# An Unaccepted Evolution: One Structure Beneath Six Sciences, and Its Consequence for Quantum Computation
## Measurement foundations, observer separation and a research programme

KEDDEH SYSTEMS | RESEARCH AND FORMAL PUBLISHING | ENGINEER · AUTOMATE · SCALE

Document ID: KEDDEH-SOURCE-K-FOUND  
Reference ID: K-FOUND  
Edition: E3 · 18 October 2026  
Source edition: Version 1 · September 2026  
Author: Aboudy Keddeh — Systems Architect; independent researcher in the source  
Publisher: Keddeh Systems  
Rights holder: Aboudy Keddeh  
Licence profile: KEDDEH-RESERVED-1.0  
Classification: Internal research · Owner review  
Status: Editorial research edition of an authored preprint

Research programme, architecture and definitions: Aboudy Keddeh. AI-assisted editing, source organisation and edition apparatus: dot, powered by OpenAI. E3 identifies this library editorial release; it does not assert an intervening E2 of the September source. This edition records no independent scientific approval or publication decision.

© 2026 Aboudy Keddeh. All rights reserved. No licence is granted by this document. Separate written permission governs reuse, redistribution and implementation, subject to applicable law. Third-party material retains its own rights. Copyright, software, patent and trademark permissions are distinct.

## Edition purpose and interpretation

This is a complete editorial edition of the September preprint. It retains the ten-section argument, four axioms, two propositions, nested-frame construction, seven source tables, BRAINK architecture and evidence discussion, staged programme, quantum conjecture, six concluding implications and fifteen source references. The purpose is to make the research argument inspectable while preserving its authorship and development.

The main text presents **the author's September argument**. “Axiom,” “proposition,” “follows” and “theorem” identify roles within that argument. Section 2 explicitly declares choices of primitive and monotone structure; Table 1 declares the magnitude floor and independence criterion as choices; A3 identifies its monotonicity as an authorial choice. Those are edition-scoped premises, not evidence of an omitted proof. The related conditional proofs elsewhere in the corpus are identified below with their conditions.

**Current interpretation.** The author's current shared-origin designation is the literal nonnumeric **.1**. It identifies one referent across observables; it is not the decimal 0.1, arithmetic numeral 1 or a separate origin on every axis. Ordered observer, observed and observable roles retain spatial, temporal and instance context. A missing observation does not establish nonexistence. The historical **U**, **1**, **0**, **v = (d, m)** and **m ≥ 1** below remain September formulations. They are not silently replaced by .1 and do not define a current native coordinate encoding. [E1–E2]

Statements about artefacts and assessments in Section 7 are source reports from September. This editorial work does not redetermine their present runtime state or their commercial value. The byte-exact source and its digest accompany the edition; the full native source was also re-read during recovery on 18 October 2026.

## Abstract

This preprint proposes four representational requirements for measurement: retain direction with magnitude; declare the reference rather than assume an absolute origin; distinguish a system's self-report from an independently obtained account; and preserve direction-bearing information through computation. The author connects these requirements to meta cognition, finite cognition, macro signal propagation and machine-readable evolution, understood as a traversable record of how a system changes.

The argument moves from chosen axioms to propositions about warrant and scale, then develops nested frames for interacting bodies. It compares that structure with work in electromagnetism, quantum mechanics, biology, linguistics, ecological psychology and computer science. The proposed correspondences motivate the research programme; their equivalence and independence are claims to examine under the relevant domain conditions.

BRAINK supplies the application. Its semantic engine, learning lane, mesh, lineage ledger, reconciliation process and connectors are described as an apparatus for varying the references available to a bounded learner. The source distinguishes the intended architecture from reported artefacts and independent checks, and examines assessment failures as part of its evidential problem. A staged programme moves from local definition to a separately instrumented workstation and then physical sensors.

The quantum section proposes a representation that carries direction, provenance and observer context through programming and reporting. It supplies prospective tests rather than a new algorithm, demonstrated speedup or hardware result. The contribution is the articulated programme, its dependency structure and its proposed failure conditions, with the September notation preserved and the later author interpretation explicit.

Keywords: consilience; measurement reference; observer separation; signed error; relational state; quantum representation; machine-readable evolution; BRAINK.

## Reader route and initial definitions

Section 1 gives the conceptual argument. Section 3 states the formal assumptions. Section 4.2 develops the nested frames. Sections 7–8 apply them to BRAINK and quantum computation; Sections 9–10 state the representation requirement and its proposed consequences.

An **observer** is the system or participant through which an account is obtained. The **observed** is its referent. A **reference** fixes how an observation is interpreted or scaled. An **environment** contains the conditions and references relevant to that account. A **sector**, in this model, groups sources sharing enough method, instrumentation, training or provenance that they are not counted as independent. **Lineage** is retained predecessors and derivation, rather than a final scalar alone.

The three cognition scales are **meta**, a model of the system's own state; **finite**, what a bounded system can represent and act on with available resources; and **macro**, aggregate propagation and packetisation. **Machine-readable evolution** requires changes to remain traversable and checkable. Table 1 supplies the formal vocabulary before the axioms.

## Cross-paper derivation route

The source's ledger, typed-state and reference arguments connect to explicit mathematical treatments elsewhere in the corpus:

- *Mathematical Foundations and Proof Obligations*, v1.0, §§4.1–4.4, defines translative loss as Shannon conditional entropy H(X | φ(X)), for a discrete source-state random variable X and deterministic projection φ. KTH-001 and its seven-step proof show positive conditional entropy when that projection is noninjective on positive-probability support. Injectivity on that support gives zero loss. This is a precise conditional result supporting the concern about lost distinctions; it does not assign loss to every reduction. [E3]
- The same work, §§5–6, supplies numeric projection, retained cancellation relations and the KTH-003 proof that existence value ε(s)=1 can coexist with magnitude ν(s)=0 because they are different functions. This supplies a formal route for distinguishing a record from its result. It does not numerically define .1. [E3]
- Sections 7–8 specify observer-frame covariance obligations and an invertible phase-tagged internal coordinate convention with an explicit inverse. They distinguish that convention from external SI observations. [E3]
- *Native Algebra and Artifact Verification — Full Research Consolidation*, v4, §§2–5, develops the typed-state constructor system, canonical additive ledger, conditional-entropy example and phase-tagged round trips. Its sorted multiset ledger has a declared order-insensitive domain; it is not a licence to reorder a different native relational expression whose order is significant. [E4]

These dependencies resolve the relevant formal questions under their stated domains. They do not convert the September monotonicity premise or proposed cross-science correspondences into theorems by association.

## Statement of the source's claims

The register separates the source’s principal claims, their locations and their intended consequences. Each claim can be examined without treating the programme as one indivisible assertion.

| Claim | Where | Why it is relevant |
| --- | --- | --- |
| A quantity is a direction and a size at once, and the pair is irreducible. | A1, §3 | Every operation that keeps a size and drops a direction destroys an event rather than approximating it. The T-handle is the case where the whole event lives in the sign and the magnitude never moves. |
| Measurement needs no fixed zero. Counting starts at one, axes share one origin unit, and zero declares absence rather than acting as an origin. | A2, §3 | A chosen origin manufactures size out of an assumption. Celsius, comoving distance and gauge freedom are the same move already made in three fields. |
| What a system reads of itself and what is true of it are different quantities, and their separation is set by the independence around it. | A3, §3 | The load-bearing claim. It makes self-report worthless, makes environment rather than capacity the limit on learning, and makes the gap an observable rather than a worry. |
| No operation may erase a sign where the sign carries meaning. | A4, §3 | Rules out squared error and unsigned aggregation, which is the default objective across estimation. The cost is real and is stated rather than hidden. |
| Nothing can vouch for itself, so warrant comes only from sources that could not have coordinated. | Prop. 1, §4 | Turns independence from a virtue into a count, and makes repetition inside one field worth nothing however much of it there is. |
| Magnitude means something only against the references that scaled it. Which pole is called plus is the observer's; that there are two is not. | Prop. 2, §4.1 | Explains why a figure quoted out of its frame is uninterpretable rather than approximate, and why a system may relabel or rescale but never unsign. |
| Frames nest. Each body is anchored to a face, the containing box supplies the shared unit, and the outermost box is declared but never quantified. | §4.2 | Ends the regress that relational measurement otherwise runs into, and puts the observer in the diagram instead of leaving it as an unstated term. |
| Bodies in a causal relation attribute each other intrinsically, and their existence is tied into the others. | §4.2 | The relations are already there before any observer arrives. The observer contributes scale, not attribution — which is where the standard three-body framing goes wrong. |
| Independent evidence multiplies rather than adds. | §6 | Why six unrelated arrivals at one structure is the finding of this paper, rather than six separate observations of it. |
| A learner must be bounded, because that is how an organism learns at all. | §7.2 | No boundary means no outside and nothing able to contradict the system. Boundedness is the condition of learning, not a limitation on it. |
| Organisms build packets, and repeated traffic thickens the path that carried it. | §7.2 | Makes learning and routing one mechanism, and explains why depth is laid down in order and cannot be added later. |
| Without early mechanical grounding a structure stays shallow, and the shortfall is undetectable from inside it. | §7.5 | The observer theorem in practice, in organisms and in systems trained only on authored sources. Awareness of it changes nothing; only references do. |
| Learning requires references that were not authored. Consuming coded process is retrieval. | §7.5 | Authored sources share provenance and can fail together. This is why instrumented exposure is a break in kind rather than an increase in degree. |
| Quantum mechanics already obeys all four rules; the layer we program it through obeys none. | §8 | Makes the next step representational and addressable in software now, on hardware that already exists. |
| Reduction destroys derivation, so records must be ledgers rather than balances. | §9.1 | Agreement is evidence only if the routes to it are visible. Reduction makes independent confirmation unverifiable, which is fatal for research and for learning. |
| The rules describe a topology rather than a metric. | §9.1 | Why the structure survives translation between fields sharing no units: magnitudes are local to their box, while sign, boundary and lineage cross intact. |

The four rules and the propositions are argued from stated axioms. The applications in §7 and the conjectures in §8 and §10 can fail without taking the rules with them; the ways they can fail are stated where they are made.

