scope, and falsifiers

The functional and binding certificates above are both valid, but they are not independent within this particular audit. Here e_C(𝔖+) is exactly the capacity-deleted observation q_C(𝔖+), and Comp(e_C(𝔖_+)) is the target observation fibre. Because that set is non-empty, P_C fails precisely when it contains completions with different budget verdicts. Consequently, under the full-support prior,

δC^{𝒜_C}(𝔖+) = 1 ⟹ |Sev_C^{𝒜C}(y)| ≥ 2 ⟹ 𝔅C^π(y)>0.

This one-way implication is created by the sealed choice of (q_C, e_C, P_C); it is not an identity between the coordinates of Definition 3.6. The converse fails when multiple hidden capacities share one budget verdict, and §A.6 supplies separately sealed witnesses for both off-diagonal profiles.

The primary certificate is declaration-relative and closes only the finite toy-C branch of Open Problem 12.3.

1. It does not derive C_- or C_+ from (X, D, Adm, Φ) and therefore does not solve the general origin-of-threshold problem.
2. It does not imply that every capacity audit has positive binding or positive functional defect.
3. If q_C exposes C, τ(γ_*), the budget verdict, or equivalent metadata, the displayed witness pair no longer lies in one observation fibre.
4. If the isomorphism criterion identifies the two capacities, or if the comparison class removes one completion, the witness disappears.
5. If every compatible completion gives the same budget verdict, then this declared P_C assigns the excised presentation value one even though C remains hidden; §A.6.2 makes this case explicit.

These are falsifiers of this certificate’s hypotheses, not exceptions added after scoring.

A.6 Off-diagonal sealed audits

These two audits are logical independence witnesses, not additional empirical claims about NC2.5 deployments. Each uses a finite full-support uniform prior and is sealed before scoring.

A.6.1 Functionally load-bearing but reconstructible: (1, 0)

Let

X_E = {s, t}, D_E = {E:s → t},

and let the finite trajectory set induced by this directed skeleton be

Γ_E = {(s), (t), (s, t)}.

Declare every trajectory in Γ_E admissible, set Φ_E(γ) = 0 for every γ ∈ Γ_E, and take C_E = 1. Thus

𝔘_E = (X_E, D_E, Adm_E, Φ_E, C_E)

is a substrate of the form required by Definition 2.1: Φ_E is monotone under extension and the viability budget is positive on every admissible trajectory. With the uniform prior π_E, define

ΩE = {𝔗{red}, 𝔗_{blue}}, 𝔗_i = (𝔘_E;i, s, t).

The semicolon separates the five-field substrate from presentation metadata. Each completed presentation has a visible colour i ∈ {red, blue}, marked input s, and marked output t. An isomorphism in the declared class must preserve (X_E, D_E, Adm_E, ΦE, C_E) as well as the colour and the marked states. The target is 𝔗{blue}, and the audited component is the transport edge E.

The observation

q_E(𝔗_i) = (i, X_E, C_E, s, t)

hides D_E, the trajectory-dependent fields Adm_E and Φ_E, and every reachability bit derived from them. The colour is independent visible data rather than a copy of E: it varies while the complete five-field substrate is identical in both presentations. Since the colours differ, q_E is injective on Ω_E, so the target fibre is a singleton and

𝔅E^{π_E}(q_E(𝔗{blue})) = 0.

For excision, let

D_E⁰ = ∅, Γ_E⁰ = {(s), (t)},

declare both singleton trajectories admissible, and set Φ_E⁰ = 0 on Γ_E⁰. The excised presentation is the well-typed empty-edge completion

e_E(𝔗_i) = ((X_E, D_E⁰, Adm_E⁰, Φ_E⁰, C_E);i, s, t).

Declare P_E(p) = 1 iff the transition structure carried by p contains a directed path from its marked input to its marked output, and seal

𝒜_E = ({E}, Ω_E, π_E, {(q_E, e_E, P_E)}, {1})

before scoring. Both completed presentations pass, while the excised target has no such path. Therefore

δE^{𝒜_E}(𝔗{blue}) = 1, 𝒫E^{𝒜_E}(𝔗{blue}) = (1, 0).

The component is functionally load-bearing but uniformly reconstructible from the declared class and observation.

A.6.2 Functionally optional but target-ambiguous: (0, 1 bit)

Retain the visible signature and burden of §A.1, but seal a different comparison class

Ω_C^{same} = {𝔖_2, 𝔖_3}, C_2 = 2, C_3 = 3,

with uniform prior π_{same} and target 𝔖_3. Let q_C^{same} and e_C^{same} be the restrictions of the capacity-hiding channel and excision map from §A.2 to Ω_C^{same}. For any completed or excised presentation p, define

Comp_{same}(p) := {𝔖’ ∈ Ω_C^{same}: 𝔖’ agrees with every field retained by p}.