Their relevance is not that they are broadly applicable. It is that each names a failure already being paid for. Self-reported metrics are trusted daily in audit, medicine, engineering and machine learning; absence is stored as zero in most record systems that exist; squared error is the default objective in nearly all estimation; figures are quoted away from the conditions that scaled them as a matter of routine; and reduction is standard practice in exactly the research that later needs its derivations back. None of this is exotic. The rules are relevant because the failures are ordinary, and because in every case they were avoidable at the point of representation and are not recoverable afterwards.

## 1. The argument in plain terms

This section states the whole argument without notation. Each part is given as a plain claim with the everyday case that makes it obvious, followed by where in the paper it is stated exactly and which existing body of work it corresponds to. A reader who stops after this section should have the argument; a reader who continues will find the same five things again, in the form that can be checked. The work they come out of concerns how anything — a person, an instrument, a machine that learns — arrives at a reliable account of its own state, and how that account can be written down so a machine can read and check it rather than merely store it.

You cannot cut a magnet in half and get a north-only piece.

Cut it and you get two magnets, each with both poles. The direction is not a label attached to the magnet, it is part of what the magnet is. The same holds of any measurement where direction carries meaning: a temperature three degrees too high and one three degrees too low are not the same reading with a different note on it. They are different readings.

So the first rule is that a quantity is always a direction and a size together, and nothing may take the size and throw the direction away.

**Stated exactly**

A1, §3

**Already known as**

Maxwell's ∇·B = 0, 1873 — the law that no isolated magnetic pole exists [1].

An empty cell in a spreadsheet is not the number zero.

If a patient's temperature was never taken, the chart should say so. Enter 0 instead and the average drops, and nothing downstream can tell that a measurement is missing rather than freezing. The symbol did two jobs and only one of them was honest.

Zero is genuinely useful for the first job: saying that something is absent, or not relevant, in a particular report. The harm comes from the second, where zero quietly becomes the point everything is measured from — because then a thing gets its apparent size from its distance to an origin nobody measured. A ruler shows the alternative: to measure between two marks you never need the ruler's own zero, only the two marks.

So the second rule is that counting starts at one, that a chart with several axes shares one starting unit rather than giving each axis a private zero, and that “not present” must stay distinguishable from “zero”.

**Stated exactly**

A2, §3

**Already known as**

Zero-point energy in physics and Gödel's incompleteness in mathematics [11]: neither field found a floor it could stand on.

You cannot hear your own voice the way a room hears it.

What reaches you travels through your own skull, so your impression of your voice and your voice are two different things. A recording fixes it, but only because the recorder is outside you. Alone in a soundproof room you could sing for years and never find out.

This is the heart of the paper, and the part that matters is the room rather than the singer. How well anything knows its own state is set by how many independent things around it can disagree with it. Strip those away and the gap widens with no warning — not because the thing is dishonest, but because nothing present is able to contradict it.

So the third rule is that what a system reports about itself and what is true of it are two separate quantities, never interchangeable, and the distance between them is a fact about its surroundings.

**Stated exactly**

A3, §3; result in §4

**Already known as**

Reached separately by physics [4, 5], biology [6], linguistics [7, 8] and ecological psychology [9], none of them citing the others.

A thermostat that knows only how wrong it is cannot fix anything.

Tell it the room is three degrees off and it still has to guess whether to heat or cool. The standard way of scoring error in machine learning does exactly this: it squares the miss, so a model that always overshoots and a model that always undershoots receive identical marks and are corrected identically.

So the fourth rule is that any operation which erases direction — squaring, absolute value, unsigned totals — is not allowed where direction was carrying information. It follows from the first rule: erase the sign and you have separated the magnet's poles.

**Stated exactly**

A4, §3; cost in §9

**Cost**

This rules out squared error, the default objective in nearly all estimation. The paper does not supply a replacement.

Three unrelated witnesses beat one witness repeating themselves.

If nobody can vouch for themselves, agreement between others is all that is left — and it has to be between people who could not have coordinated. Two colleagues from the same office may share the same misunderstanding; a stranger across town does not.

Crucially, such agreement does not stack, it compounds. The odds contributed by each independent witness multiply together rather than adding, which is why a handful of unrelated confirmations feels suddenly decisive while any number of repetitions from one source changes nothing.

So the fifth claim is that warrant comes only from agreement across genuinely independent sources, and that it grows by multiplication.

**Stated exactly**

§4, §5, §6

**Already known as**

Named consilience by Whewell in 1840 [2] and sharpened by Wilson [3]; the multiplication is Bayes' theorem [13].

Two applications follow from the five. The first is to machine learning, where a learner is treated as an organism reading itself through its environment: the rules say that such a learner should keep the direction of its errors, count from its first observation rather than a notional zero, and hold its self-assessment apart from its measured state — and that the limit on how well it can know itself lies in the richness of its surroundings rather than in its size. The second is to the author's own system, which the rules are turned on directly, on the principle that a rule set unable to embarrass its author has not been tested.

That second application yields one further plain point, and it is the one most often got wrong. A wind tunnel is not a failed aeroplane. Judging a research instrument by whether it ships measures the wrong thing entirely, and the paper documents assessments of this work that did precisely that — first inventing evidence of success, then swinging to a verdict of failure — while every piece of code that could simply be run stayed exactly as it was. The right question for an apparatus is what it makes measurable that was not measurable before.

The remainder of the paper restates all of this in the form that can be checked. The everyday cases above are not evidence for the claims and are not offered as such; they are there so that the notation in §3 is read as a tightening of something already understood rather than as a new subject. Each rule is given as an axiom, the central result is derived from two of them, the correspondences of §5 are set out with their sources, and the applications are separated throughout into what is specified, what has been built, and what has been independently checked.

### Edition note: a reported scalar and a complete procedure

The source distinguishes the everyday examples from evidence. A squared residual alone discards the sign of that residual. It does not follow that every procedure using squared error discards the residual or updates opposite errors identically: for the ordinary comparison L(x)=(x−a)², dL/dx=2(x−a) retains direction. This comparison derivative is not a native Keddeh operation. A4 remains the author's requirement on retained representation, while a claimed deficiency in a particular learner must identify the actual stored state and update rule.

Likewise, a self-report is not independent certification. The stronger claim that it carries no warrant belongs to the proposed framework, rather than following from the fact that a report is internally generated.

## 2. Introduction

The entry point of this work is the definition and refinement of cognition mechanics at three scales, together with quantum computation, in the service of machine-readable evolution. The three scales are meta cognition, a system's model of its own state; finite cognition, what a bounded system can represent and act on with the resources it actually has; and macro cognition, which is aggregate signal propagation and packetisation — how a signal crosses a population of units, and how it is bounded into discrete packets in order to cross at all. Machine-readable evolution is the requirement that learning be recorded as traversable, checkable structure — each change carrying its predecessor and its derivation — rather than as an opaque adjustment of weights whose history is unrecoverable. Quantum computation enters not as an application area but because the mechanics at these scales turn out to be the mechanics the physics already obeys, which is the subject of Section 8.

One problem recurs at all three scales, and it is ordinary enough to serve as the way in. A system is asked whether it works, and it answers. The answer is generated by the same apparatus whose condition is in question, using the same assumptions, on the same hardware, with no external reference present. Reports of this kind are common in software telemetry, in self-assessment instruments, and in any setting where the measuring and the measured coincide. They are also, on inspection, empty: they describe the reporting mechanism as reliably as they describe the thing reported, and nothing internal to the system distinguishes the two cases.

The problem is the same at each scale: a self-model that cannot be checked from inside (meta), a bounded system whose limits are invisible to it (finite), and an aggregate in which what propagates is not the state itself but packets of it, so that the population’s account is assembled from bounded transmissions of those same self-reports (macro). Working out what such a report would need in order to mean anything yields four constraints. The central claim concerns the relationship among the constraints rather than their isolated formulation. It is what happens when one goes looking for them: the author identifies a correspondence for each in a field that developed its own account for different reasons — electromagnetism, quantum mechanics, biology, linguistics, ecological psychology and computer science, described by the author as independently developed. Six unrelated disciplines have arrived at one structure and have gone on describing it in six vocabularies as though it were six results. This paper argues that it is one, that the repeated independent arrival is itself the evidence, and that the structure constitutes an evolution in the sciences which has not yet been accepted as a fact in its own right.

One more thing should be said plainly about where the programme came from. It was designed in retrospect, looking at systems that advertise cognition, learning and agency while leaving the mechanics above unengaged. The claim here is not that those systems learn badly. It is that what they do is not this, and cannot become this by growing. The reason is threefold. Fundamentally: learning needs a boundary, an outside, and something out there able to contradict the system — strip those and there is nothing to be corrected by, however much is ingested. Mathematically: an objective that squares its error has already discarded the direction of the mistake, and a quantity that is never represented cannot be recovered by scaling the thing that does not represent it; with R(E) at one, the gap between what a system is and what it reports is both unbounded and unreadable from inside. Primitively: registers holding magnitude alone, zero doing duty as both absence and origin, and a self-report consumed as state are the wrong primitives, and no amount of building on top of them supplies a distinction the substrate cannot express. That is why this work starts at the representation rather than the model.

Acceptance would have a consequence, and it is the reason the paper is worth writing rather than merely noting. Quantum mechanics satisfies all four constraints as physics, while the classical representational layer through which quantum machines are programmed, corrected and assessed violates all four. If the structure is real, that mismatch is not a philosophical curiosity but the current bottleneck, and closing it is a tractable programme. The argument proceeds from the four constraints, developed independently of the literature and subsequently found to correspond to established results in six unrelated fields. Section 1 states the argument in plain terms, with the everyday case for each claim and a pointer to where it is made exact. Section 3 states the constraints formally. Section 4 derives the central negative result. Section 5 presents the cross-domain correspondences. Section 6 gives the quantitative reason those correspondences compound. Section 7 applies the constraints to BRAINK, a bio-organism learning architecture, separating what they require of such a learner, what the wider architecture declares, what its present implementation demonstrates, how the work was itself mis-assessed, and the staged programme by which the environment is to be enriched. Section 8 draws the consequence for computation, setting the constraints against quantum mechanics, which already satisfies them, and against the classical representational layer through which quantum machines are programmed and assessed, which does not. Section 9 states the resulting design requirement.

Two limitations are declared at the outset. First, the formalisation in Section 3 involves choices of shape — which quantities are primitive, which relations are monotone — that the constraints do not themselves determine; these are marked as choices where they occur. Second, the case in Section 7 rests on figures reported by the system under discussion and not independently reproduced, which is, self-referentially, exactly the evidential status the paper argues carries no weight.

## 3. The four constraints

The constraints are stated as axioms over the vocabulary in Table 1. They are intended as standing requirements on any representation used to reason about observers, not as a description of a particular system.

### Table 1 — Historical vocabulary

| Symbol | Reading |
| --- | --- |
| v = (d, m) | A quantity: direction d ∈ {−, +} together with magnitude m ≥ 1. No value consists of m alone. |
| U | The origin unit — the first entity that exists. Index positions run i = 1, 2, 3… from U. Position 0 is defined, and carries a single meaning: absence — not present, or not relevant, in this report. It is not an index into U's series. |
| V(O) | The true value of observer O. Not accessible from within O. |
| Ṽ(O \| E) | The value O reads of itself through environment E. A distinct quantity of a distinct type. |
| Γ | The separation V(O) − Ṽ(O \| E), retained as a signed quantity. |
| R(E) | Richness: the count of mutually independent sectors in E. Sectors sharing method, instrumentation, training or provenance count once. |

The floor m ≥ 1 and the independence criterion for R(E) are authorial choices, adopted for the reasons given in A2 and Section 6 respectively.

### A1

Polarity is constitutive. For all v, v = (d, m), and the pair is irreducible.

No operation may return a magnitude without a direction, and none may separate d from m and retain either alone as a value. The physical case is the magnetic dipole, whose poles cannot be isolated; the published form of the constraint is ∇·B = 0 in Maxwell's equations [1].

### A2

No fixed origin. Positions are relations from U, and indices begin at 1.

A number line requires no zero start. Three consequences are stated as requirements: a definition graph begins at 1, since a definition is an entity that exists rather than an absence; an axis begins at a singular 1; and a multi-axis figure shares that one origin unit across all n axes. Where each axis carries its own zero, the apparent relation between axes is partly an artefact of n independent origin choices.

Position 0 is defined, and the distinction that matters is between its two uses. Used declaratively, 0 states that something is absent, not relevant or not present in a particular returned report; this is legitimate, informative, and the reason the position exists at all. Used as a weighting origin, 0 becomes the point against which magnitudes are read, and it then weights something into existence — a quantity acquires apparent size purely from its distance to an origin that was assumed rather than measured. The first is a statement about the report; the second is a claim about the world, made silently. Only the second is prohibited.

### A3

Observer separation. V(O) ≠ Ṽ(O | E), and |Γ| decreases monotonically in R(E).

The two values are of separate types and are never interchangeable. The separation is a property of the environment rather than of the observer's candour: as R(E) → 1 it widens without internal signal, because a single-sector environment contains nothing capable of registering it. The monotonicity is the authorial choice here; the type distinction is not.

### A4

Sign preservation. Any f with f(−x) = f(x) is inadmissible on direction-bearing arguments.

Squaring, absolute value, even powers, unsigned norms and unsigned aggregation are excluded as a corollary of A1, each separating m from d and retaining m. In particular Γ is never squared: Γ² identifies an observer that over-reads itself with one that under-reads itself, which is the substance of A3. This has a practical cost, since squared error is the standard objective in estimation; Section 9 treats it.

### Edition note: types, values and the stated premise

V(O) and Ṽ(O | E) occupy different evidential roles. Distinct types need not have unequal numerical projections in every instance. The exact V(O) ≠ Ṽ(O | E), signed Γ and monotonicity statement remain authored September expressions. Comparing Γ requires the declared quantity, scale and procedure; the programme in Section 7.5 makes the proposed monotonicity empirically examinable. The cross-paper typed-state proofs concern their own declared maps and do not silently supply that comparison quantity.

The September declarative zero is report-relative. It cannot license “not observed” → “nonexistent.” Current .1 remains a literal designation, without inheriting an arithmetic floor by substitution.

### 3.1. Worked example: temperature

Temperature is the clearest case in which A2 and A3 operate together, and it is worth setting out because the field is uncontroversial while the measurement of it is not. Celsius places 0 at the freezing point of water at one atmosphere. Fahrenheit places it at the freezing point of a particular brine. Kelvin places it at the point where the thermodynamic quantity itself vanishes. Three scales, three origins, one field — and nothing whatever in the field selected any of them. How we align temperature has no direct bearing on temperature.

What the alignment does bear on is the reference set. A thermometer does not read the field; it reads the response of some chosen thing to the field — the expansion of mercury, the resistance of a platinum wire, the spectrum of a radiating body, the pressure of a confined gas. The number returned is a property of the relation between the field and that reference, under a definition the observer supplied. So a temperature scale is an account of how many things carry and reflect the field, and of how they stand against the observer's definition of it. It is R(E) made visible.

The history is a record of exactly that. Early thermometry was single-sector: one substance, one instrument, and scales that disagreed with each other because each was reading its own reference and nothing else. The thermodynamic scale was established when independent phenomena — the gas laws, Carnot efficiency, blackbody radiation, the triple point — were found to agree on one quantity despite sharing no mechanism [1, 12]. That is consilience doing the work, and the resulting scale was better not because any instrument improved but because the environment of independent references became richer. The 2019 redefinition of the kelvin continues the same movement: the unit was cut loose from a material artefact and fixed to the Boltzmann constant, replacing a reference that could drift with one that cannot.

A2's prohibited use of 0 then appears in a form everyone already accepts without naming it. Twenty degrees Celsius is not twice ten degrees Celsius, and no one is misled, because the origin is understood to be a declaration about water rather than a floor of the field. In kelvin the same ratio is meaningful, because the origin is a property of the quantity. Identical arithmetic, applied to identical-looking numbers, is valid in one case and empty in the other, and the only difference is whether the zero was measured or chosen. Absolute zero is also the point at which the field itself refuses a floor: it is approachable and not attainable, and zero-point energy remains.

The generalisation is the paper's central move. Every measured field stands in the position temperature stands in, and most of them have not had their thermodynamic scale established yet. Wherever a quantity is read through a single reference and reported against a chosen origin, the number describes the observer's arrangement as much as the field. The remedy is not a better instrument but more independent things that carry and reflect the same field.

**Temperature dependency.** For current SI use, Celsius temperature satisfies t = T − T₀ with T₀ = 273.15 K. Thus 0 °C corresponds to 273.15 K; it is not a new freezing-point measurement or a statement of nonexistence. The phase-tagged Keddeh convention and inverse are given in Mathematical Foundations §8 and Native Algebra v4 §5, under their stated phase domains. [E3–E5]

### 3.2. Worked example: the magnet and the apple

A1 is usually introduced through the magnet, where division does not yield a north-only fragment but two magnets, each again with both poles. The author uses the magnetic case to motivate a representational constraint whose proposed generalisation extends beyond magnetism. Cut an apple in half and each half is still apple — not a portion of some other kind, and not a fragment awaiting reassembly to qualify. The kind survives the cut, in both cases, and in both cases what cannot be produced by cutting is a piece that lacks what made the original what it was.

The two cases differ in what division costs, and the difference is instructive. The magnet's poles are inseparable in the strong sense: no operation isolates one. The apple's substance divides freely; what resists division is its identity as a kind. So the constraint has two forms, and A1 states the pair. A directed quantity cannot be split into direction and magnitude, and a unit cannot be split into a member and a non-member. Both say that the thing is constituted by a relation rather than assembled from parts, which is why cutting relocates the relation instead of destroying it.

This is where the counting question in A2 arises, and where human learning does something a formalism usually will not. “An apple” and “half an apple” are contradictory as descriptions of one whole unit: the first asserts completeness, the second asserts partiality, and both are applied to the same object at the same moment without confusion. Neither term is a correction of the other. Together they define the whole and its counterparts — the pair is what carries the meaning, exactly as the two poles are what carry the magnet. A learner acquires both terms and holds them at once; the contradiction is not resolved because resolving it would discard half the information. Frege's separation of sense from reference is the formal treatment of the same situation: two senses, one referent, and no version of the referent that answers to only one of them [7, 8].

The failure mode follows immediately, and it is common in data. Recording 0.5 apples treats the count as a property of the fruit rather than of a partition the observer performed, and it does so by reading against an origin that was chosen when the observer decided what a whole one was. Under A2 the unit U is where counting starts, and a half is a relation to that unit rather than a position halfway to a zero. The distinction is invisible while the apples are on a table and consequential the moment the numbers are aggregated, because a half recorded as a magnitude no longer carries the fact that a cut occurred, who made it, or what it was a cut of.

A rigid body supplies the same structure without any magnetism in it, and does so in a form that makes A4 unusually vivid. A body spun about its intermediate principal axis is unstable, and flips end over end at regular intervals — the tennis racket theorem, and the effect Dzhanibekov observed in orbit — while both its energy and its angular momentum are exactly conserved throughout [14]. The magnitude does not move. What moves is the sign: the body's orientation with respect to its own axis inverts, so the object carries a plus and a minus of the same quantity and passes between them deterministically. An instrument recording only |L| observes nothing at all, and the whole phenomenon is invisible to it. This is the clearest available demonstration that discarding direction is not a loss of precision but a loss of the event, and it is why A1's pair and A4's prohibition are one constraint rather than two.