Let P_C^{same} be the budget-verdict criterion of §A.2 with Comp replaced by Comp_{same}:

P_C^{same}(p) = 1 ⟺ Comp_{same}(p) ≠ ∅ and ∃b ∈ {0, 1} ∀𝔖’ ∈ Comp_{same}(p), b = v(𝔖’).

Seal

𝒜C^{same} = ({C}, Ω_C^{same}, π{same}, {(q_C^{same}, e_C^{same}, P_C^{same})}, {1}).

before scoring. Both completions satisfy τ(γ_*)>0, so every compatible completion has verdict one and

P_C^{same}(e_C^{same}(𝔖_3)) = 1, δ_C^{𝒜_C^{same}}(𝔖_3) = 0.

The capacity-hiding observation is constant on the two non-isomorphic completions. Its target fibre therefore has two uniform members and one bit of binding:

𝔅C^{π{same}} (q_C^{same}(𝔖_3)) = 1 bit, 𝒫_C^{𝒜_C^{same}}(𝔖_3) = (0, 1 bit).

Taken together with the primary (1, 1 bit) audit, these off-diagonal witnesses show that neither profile coordinate determines the other. They do not show declaration-invariance; F4 remains applicable.

A.7 Executable reference verifier

The following standard-library Python program implements the primary capacity certificate and the two off-diagonal audits, then checks all three profiles. In the transport audit, TransportSubstrate records the full five-field substrate together with the colour and marked-state metadata; q_E hides the transition-dependent fields, and e_E constructs the well-typed empty-edge completion. For the primary capacity audit, the prose criterion requires a bijection preserving (X, D, Adm, Φ, C). In that two-object class, both objects share the same fixed labelled visible signature, the directed skeleton has only the identity automorphism, and only C varies. The code’s isomorphic function therefore implements the declared criterion by comparing q_C and C while ignoring the presentation name.

from fractions import Fraction
from math import log2
from typing import NamedTuple

class Substrate(NamedTuple):
    name: str
    capacity: Fraction

class TransportSubstrate(NamedTuple):
    colour: str
    states: tuple
    edges: tuple
    admissible: tuple
    burden: tuple
    capacity: Fraction
    source: str
    target: str

X = ("x0", "x1")
D = (("x0", "x1"),)
ADM = (("x0",), ("x1",), ("x0", "x1"))
PHI = (
    (("x0",), Fraction(0)),
    (("x1",), Fraction(0)),
    (("x0", "x1"), Fraction(3, 2)),
)
VISIBLE_SIGNATURE = (X, D, ADM, PHI)
PHI_GAMMA = Fraction(3, 2)

S_MINUS = Substrate("S_minus", Fraction(1))
S_PLUS = Substrate("S_plus", Fraction(2))
OMEGA = (S_MINUS, S_PLUS)
PRIOR = {S_MINUS: Fraction(1, 2), S_PLUS: Fraction(1, 2)}
TARGET = S_PLUS

S_THREE = Substrate("S_three", Fraction(3))
OMEGA_SAME_VERDICT = (S_PLUS, S_THREE)
PRIOR_SAME_VERDICT = {
    S_PLUS: Fraction(1, 2),
    S_THREE: Fraction(1, 2),
}
TARGET_SAME_VERDICT = S_THREE

X_EDGE = ("s", "t")
TRANSPORT_EDGE = (("s", "t"),)
EDGE_TRAJECTORIES = (("s",), ("t",), ("s", "t"))
ADM_EDGE = EDGE_TRAJECTORIES
PHI_EDGE = (
    (("s",), Fraction(0)),
    (("t",), Fraction(0)),
    (("s", "t"), Fraction(0)),
)
CAPACITY_EDGE = Fraction(1)
ADM_EDGE_EXCISED = (("s",), ("t",))
PHI_EDGE_EXCISED = (
    (("s",), Fraction(0)),
    (("t",), Fraction(0)),
)