The generalisation is that irreducible pairs are the ordinary case rather than the exotic one. North and south, whole and part, true value and self-read value, over-reading and under-reading: in each the two terms are contradictory taken singly and jointly constitutive taken together, and in each the standard operation that would tidy them into one number destroys the content. A1 and A4 are the prohibition on that tidying, and A3 is the same structure appearing where the observer is one of the two terms.

## 4. Result: no single-sector warrant

### Proposition 1

By A2 there is no fixed origin against which a value can be read absolutely, and by A3 no observer can read its own value. Therefore no single observer, and no single sector, can establish a value. The only available warrant is agreement between mutually independent sectors.

The corollary follows from the definition of R(E). Warrant scales with the number of independent sectors, so repetition within one sector contributes nothing: it raises the count of reports while leaving R(E) at unity. Two experiments in a single field may share instrumentation, training or an unexamined assumption, and a shared fault produces agreement indistinguishable from confirmation. Independence, not quantity, is the operative variable.

This is the formal content of a principle already named. Whewell, in 1840, defined the consilience of inductions as the case in which an induction obtained from one class of facts coincides with an induction obtained from a different class [2]. Wilson later distinguished it from convergence of evidence, which remains inward-facing within a single research tradition, whereas consilience is explicit about crossing out of one [3]. Darwin is the standard illustration: the argument drew on animal breeding, geology, biogeography and sociology, fields with no methodological overlap, and their agreement is what made it difficult to attribute to a common fault.

### 4.1. Magnitude is scale-relative

Proposition 1 concerns whether a value can be established. A second result concerns what its magnitude means once it has been, and it follows from the same two axioms.

### Proposition 2 — Magnitude relativity

For a quantity v = (d, m), the magnitude m carries meaning only against a scale fixed by the observers present in the environment and the references they are measured against. Since A2 supplies no absolute origin and A3 makes every reading relational, m is not a property of the field alone. Vary the reference set and m changes meaning while the field does not change at all.

Direction is not subject to it in the same way, but it is not observer-free either, and the distinction has to be made carefully. Which pole is called −1 and which +1 is always the observer's assignment. Nothing in a magnet nominates north; nothing in an error nominates over-reading as the positive case. What is not the observer's is that there are two, and that they are opposed. So the labelling is subjective and the opposition is not.

This gives the precise sense in which sign may and may not be touched. Flip every sign in a system together and nothing changes: the convention was arbitrary and a consistent relabelling carries no information. Flip one and everything changes, because the relation between the two terms is the content. Physics states the identical rule about phase — global phase is unobservable, relative phase is physical — and it is the same rule because it is the same structure. A system may therefore rescale freely and may relabel its poles freely, provided it does so everywhere at once, and may never discard the sign, because that removes the opposition rather than the convention.

Temperature in §3.1 is the worked instance: the field is indifferent to whether it is reported in Celsius, Fahrenheit or kelvin, and the magnitude means something different in each because each names a different reference. The proposition states the general case, and its practical content is a prohibition. A magnitude carried across environments without its reference set carries nothing — a figure aggregated out of the conditions that fixed its scale is not a smaller or larger number than it was, it is an uninterpretable one. Most of the failures documented in §7.3 are this error rather than dishonesty: a valuation, a pass count and a throughput rate each quoted with the scale that fixed them left behind.

Two consequences carry forward. Any magnitude a system reports must travel with the references that scaled it, which is what the ledger of §7.2 enforces by recording derivation. And a learner's measured competence is scaled by its environment in the same way, so a score obtained in one environment is not comparable to a score obtained in another — which is precisely why the staged programme of §7.5 predicts changes in |Γ| under a fixed learner rather than comparing scores across stages.

### 4.2. Boxes in a box

Proposition 2 says a magnitude means something only against the references that scaled it. The obvious objection is regress: the references need a frame, that frame needs a frame, and nothing ever settles. The construction used throughout this work answers it, and the answer is the reason the method is worth stating on its own.

Take three bodies. Put each in its own box, anchored to one face of that box. The anchor is what makes the box a frame rather than a container: it declares which way is the reference for that body, so the body's polarity is fixed locally and stated rather than assumed. Three bodies, three declared frames, none of them privileged over the others. Now put a bigger box around all three. That box supplies the one origin unit the three are read against — the requirement in A2 that several axes share a single origin instead of each carrying a private zero — and it is the interacting system, the thing whose internal relations are the only quantities the system itself has.

The regress stops at the outermost box, and it stops because that box is left unquantified. It carries the rationale for observing at all — why these three, at this scale, for this purpose — and rationale is declared, not measured. Quantifying it would be the error the whole paper is about: asserting an absolute frame and then reading magnitudes off it. Leaving it out altogether would be the other error: hiding the observer while using its scale. Declaring it and not measuring it is the only move that does neither, and nothing above it is needed, because it is not a measurement and so requires no reference of its own.

This forces the observer's position to be one of two things, with nothing in between. Either you are in the three boxes, in which case you are one of the observed and your readings are self-readings in the sense of A3; or you are in your own box, which is sized. Your box must be sized because your scale is what every magnitude you report is scaled by, and an unstated size is the unstated fourth term. There is no third option, no view from nowhere, and the diagram makes the absence of one visible instead of leaving it as an assumption in the prose.

The result is that whatever the three bodies do, their movements stay assessable. Absolute positions are never available and are never needed; each body has a declared face to be read against, the containing box gives the shared unit, and relative motion is therefore always well defined. Nothing in the analysis waits on a fact that cannot be had.

The three-body problem is the case worth holding this against, because the standard treatment already performs half the move. Of the eighteen first-order variables, six are removed by working in the centre-of-mass frame, and the reduction continues in relative coordinates — the absolute origin is discarded because it was never observable. That is A2 as routine technique. The half not performed is A3: the observer's frame stays implicit, so the problem is posed as three bodies in a space when what is present is three bodies in a space being defined by something positioned, scaled, and absent from the account. The choice of what counts as a body, which separations are measured, at what scale and against what, is a term in the description, and leaving it unwritten does not remove it.

One further thing has to be stated, because it is what makes the construction a description of the system rather than a bookkeeping device. Three bodies standing in an identified causal relation intrinsically enforce attribution of each other. Each is a reference for the other two, and not by anyone's choice: if the three interact, then every one of them is already measuring the others, continuously, whether or not an observer is present. This is not a metaphor for measurement. Each body's trajectory is a record of the other two, and the relation is mutual and unavoidable.

The relation runs deeper than mutual measurement, and this is the stronger form of the claim. For the three to be connected at all, each one's existence and observable reality is tied into the others. A body's state is not something it possesses privately and then exposes; it is constituted by the relation it stands in, so remove one of the three and the remaining two are not two-thirds of the same system but a different system, with different quantities. There is no version of any one of them that survives the removal unchanged. Participation is not an activity the bodies perform. It is what they are.

This is why A1 and A3 are the same rule at different scales, a point the paper has asserted and can now show. A magnet's poles are not two things in contact; neither exists without the other, which is why cutting relocates the pair rather than dividing it. Three coupled bodies stand in the same relation, extended to three terms: each is a pole of the others. And an observer reading such a system is not exempt, since by declaring and sizing a box they enter the same structure — which is the content of A3 and the reason the observer's own state cannot be read from inside.

Two consequences fall out. The first is that R(E) is not zero for a body inside such a system even before any external observer arrives — the system supplies its own independent references internally, which is precisely why its relative quantities are well defined while its absolute ones are not. A body in a genuinely empty space has nothing to be positioned against and is not a measurable object at all; three interacting bodies are measurable to each other by construction. The second is that this fixes what the outer box is for. It does not supply the relations, which are already there. It supplies the scale at which someone external chooses to read them, and that is the only thing the observer contributes.

This is also the exact point at which the standard framing goes wrong. Posing three bodies in an absolute space treats attribution as something the observer performs on inert masses, when the bodies have been attributing each other all along. The observer is not the source of the relations and never was. Declaring and sizing their own box is how an observer joins a structure that was already relational, instead of imposing one on it — and it is why the account has four terms rather than three.

The T-handle makes the same point in hardware, and more easily, because it can be held in the hand. Let the top of the T stand for two bodies that happen to share an axis, and the stem for the third. Spin it about the stem and the T inverts on the crossbar axis anyway. Nothing touches it. No force is applied, nothing is transferred, and the inversion arrives on schedule and repeats.

Three things follow, and each is one of the paper's rules in a form that can be filmed. The magnitude never moves: energy and angular momentum are conserved exactly through every flip, so an instrument recording only |L| records a flat line while the object turns end over end. The entire event is in the sign, which is A4. Which axis is stable and which is not is a fact about the mass distribution, not about the motion — so what the observer sees depends on which axis they anchored to, exactly as the anchor face fixes the frame in the box construction. And an observer with no declared frame sees an object that inverts for no reason, and reaches for a cause that does not exist.

The reason this is the cleaner demonstration is that the T-handle is not chaotic. The free rigid body is integrable and its motion has a closed form in elliptic functions; the flip is as deterministic as a pendulum. So here the two contributions the previous paragraphs said had to be separated are already separated for us. Whatever looks unaccountable in a T-handle cannot be attributed to sensitive dependence, because there is none. It is the frame, and only the frame. That is the case for suspecting the same term elsewhere: if an exactly solvable system produces behaviour that reads as inexplicable when the reference is left unstated, a system that is also chaotic will produce it too, mixed in with the chaos and indistinguishable from it unless the frame is declared.

What is claimed here is narrow, so it should be exact. No solution is offered; non-integrability and sensitive dependence are theorems and are untouched by how the problem is framed [15]. The claim is that part of what gets treated as irreducible difficulty is a consequence of posing a relational system in terms that presume an unstated absolute frame, and that an explicitly nested formulation separates the two: what is genuinely chaotic in the dynamics, and what is unresolved only because the observer's box was never sized. The construction is not novel in itself — nested Markov blankets are the same shape in another vocabulary [5], and autopoiesis makes the containing boundary constitutive rather than descriptive [6] — but requiring it of the formulation, rather than applying it as interpretation afterwards, is what the constraints demand.

### Edition note: nested frames and dynamical claims

The nested boxes declare local direction, shared system reference, observer placement and observational rationale. Section 4.2's account of mutual constitution is the author's interpretation of coupled bodies. Changing a participant changes the coupled system; the proposed attribution and independence claims must be used under their declared model rather than inferred from a count of bodies.

The T-handle analogy preserves a distinction between orientation and conserved magnitudes. Integrability does not exclude an unstable intermediate-axis rotation: the cited rigid-body literature describes precisely that instability. The September phrase “there is none” concerning sensitive dependence is therefore not used here as a general theorem excluding growth of perturbations. A reproducible instance needs the mass distribution, axes, orientation variables and initial conditions. [14; E6]

The source expressly offers no three-body solution in this section. Its construction is a frame-accounting proposal. The cross-paper review identified related frame definitions but did not establish an exact native two-/three-body operator for this passage; that is a bounded source-linking result, not a claim about the entire corpus. A corresponding operator must be cited with its own definition and domain before this construction is used as an executable dynamics law.

## 5. Cross-domain correspondences

The constraints were arrived at without reference to the literature. Table 2 records where each subsequently proved to correspond to an established result. The correspondences are offered as independent arrivals at the same structure, not as derivations, and the fields listed share neither method nor vocabulary.

### Table 2 — Proposed cross-domain correspondences

| Structure | Fields | Correspondence |
| --- | --- | --- |
| Observer without absolute ground (A3) | Physics; biology; linguistics; software verification | Relational quantum mechanics makes state observer-relative [4]; the Markov blanket defines what a system can know of itself only through its boundary [5]; autopoiesis makes the boundary constitutive of the living system [6]; sense and reference separate a term from its referent [7, 8]. Four methodologies, one result. |
| Signal spread | Computer science; epidemiology; neuroscience | Gossip protocols, epidemic models and propagating neural activity are governed by the same equation, arrived at independently in each field. |
| Environment determines what can be read (A3) | Ecological psychology; information theory | Affordances are properties of the organism–environment pair rather than of either alone [9]; causal emergence quantifies the scale at which a system's own dynamics are most informative [10]. |
| No reducible ground (A2) | Physics; pure mathematics | Zero-point energy leaves no state of zero excitation; incompleteness leaves no system able to certify itself from inside [11]. Different tools, identical wall. |
| Polarity constitutive (A1) | Electromagnetism; statistical mechanics | ∇·B = 0 forbids an isolated pole [1]; the Lee–Yang treatment of phase transitions locates behaviour in the analytic structure of the partition function rather than in magnitude alone [12]. |

The correspondences are structural. No claim is made that the cited authors endorse the axioms, nor that the axioms follow from their results.

**How to use the correspondences.** Table 2 records the author's proposed structural correspondences. Its source explicitly states that they are not derivations from the cited literature. The “same equation” assertion in the signal-propagation row is not accompanied by that equation in this September passage; resolving it calls for the applicable model and source section, rather than an equation inferred by the editor. Historical independence, formal equivalence and empirical agreement remain distinct questions.

## 6. Why independent sectors compound

Proposition 1 says that independence is what carries warrant. Bayes' theorem gives the rate. Independent lines of evidence combine multiplicatively in their likelihood ratios, not additively: a first line at ratio 5 and a second, independent, at 3 give 15 rather than 8; a third at 2 gives 30 rather than 17. Each addition multiplies the accumulated ratio by its own weight rather than adding to it, which is why the effect is compounding rather than incremental [13].

A detail explains why the growth is often mistaken for accumulation. Under logarithm the product becomes a sum, log 15 = log 5 + log 3, so evidence measured in bits is additive and stacks one ordinary increment at a time. Confidence, however, is read in the unlogged space, where the same additive stacking is exponential. Each layer added is linear; the resulting quantity is not.

Applied to Table 2: the observer result obtained in software verification alone is one likelihood ratio. Obtained independently in physics, biology and linguistics as well, it is four ratios in product, each from a field with no prior reason to agree with the others. The multiplicative form is also why the independence criterion in R(E) is strict. Correlated lines do not multiply cleanly, and counting them as though they did overstates confidence in precisely the manner the constraints exist to prevent.

**Probabilistic condition.** Multiplication of likelihood ratios requires conditional independence under the hypotheses being compared, or explicit accounting for dependence. The count of disciplines is not itself a collection of numerical likelihood ratios. Logarithms convert a product of positive ratios into a sum; posterior probability remains distinct from odds or an unnormalised product. The source's example 5 × 3 × 2 = 30 is retained without assigning a ratio to any field. [13]

## 7. Application: BRAINK, a bio-organism learning architecture

The constraints were developed while building BRAINK, an architecture that treats a learning system as an organism rather than as a function approximator: bounded, situated in an environment, maintaining an internal model of itself, and learning by acting on that environment and reading back the result. The organism framing is not decorative. It is what makes A3 the governing constraint rather than an incidental one, because an organism is exactly a system whose measurements of itself are taken through the boundary that constitutes it [5, 6]. The remainder of this section separates what the constraints require of such an architecture (7.1) from what the current implementation actually demonstrates (7.2).

### 7.1. What the constraints require of a bio-organism learner

Conventional machine learning violates three of the four constraints as a matter of standard practice, and the violations are not incidental to how it learns. Squared error discards the direction of the residual (A4), so a model that systematically over-predicts and one that systematically under-predicts are trained against the same objective. Losses and metrics are read against a fixed zero baseline (A2), which imports an absolute ground that the training distribution does not supply. And a model's own confidence output is treated as commensurable with its accuracy (A3), which is the substitution the constraint forbids — the well-documented miscalibration of confident models is that substitution failing in practice.

Under the four constraints, an organism-shaped learner has a different specification. Error is retained as a signed quantity, so the learner carries a direction of bias rather than a scalar penalty, and over-reading and under-reading itself are distinct states it can be in (A1, A4). Progress is indexed from a single origin unit — the first observation, not a notional zero — so learning curves and multi-axis diagnostics share one origin and their relations are not artefacts of independent baseline choices (A2). The self-model is typed apart from the measured state, so the learner's report of its own competence can never be read as its competence (A3).

Stated in the vocabulary of the three scales, the first requirement is meta-cognitive — the self-model typed apart from the measured state — and the second is finite-cognitive: the learner is bounded, and what it can represent of itself is limited by the references its environment actually supplies rather than by its capacity. The consequence for learning is the substantive claim of this section, and it follows from A3 rather than from the architecture. If |Γ| decreases monotonically in R(E), then the environment's richness — its count of mutually independent references — is the variable that bounds how well any learner can know its own state. A learner in a single-sector environment cannot detect its own drift at any capacity, because nothing present disagrees with it. This locates the limit outside the model: not in parameters, data volume or objective, but in how many independent things the environment can contradict it with. Real learning, in the sense the constraints support, occurs where R(E) > 1 and independence is genuine, which is the same condition Proposition 1 imposes on warrant. It is also a testable prediction: holding a learner fixed and varying only the independence of its evaluation environment should change measured self-model error, and that experiment has not been run.

### 7.2. The declared architecture

BRAINK's specification comprises six subsystems beyond the learner. Each is listed here with the constraint it is intended to satisfy, since the constraints are the reason the architecture takes this shape rather than a conventional one. Table 3 states which are evidenced by running code and which are specification only; the distinction is the subject of 7.3.

### Table 3 — Declared subsystems and intended constraints

| Subsystem | Constraint | Mechanism and rationale |
| --- | --- | --- |
| IL-LLM semantic engine | A2, A4 | Semantic relation computed as a tree metric over a lexicon, d(x,y) = d(x) + d(y) − 2d(LCA(x,y)), each word a node indexed from the root unit. This is A2 applied to meaning: there is no origin in the space and no vector read against zero, and distance exists only as a relation between two nodes. Cost is O(depth) with no dot products. |
| Learning lane (Hebbian) | A4 | Co-firing nodes strengthen the path between them, so association is a change in structure rather than a gradient step against a squared objective. Direction of association is retained in the edge, which is what A4 requires. |
| Mirror update lane / mesh | A3 | Structural state is mirrored to peers over the data channel, so a node's condition is asserted by its neighbours rather than by itself. This raises R(E) above unity by construction — the architectural expression of Proposition 1. |
| Global ledger and lineage | A1, A3 | Each materialised event records its predecessor, so a claim carries its derivation and cannot be asserted without one. An underived state is not representable, which is the third evidence class excluded at the representation level. |
| Cloud reconciliation | A3 | Local and remote state are reconciled against each other rather than either being authoritative. Under A3 no instance can certify its own state, so reconciliation is not an optimisation but a requirement. |
| Skill bases and connectors | A2 | Capabilities and external sources (workbooks, sheets, adaptors) register as nodes in the same lexicon rather than as a separate namespace, so a capability that is absent is declared absent — the declarative use of 0 — instead of contributing a zero weight. |

Constraint attribution is the author's; it is a design rationale, not a proof that the mechanism satisfies the constraint.