T_RED = TransportSubstrate(
    "red",
    X_EDGE,
    TRANSPORT_EDGE,
    ADM_EDGE,
    PHI_EDGE,
    CAPACITY_EDGE,
    "s",
    "t",
)
T_BLUE = TransportSubstrate(
    "blue",
    X_EDGE,
    TRANSPORT_EDGE,
    ADM_EDGE,
    PHI_EDGE,
    CAPACITY_EDGE,
    "s",
    "t",
)
OMEGA_EDGE = (T_RED, T_BLUE)
PRIOR_EDGE = {T_RED: Fraction(1, 2), T_BLUE: Fraction(1, 2)}
TARGET_EDGE = T_BLUE

def q_C(substrate):
    return VISIBLE_SIGNATURE

def isomorphic(left, right):
    return q_C(left) == q_C(right) and left.capacity == right.capacity

def budget_positive(substrate):
    return substrate.capacity - PHI_GAMMA > 0

def completed_presentation(substrate):
    return (q_C(substrate), substrate.capacity)

def e_C(substrate):
    return (q_C(substrate), None)

def compatible(presentation, substrate):
    visible, capacity = presentation
    return q_C(substrate) == visible and (
        capacity is None or substrate.capacity == capacity
    )

def P_C(presentation, omega=OMEGA):
    completions = [s for s in omega if compatible(presentation, s)]
    answers = {budget_positive(s) for s in completions}
    return bool(completions) and len(answers) == 1

def q_C_same(substrate):
    return q_C(substrate)

def e_C_same(substrate):
    return e_C(substrate)

def P_C_same(presentation):
    return P_C(presentation, OMEGA_SAME_VERDICT)

def fibre(observation, omega=OMEGA):
    return [s for s in omega if q_C(s) == observation]

def entropy_bits(probabilities):
    return sum(-p * log2(p) for p in probabilities if p > 0)

def q_E(substrate):
    return (
        substrate.colour,
        substrate.states,
        substrate.capacity,
        substrate.source,
        substrate.target,
    )

def completed_E(substrate):
    return substrate

def e_E(substrate):
    return substrate._replace(
        edges=(),
        admissible=ADM_EDGE_EXCISED,
        burden=PHI_EDGE_EXCISED,
    )

def reachable(edges, source, target):
    frontier = [source]
    seen = {source}
    while frontier:
        current = frontier.pop()
        if current == target:
            return True
        for left, right in edges:
            if left == current and right not in seen:
                seen.add(right)
                frontier.append(right)
    return False

def P_E(presentation):
    return reachable(
        presentation.edges,
        presentation.source,
        presentation.target,
    )

target_fibre = fibre(q_C(TARGET))
normalizer = sum(PRIOR[s] for s in target_fibre)
posterior = [float(PRIOR[s] / normalizer) for s in target_fibre]
binding = entropy_bits(posterior)
delta_C = 1 - int(P_C(e_C(TARGET)))

same_fibre = [
    s
    for s in OMEGA_SAME_VERDICT
    if q_C_same(s) == q_C_same(TARGET_SAME_VERDICT)
]
same_normalizer = sum(PRIOR_SAME_VERDICT[s] for s in same_fibre)
same_posterior = [
    float(PRIOR_SAME_VERDICT[s] / same_normalizer) for s in same_fibre
]
same_binding = entropy_bits(same_posterior)
delta_same = 1 - int(
    P_C_same(e_C_same(TARGET_SAME_VERDICT))
)

edge_fibre = [
    s for s in OMEGA_EDGE if q_E(s) == q_E(TARGET_EDGE)
]
edge_normalizer = sum(PRIOR_EDGE[s] for s in edge_fibre)
edge_posterior = [
    float(PRIOR_EDGE[s] / edge_normalizer) for s in edge_fibre
]
edge_binding = entropy_bits(edge_posterior)
excised_edge = e_E(TARGET_EDGE)
delta_E = 1 - int(P_E(excised_edge))

assert all(PRIOR[s] > 0 for s in OMEGA)
assert sum(PRIOR.values()) == 1
assert dict(PHI)[("x0",)] <= dict(PHI)[("x0", "x1")]
assert set(target_fibre) == set(OMEGA)
assert not budget_positive(S_MINUS) and budget_positive(S_PLUS)
assert P_C(completed_presentation(TARGET))
assert not P_C(e_C(TARGET))
assert q_C(S_MINUS) == q_C(S_PLUS)
assert not isomorphic(S_MINUS, S_PLUS)
assert delta_C == 1
assert binding == 1.0

assert all(PRIOR_SAME_VERDICT[s] > 0 for s in OMEGA_SAME_VERDICT)
assert sum(PRIOR_SAME_VERDICT.values()) == 1
assert set(same_fibre) == set(OMEGA_SAME_VERDICT)
assert q_C_same(S_PLUS) == q_C_same(S_THREE)
assert all(budget_positive(s) for s in OMEGA_SAME_VERDICT)
assert P_C_same(completed_presentation(TARGET_SAME_VERDICT))
assert P_C_same(e_C_same(TARGET_SAME_VERDICT))
assert not isomorphic(S_PLUS, S_THREE)
assert delta_same == 0
assert same_binding == 1.0