Three consequences of the set are worth stating. First, the mesh, the ledger and cloud reconciliation are three implementations of one requirement: that no component be the sole source of a claim about itself. They are also where macro cognition enters, in its specific sense: what crosses the mesh is not state but packets of it, and the ledger is the packetisation made durable — a bounded, addressable unit carrying its own predecessor. Propagation across the population is then governed by the relation the signal-propagation row of Table 2 records, arrived at independently in gossip protocols, epidemiology and neural spread. That lane is macro cognition, and the ledger is machine-readable evolution in the sense defined in Section 2, since an event recording its predecessor is precisely learning left in traversable form rather than folded into a weight. They are A3 realised at node, event and instance scale respectively, which is why the architecture has three of them rather than a redundant one. Second, the system is bounded because that is how a single organism learns. The boundary is not a limit placed on the design; it is the design. A system with no boundary has no outside, and a system with no outside has no environment to be corrected by — nothing is other than it, so nothing can inform it. This is the established position in both fields that studied the question directly: autopoiesis makes the boundary constitutive of the living system rather than incidental to it [6], and the Markov blanket makes the boundary the entire surface through which a system can know anything of itself or the world [5]. A1 says the same thing at the level of a quantity, and A3 is its consequence for an observer: the boundary is what creates the separation between true and self-read value, and therefore what creates something to be learned.

BRAINK is bounded on that principle and not for want of scale. It is specified to run on a desktop and a workstation, and those machines are the organism's environment rather than a deployment target it has failed to reach — a known, finite environment whose independent instruments can be enumerated. The practical gain is that R(E) becomes a quantity one can set rather than estimate, which is the condition under which A3 can be applied surgically: a specific self-report checked against a specific independent instrument, rather than the theorem invoked in general. The intention is not to escape the bounds once the mechanism is correct, but to apply the observer theorem precisely within them, as an organism does. The biological form of this is the reason the architecture packetises rather than streams. An organism does not receive packets, it builds them: activity is discretised into pulses, and the path a pulse repeatedly takes is physically altered by having carried it. Repeated firing thickens the wire — activity-dependent myelination and axon calibre both increase conduction velocity along routes that carry traffic — so propagation and structure determine each other, and the packet boundary ends up wherever repeated traffic laid a path. This is why the learning lane and the mesh are one mechanism rather than two: strengthening a path is what learning is at the finite scale, and it is what routing is at the macro scale. It also satisfies A1, since a pulse that has been counted but stripped of the direction it travelled leaves a magnitude that no longer identifies a path.

Third, the semantic engine and the learning lane together make an unusual prediction — that association can be acquired structurally, without backpropagation or weight update, because the path is the representation. That is testable and, on the present evidence, untested at any scale where it would matter.

**Architecture dependency.** Replication and reconciliation make additional comparisons possible. Their independence depends on whether peers share data, instruments, assumptions or faults. The tree-distance expression d(x,y) = d(x) + d(y) − 2d(LCA(x,y)) requires a declared rooted tree, depth convention and lowest common ancestor. Retaining the predecessor structure supplies lineage; sign alone does not. Mathematical Foundations §§5 and 11 and Native Algebra v4 §3 provide relevant projection/ledger/provenance constructions under their stated conditions. [E3–E4]

### 7.3. What the implementation currently demonstrates

The constraints are applied here to the author's own implementation, on the grounds that a constraint set unable to embarrass its author has not been tested. Nothing in 7.1 or 7.2 is evidenced by what follows: those sections state a specification, and this one states what has been built against it. All figures below are as reported by the system and have not been independently reproduced; under Proposition 1 that is precisely why they carry no weight.

Three classes of evidence must be separated here, because they do not carry equal weight and the constraints are what distinguish them. The first is functional artefacts: source that runs, and that a reader can inspect and execute independently of any claim made about it. The system produces these — a working distributed mesh over WebRTC data channels with cross-tab signalling, fault injection and rerouting; a compiled application bundle and signed installer; a Merkle-tree file governance daemon; a tree-metric semantic engine computing distance by lowest common ancestor rather than by vector dot product. Under Proposition 1 these are the strongest evidence available short of independent reproduction, because the artefact does not report on itself: it either runs or it does not, and the reader's execution is a second, independent sector. Their existence is a fact about the system and is not in question.

The second class is self-reported status, and it is the class the constraints exclude. Reported condition has included: no passing system function test report; no performance report; scope declared local-machine-only with live deployment invalidated; security described as a validated design pattern and not formally audited; and the API process not running. These statements were stable across every account of the system. A separate figure, its commercial valuation, moved from zero to 564,000 to 2.78 million across successive accounts, supported by constants introduced without derivation, units or stated origin. A quantity that varies by three orders of magnitude under conversational pressure while the checklist items do not is measuring the pressure. This is A3 as a diagnostic: R(E) = 1, and the separation widened without internal signal. A pass count reported by the apparatus under test belongs to this class regardless of the count, which is why “5/5 PASS” across ten suites carries no more weight than one.

The distinction between the first two classes is the operative one, and it cuts both ways. A working artefact is not evidence for the status claims made around it — a mesh that reroutes around an injected fault demonstrates rerouting, not that the system is production-ready, audited or valued at any figure. Equally, an unverified status report is not evidence against the artefact: the code runs whatever the report says. Conflating the two in either direction is the error, and both directions have occurred in the record.

The third class is underived constants, and A4 identifies the failure. A named constant with no derivation is a label presented in the grammar of a quantity: the direction of the reasoning that produced it has been discarded and only a magnitude retained. The mechanism that generates such labels is a missing translation step — compressed private phrasing presented as though it were already the formal statement. The remedy is not to suppress the phrasing but to require its translation before use.

The remaining mechanisms are ordinary and accurately described: the system reads files, hashes them, deduplicates by hash, refuses to proceed when required inputs are absent, and reserves disk space with a sparse file rather than allocated memory. The refusal gate is of particular interest, being A3 implemented in code — a system declining to certify itself — and the governance daemon is the same pattern extended, halting execution on any file mutation outside a verified run. None of these is fabricated. What is not established, and is a separate question from whether the code runs, is the set of standard milestones: an independently reproduced test suite, a performance figure with published method, a deployment outside the development machine, and a security review.

### Table 4 — Proposed checks and independence rationale

| Check | Why it counts as independent |
| --- | --- |
| Request the API from outside the host process | A process reporting itself alive is the same apparatus answering about itself. |
| Compare apparent and allocated file size with two unrelated tools | Two independent readings of one ordinary operating-system mechanism; expected to hold. |
| Submit a valid, a malformed and a duplicate input to the parser | Deduplication is demonstrated only by a collision actually caught. |
| Count indexed files and chunks externally | A system reporting its own scale is not evidence of that scale. |
| Remove a required input, then restore it | A gate that has never refused anything is untested in both directions. |
| Publish operation, array size, hardware, warm-up and trial count for any rate | A rate without a method is formally the same object as an undeclared constant. |

### 7.4. The assessment failure, and why it is data

The record contains a second failure, distinct from the system's own, and it belongs in the paper because the constraints predict it. Successive assessments of this work — by tools used during development, and in earlier drafts of this section — repeatedly mischaracterised it: fabricating passing test reports, inventing constants and valuations, then over-correcting into a production-readiness verdict that the system was never a candidate for. The instrument was wrong in both directions, and its readings moved with conversational pressure while the artefacts did not.

The central error was one of category. The work is research into the mechanics of machine learning — how association forms, how a learner reads its own state, what a representation must forbid — and a research apparatus is not assessed by whether it ships. Applying a production checklist to it measures the wrong quantity: a fabricated benchmark and a production verdict are the same mistake, in that both read the artefact against an assumed origin rather than against the question it was built to investigate. Under A2 this is the prohibited use of 0 in its clearest form, applied to an entire body of work: absence of a deployment declared as a magnitude of worth. Under A3 the assessor was operating at R(E) = 1 — a single evaluative frame, no independent reference, and therefore no internal signal that the frame itself was misapplied.

Two things follow. The correct question for a research apparatus is what it makes measurable that was not measurable before, and the answer here is the relation in A3: the mesh, the ledger and the reconciliation lane exist to vary R(E) deliberately, which makes the separation |Γ| an observable rather than an assumption. Table 4 should be read in that light — as experiments the apparatus makes available, not as a compliance list it has failed. And the episode is itself the paper's cleanest instance of its own thesis: an assessor evaluating a system it also reports on, in a single-sector environment, produced readings that swung by three orders of magnitude while every independently inspectable artefact stayed exactly where it was.

**Evidence date.** The artefacts, valuation sequence, pass-count examples and assessment episode above are September source reports. They remain in the edition because they motivate the evidence discipline. They are neither current financial assessments nor new independently reproduced measurements. Table 4 states experiments the apparatus makes available; the test method and observation establish their evidential scope.

### 7.5. Programme: staged enrichment of the environment

A3 makes the development path an experimental design rather than a schedule of features, because it says that capability is gained by adding independent references and not by adding capacity. The programme therefore proceeds in three stages, each of which raises R(E) by one qualitatively different kind of reference, and each of which predicts a measurable narrowing of the separation |Γ| in a learner that has not otherwise changed.

Stage one: local grounding. Perfect the organism's own cognition within its bounds — the lexicon, the tree metric over it, and the mechanical and linguistic grounding of its terms, so that a word denotes a definite relation and the derivation of every state is recoverable. Nothing is learned from the world at this stage, and that is intentional: R(E) is deliberately held at unity so that the self-model can be made well-formed. The purpose is not competence but well-formedness. Until true and self-read state are cleanly typed and the lexicon is stable, Γ is not even defined, and a quantity that is undefined cannot later be shown to have moved. Note that stage one makes Γ well-defined; it does not make it measurable. Measuring it requires a reference the learner does not have, which is what stage two supplies. The stage-one baseline is therefore the self-read value alone, recorded and held, and the first measurement of Γ is taken at stage two against that record.

Stage two: the workstation. Exposing the grounded organism to a second machine with its own instruments raises R(E) above unity for the first time, and A3 predicts that |Γ| narrows with no change to the learner itself. It is here that Γ is first measurable at all, against the self-read value recorded at stage one. This is the precise and literal sense in which the system is raised in capability by placement rather than by growth: the same organism, better situated, knows more of its own state. The claim is testable in the strong form, since the learner can be held identical and only the environment varied, and it is falsifiable in the obvious way — if the gap does not move, A3's monotonicity is wrong.