assert all(PRIOR_EDGE[s] > 0 for s in OMEGA_EDGE)
assert sum(PRIOR_EDGE.values()) == 1
assert set(ADM_EDGE) == set(EDGE_TRAJECTORIES)
assert set(dict(PHI_EDGE)) == set(ADM_EDGE)
assert all(value == 0 for value in dict(PHI_EDGE).values())
assert dict(PHI_EDGE)[("s",)] <= dict(PHI_EDGE)[("s", "t")]
assert CAPACITY_EDGE > 0
assert T_RED._replace(colour="blue") == T_BLUE
assert q_E(T_RED)[1:] == q_E(T_BLUE)[1:]
assert len({q_E(s) for s in OMEGA_EDGE}) == len(OMEGA_EDGE)
assert edge_fibre == [TARGET_EDGE]
assert completed_E(TARGET_EDGE) == TARGET_EDGE
assert P_E(completed_E(TARGET_EDGE))
assert excised_edge.states == X_EDGE
assert excised_edge.edges == ()
assert excised_edge.admissible == ADM_EDGE_EXCISED
assert excised_edge.burden == PHI_EDGE_EXCISED
assert set(dict(excised_edge.burden)) == set(excised_edge.admissible)
assert excised_edge.capacity == CAPACITY_EDGE
assert excised_edge.colour == TARGET_EDGE.colour
assert excised_edge.source == "s" and excised_edge.target == "t"
assert not P_E(excised_edge)
assert delta_E == 1
assert edge_binding == 0.0

print("target:", TARGET.name)
print("witness:", S_MINUS.name)
print("budget verdicts:", int(budget_positive(S_PLUS)), int(budget_positive(S_MINUS)))
print("fibre size:", len(target_fibre))
print("delta_C:", delta_C)
print("binding_bits:", binding)
print("profile:", (delta_C, binding))
print("edge_profile:", (delta_E, edge_binding))
print("same_verdict_capacity_profile:", (delta_same, same_binding))

Expected output:

target: S_plus
witness: S_minus
budget verdicts: 1 0
fibre size: 2
delta_C: 1
binding_bits: 1.0
profile: (1, 1.0)
edge_profile: (1, 0.0)
same_verdict_capacity_profile: (0, 1.0)

References

Barziankou, M. (2026). Navigational Cybernetics 2.5  -  Axiomatic Core, Version 2.1. The Urgrund Laboratory, PETRONUS. DOI: 10.17605/OSF.IO/NHTC5.

Barziankou, M. (2026). ONTOΣ XIV  -  Nested Substrates and the Frame-Relative Derivative, with formal companion Cross-Layer Forgetful Separation: A Categorical Foundation. Bundle DOI: 10.17605/OSF.IO/KAGMH.

Barziankou, M. (2026). NC2.5 ↔ the Adm_t Class: Point-Indexed Admissibility Gates and the Missing-Trajectory-Grounding Hypothesis. NC2.5 Empirical Probes, Part II. DOI: 10.17605/OSF.IO/7SQMY.

Barziankou, M. (2026). Nothing Is Solid: Atomicity, Emptiness, and the Source of Structural Gravity. Essay Through a Life, Part VIII. Publication-pair DOI shared with the present formal note: 10.17605/OSF.IO/5VJMR. Companion essay; non-formal.

Cover, T. M., and Thomas, J. A. (2006). Elements of Information Theory, 2nd ed. Wiley.

Mac Lane, S. (1998). Categories for the Working Mathematician, 2nd ed. Springer.

MxBv, Poznań, Poland.

The Urgrund Laboratheory.

Maksim Barziankou (MxBv)  -  PETRONUS  -  [e-mail]

July 2026.

CC BY-NC-ND 4.0.

Publication-pair DOI (shared with Nothing Is Solid: Atomicity, Emptiness, and the Source of Structural Gravity): 10.17605/OSF.IO/5VJMR

Grounded in Navigational Cybernetics 2.5 v2.1, axiomatic core DOI: 10.17605/OSF.IO/NHTC5.

Publication pair: this formal note and Nothing Is Solid: Atomicity, Emptiness, and the Source of Structural Gravity (Essay Through a Life, Part VIII) form a single two-part work under the shared DOI 10.17605/OSF.IO/5VJMR.

© MxBv / PETRONUS · CC BY-NC-ND 4.0

Originally published at https://petronus.eu/blog/severance-defect-and-the-binding-functional/

SHA-256: b0077c08ab4bc9aef254023dd539439482b3cf8f4e1fdaa8797405b7de9bd7cc

---

NC2.5 · Urgrund · MxBv · PETRONUS · [e-mail] · CC BY-NC-ND
Originally published https://medium.com/@MxBv/severance-defect-and-the-binding-functional-c7b4f21dea78