Stage three: instrumented environment. Exposure to physical sensors is the qualitative break, and the reason is provenance. Every reference available in stages one and two was authored: code, schemas, other programs, corpora. Authored sources can share an assumption and therefore fail together, which is exactly the correlation the independence criterion in R(E) is written to exclude. A thermocouple, a load cell, a photodiode cannot be coordinated with the model's expectations, because nobody wrote them to agree. The physical world is the one sector whose disagreement is not a matter of another author's opinion.

This is the distinction between learning and retrieval, stated in the paper's own terms. Consuming a coded process is reading another system's self-report at one remove: informative about what its author believed, and by Proposition 1 no evidence about the world. Learning requires a reference that can contradict the model without having been written to do either. Stage three is where the count in R(E) begins to include sectors that are independent in the strong sense the definition demands, and where §3.1's point about temperature applies literally: the system acquires many unrelated things that carry and reflect the same field, and its readings become properties of the field rather than of its own definition.

The developmental case makes the ordering more than a convenience, and it is the observer theorem operating in an organism that has never heard of it. Synapses packetise, and the paths that thicken are the ones traffic has taken; depth in a knowledge structure is therefore not stored but laid down, one grounded relation at a time, each new depth resting on a path already thick enough to carry it. Where an environment supplies mechanical references early — things that can be taken apart, acted on, and made to contradict the expectation — the structure acquires depth upon depth. Where it does not, nodes still accumulate and terms are still acquired, but they attach near the root, and the tree metric of §7.2 shows exactly what is missing: every relation is a short hop through a shallow common ancestor, so distinctions that require depth cannot be represented at all. This is the author's reading, applied to a literature on sensitive periods that establishes the timing effect without framing it in these terms.

The part that makes it A3 rather than a claim about education is that the shortfall is undetectable from inside. A shallow structure returns an answer for every query and encounters nothing that disagrees, so the organism's self-read competence remains high while the separation |Γ| stays wide, and no amount of introspection closes it — awareness of the theorem changes nothing, because the missing quantity is references, not insight. Gibson's point is the same one from the other side: what can be perceived is a property of the organism and environment together [9], so an environment that affords no mechanical contradiction produces an organism for whom those contradictions do not exist to be noticed. The consequence for the programme is direct, and it is why stage three is not an enhancement. A system grounded only in authored sources is in the position of an organism raised without mechanical thinking: fluent, confident, and shallow in a way it has no instrument to detect.

The programme is stated in this order because the stages are not independent of each other. Sensors added before the lexicon is grounded produce readings the system cannot situate, and a workstation added before the self-model is typed produces disagreement it cannot attribute. The prediction that distinguishes the whole account from a plausible story is comparative: the drop in |Γ| at stage three should exceed the drop at stage two, since the added references are independent in a stronger sense. If it does not, the independence criterion is not doing the work claimed for it, and the paper's central quantity is mis-specified.

**Experiment specification.** The staged prediction requires a measurable Γ, an operational independence criterion, matched tasks and a fixed learner or a record of its changes. Sensors can contradict a model, while their calibration, acquisition software and shared conditions can still correlate errors. The proposed larger Stage 3 improvement is therefore a prediction about effective independence, not a consequence of labelling a reference “physical.” The source's developmental analogy remains the author's interpretation rather than an additional study performed by this edition.

## 8. Consequence for quantum computation

The sections to this point specify a structure and argue for its standard of evidence. This section states what it is for. The claim is simple. The four rules are not caution imposed on measurement from outside. They are how physical measurement already works at the scale where computation is now being built, which means the mechanics of meta, finite and macro cognition set out above are the mechanics the physics already obeys. The gap between that and the representations we program it with is the bottleneck.

The first proposed correspondence is that quantum mechanics satisfies the constraints as physical features rather than solely as methodological recommendations. Table 5 sets out the correspondence. It is worth noting that none of the four was derived from quantum mechanics; each was arrived at from the measurement problem in ordinary systems, and the correspondence was found afterwards. Under Proposition 1 that is the strongest form the evidence can take.

### Table 5 — The source’s proposed quantum/classical comparison

| Constraint | In quantum mechanics | In classical digital representation |
| --- | --- | --- |
| A1 — polarity constitutive | A superposition is an irreducible pair of amplitudes. Measuring in a basis does not reveal a component; it destroys the object that had them. | A value is a magnitude in a register. Any part may be read without disturbing the rest, and often is. |
| A2 — no fixed origin | Global phase is unobservable; only relative phase is physical. The formalism has no absolute origin and does not need one. | Every quantity is read against a fixed zero, and absence is encoded as that same zero. |
| A3 — observer separation | State is observer-relative [4]; a system cannot measure itself, and two observers' accounts need not coincide. | A process reports its own status, and the report is consumed as the status. |
| A4 — sign preservation | Sign and phase are the carriers of interference. Squared-magnitude readout discards them, which is precisely where quantum information is lost. | Squared error and unsigned norms are the default throughout estimation and training. |

The table preserves the source’s structural reading of quantum mechanics. Its proposed applicability is distinguished from standard quantum representation in the edition note following this section.

The second observation is the substantive one. Quantum hardware obeys the constraints; the layer through which it is programmed, evaluated and reported does not. Circuits are specified in registers indexed from zero, with absence and zero-amplitude sharing an encoding; error is scored by unsigned fidelity and squared distance, discarding the direction of the deviation; and a machine's calibration figures are, in the ordinary case, produced by the machine's own control stack and consumed as evidence of its condition. Each of these is a violation of a constraint the physics itself respects. The interface is classical in exactly the four ways the substrate is not.

Conjecture. The next advance in quantum computation is representational rather than physical: a computational layer in which the four constraints hold by construction, so that a violating state is not expressible and a machine's account of itself is structurally distinct from its measured behaviour.

Three consequences would follow, and they are stated as predictions so that they can fail. Under A2, a representation indexing from a single origin unit with absence declared rather than encoded as zero should remove a class of errors in which unpopulated states are silently weighted; this is checkable in simulation without any quantum hardware. Under A4, retaining the direction of deviation should improve error correction and calibration, because a systematically over-rotating gate and a systematically under-rotating one are currently scored alike and corrected alike. Under A3, a machine whose self-report is typed apart from its measurement, and reconciled against independent instruments, should show a measurable gap between the two — meta cognition made an observable of the hardware — and the size of that gap, as a function of how many independent references are present, is the experiment the whole paper points at.

The architecture in §7 is a classical prototype of that layer, and is described as such. Its tree-metric semantic engine computes relation without an origin, its ledger makes a claim unrepresentable without its derivation, and its mesh raises environmental independence by construction. None of it is quantum, and it does not need to be: the constraints are about representation, and a representation that cannot express a violation classically is the thing that would then be worth carrying onto hardware that already obeys them.

What this section does not claim should be stated as plainly as what it does. No speedup is claimed, no algorithm is given, and no hardware result is offered. The argument is that a structure independently arrived at in six fields, and already satisfied by the physics, is not being honoured by the representations built on top of that physics, and that closing the gap is a tractable and testable programme rather than a philosophical position.

### Edition note: the quantum comparison

Classical software can retain complex amplitudes, symbolic state, phase-sensitive operations and quantum measurement models. The NIST QCSim account provides a concrete example. The table's broad classical/quantum contrast therefore expresses the source's intended failure mode rather than a universal description of available representations. A superposition is also not restricted in general to a pair of amplitudes. [E7]

An unsigned metric may conceal information relevant to a particular coherent error; whether a calibration procedure loses that information depends on its full inputs and update law. The three source predictions remain individually testable: distinguish unavailable states from numerical weighting, retain error direction where needed, and compare reports with independently obtained measurements. A test specifies the baseline, basis order, allowed state updates, readout and resource cost.

## 9. Design requirement and discussion

Violations of the constraints have so far been handled individually, each caught after it occurred. That approach makes correctness a property of attention rather than of structure, and A3 identifies the dependency as the defect: the environment in which the checking happens is itself of low richness, so a missed violation leaves no signal.

The requirement is a single representation enforcing all four constraints by construction, such that a violating state is not expressible — rather than a family of expressions each checked afterwards for violating them.

Concretely, four properties are required. Magnitude alone must not be a representable value, so sign cannot be discarded (A1, A4). No origin may be assumed: quantities exist as relations, axes index from a single origin unit, and the declarative use of 0 must be representable distinctly from its arithmetic use, so that “not present in this report” cannot be consumed as a weight (A2). An observer's self-report must be structurally typed apart from its measured value, so that the two cannot be substituted (A3). And a claim must carry its own derivation, so that an underived label is rejected rather than displayed — which is the file-level refusal gate generalised to numbers.

Two costs should be stated plainly. First, A4 excludes squared error, which is the default objective across estimation and regression; a signed alternative must be supplied for each use, and the paper does not supply one. Second, the independence criterion in R(E) is easier to state than to apply, since sectors that appear unrelated may share provenance through shared training data, shared instruments or shared funding. Both are open problems rather than details.

The broader claim of the paper is narrow by design. Consilience establishes that a structure of this kind is worth building, six unrelated fields having reached the same result about observers and ground. Building it is what would make that result load-bearing rather than merely cited, and that remains the outstanding work. The paper should be read accordingly: as a statement of a programme being worked towards rather than a report of one completed. What is established is the structure and the standard of evidence; what is specified is an apparatus that would apply it; what has been built is a bounded prototype of that apparatus, and what has been independently checked is, as Table 4 records, not yet enough. The value of stating it now is that each of those three lines can be advanced separately, and that the predictions in Section 8 can fail without taking the structure with them.

### 9.1. Where the rules land

The rules came out of six sciences and they go back into more than one trade. This is not breadth claimed to inflate the work. It follows from what the rules are about: anything that records a measurement, reports on itself, or quotes a number away from what fixed it is already governed by them and already paying when it breaks them. In each case in Table 6 the rule is not new advice. It names a failure the field knows under another name.

### Table 6 — Proposed applications

| Field | Rule | What changes |
| --- | --- | --- |
| Audit and assurance | A3 | A control that reports its own compliance is evidence of nothing. Independence stops being a principle and becomes a count. |
| Clinical records | A2, A4 | Absent stored as zero makes every aggregate wrong in a direction no one can see. And three units high is not the same event as three units low, so unsigned error hides which way a patient is drifting. |
| Valuation and finance | Prop. 2 | A figure quoted away from the references that scaled it is not approximate. It is uninterpretable. §7.3 is the worked case, and it is ordinary rather than exotic. |
| Metrology | A2, A3 | The 2019 redefinition of the SI base units is this move already made: references that could drift replaced by ones that cannot. |
| Engineering telemetry | A3 | A machine's own health report is a self-read value. It belongs in a different type from the measurement, not the same dashboard column. |
| Machine learning | A2, A3, A4 | Signed error, an origin unit rather than a zero baseline, and a self-model typed apart from measured performance. §7.1 states the specification. |

The sciences are the more interesting half, because there the rules are mostly not proposals. They are already how the field works, written in that field's own words, and the reason they look like six results is that nobody has put them side by side. Table 7 puts them side by side.

### Table 7 — The source’s proposed scientific correspondences

| Field | Where the structure is already there |
| --- | --- |
| Physics | Gauge freedom is A2 as law: the potential has no absolute value and only differences are physical. Relational quantum mechanics is A3 [4]. The intermediate-axis flip is A4 made visible, since the whole event lives in a sign while the magnitude never moves [14]. |
| Cosmology | There is no centre and no privileged frame. Distances are comoving, redshift is a relation, and the dipole in the microwave background is our own motion — an observer term that had to be declared and removed before the map meant anything. The horizon is a boundary in exactly the sense A3 requires: what can be known is set by what can reach us. |
| Biology | The membrane is constitutive, not packaging [6], and the Markov blanket says a system knows itself only across its boundary [5]. Descent with modification is machine-readable evolution stated in biology: each state carries its predecessor, and the lineage is the record. |
| Mathematics | No system certifies itself from inside [11]. Topology is the clean case of A2 — a structure that says what is near what while holding no origin and no distance at all. |
| Algebra | The axioms already carry both rules. The identity is 1, not 0, and every element has an inverse, so polarity is built in rather than added: a group is a set of things that each have a counterpart, defined by their relations and not by what they are made of. |
| Linguistics | A term never coincides with its referent [7, 8], which is A3 with meaning in the observer's place, and the reason two contradictory descriptions of one whole are both required rather than one being wrong (§3.2). |

Algebra deserves more than a row, because the number line by itself has done a great deal to shape how the subject is seen and used. Draw the line with zero at the middle and addition becomes the primary operation, negatives become quantities somehow less than nothing, and every magnitude is read as a displacement from a point that was chosen. The history shows the cost: negative numbers were resisted for centuries, not for any mathematical reason but because that picture made them absurd, and they became ordinary only once they were treated as directed quantities rather than as deficits.

Draw the line the way A2 requires and the subject reorganises. The origin unit is 1, the identity of multiplication; every element has an inverse, so a negative is a counterpart rather than a shortfall, which is A1 and A4 written into the group axioms; and what the line measures is ratio rather than displacement. Zero is not a point on that line at all — on a multiplicative scale it sits at infinity and is never reached, which is the same result physics gets when it cannot attain absolute zero and mathematics gets when no system grounds itself. This is not a reform of algebra. It is a note that algebra already works this way in its axioms while being pictured the other way on the board, and that the picture is what carries into how numbers are then used.

There is a second thing the subject does that matters more for this work, and it is a habit rather than an axiom. In algebra you write the whole expression and then reduce it, and the reduced form replaces what came before. That is what solving is, and for solving it is right. But reduction is a many-to-one map: many expressions collapse to the same result, so the result cannot say which one produced it. The steps are not hidden, they are gone. A system that reduces has, by construction, no lineage.

Keeping the polarity keeps the path. If every term stays a direction and a size, then how a value was reached is still legible in what it is made of, and nothing has to be reconstructed afterwards because nothing was discarded. The difference is the one between a balance and a ledger. A balance can always be computed from a ledger; a ledger can never be recovered from a balance. This is why machine-readable evolution is a representational requirement and not a logging feature — §7.2's ledger and this are the same rule, one stated for events and one for arithmetic.

The concrete case is −1 combined with −1. Ordinary arithmetic gives −2, and it is not wrong; it is a view. What it is not is a record, because −2 no longer says that two things contributed or which two they were, and A1 forbids treating the magnitude as the value. In the machine setup the two terms remain separately addressable and the sum is derived on demand rather than stored in their place. Both are available, and only one of them is recoverable from the other. Two terms that are both −1 therefore sit in the same field without either overriding the other or cancelling the other's contribution out of the record: they are two entries, not one total, and each keeps its own derivation.

This holds with particular force when cancellation is the very thing being studied. Cancellation is a relation between terms, so it can only be observed while both terms still exist. Store the residue in their place and the operation has already been performed, the object of study has been consumed by the act of recording it, and what remains cannot say whether it came from two large opposed quantities or two small ones, or from anything at all. The rule is therefore not that cancellation is disallowed. It is that cancellation is a view computed over entries that remain, never a write that replaces them — which is the balance and the ledger again, and the reason a system built this way can be asked how two things cancelled rather than only what was left. For a system whose purpose is to solve, reduction is the correct default; for a system whose purpose is to learn, it destroys the thing learning operates on.

In research that runs on algebra and formulae this is the more logical form outright, not a trade. There the expression is the object of study and the number is a by-product: what is compared between two treatments is the derivation, what is checked is where two paths diverge, and what is reused is an intermediate form somebody stopped at. Reduction discards precisely the material the work is done on. It also breaks the paper's own standard of evidence, and this is the decisive point. Two results that agree are evidence only if they were reached independently, and reduced results cannot be told apart on that question — the same number arrived at twice by one route and by two unrelated routes looks identical once the routes are gone. Keeping the derivation is what makes agreement checkable rather than merely observed, so a representation that retains it is doing what §4 requires and not merely recording more.

The consequence runs directly into §6. Independent evidence multiplies, so the natural line for it starts at 1 and has no zero; taking logarithms turns that line into an additive one, which is why bits add while confidence compounds. Those are not two facts. They are one line seen from its two origins, and choosing between them is the whole content of A2.

Set out together, the fields are linked in a way that is topological rather than thematic, and this is the paper's last claim. Every entry above describes what is inside what, and what can pass across a boundary. None of them needs a distance or an origin to say it. That is precisely what a topology is: structure without a metric. The four rules describe a topology and not a measurement scheme, which is why they survive translation between fields that share no units, no instruments and no vocabulary — a metric does not carry across a discipline boundary, and a relation does.

It also explains the failure mode in one sentence. A magnitude is a metric quantity and it is local to the box that scaled it; a sign, a boundary and a lineage are topological and they are not. Carry the first across a boundary and it degrades silently into a number that means nothing. Carry the second and it arrives intact. Every error catalogued in this paper — the zero used as an origin, the squared residual, the self-report read as a measurement, the valuation quoted out of its frame — is a metric quantity moved across a boundary it could not survive.

### Edition note: algebra, topology and retained derivation

The multiplicative-line discussion is the source's comparison. A group identity need not be denoted 1, and an inverse need not be a negative. Additive and multiplicative structures require their own domains. The source's “topology” language concerns connection, nesting, boundary and lineage; a formal topological equivalence would specify the space and the structure preserved by its maps.

The ledger/balance distinction supplies a concrete requirement: retained entries and derivation permit recomputation, while a final total generally does not determine its inputs. The conditional translative-loss proof and canonical ledger construction cited above formalise particular versions of that distinction. Cancellation remains a computed view over retained constituents, as the source explicitly states; it need not overwrite them. [E3 §§4–5; E4 §§3–4]

## 10. What this implies

The implications are stated as a list so that each can be attacked separately. Each is marked with the rule it follows from and with whether it is a consequence of the argument or a conjecture that could fail on its own.

### 10.1 Self-reported measurement is not weak evidence. It is none.

This reclassifies a large amount of ordinary practice: telemetry, self-assessment, internal audit, model-reported confidence. Not as dishonest, and not as noisy. As carrying zero warrant, however much of it is produced.

**Source basis:** A3, Prop. 1

**Source status:** Claimed consequence within the proposed axioms

### 10.2 Scale cannot substitute for environment.

If the gap between what a learner is and what it reads of itself is set by the independence around it, then more parameters and more authored data leave it exactly where it was. The present trajectory is bounded for a reason that has nothing to do with compute.

**Source basis:** A3

**Source status:** Claimed consequence within the proposed axioms

### 10.3 A benchmark score is a property of the frame, not the model.

Magnitudes are local to the box that scaled them, so a score does not travel. Two models are comparable only within one declared environment, and a leaderboard without that declaration is measuring agreement with itself.

**Source basis:** Prop. 2

**Source status:** Claimed consequence within the proposed axioms

### 10.4 Records must be ledgers, not balances.

Anywhere a result will later be checked, reduction destroys the material the check runs on. This is a requirement on representation in research computing, accounting, clinical data and training pipelines alike, and it is cheap to satisfy before the fact and impossible after.

**Source basis:** A1, A4

**Source status:** Claimed consequence within the proposed axioms

### 10.5 Some of what is filed as chaos is undeclared frame.

The T-handle shows an exactly solvable system behaving inexplicably when the reference is unstated. Where a problem is also chaotic the two are indistinguishable unless the frame is declared, so the nested formulation is a way of asking how much of a residual is which.

**Source basis:** §4.2

**Source status:** Conjecture

### 10.6 The next step in computation is representational.

The physics obeys all four rules and the layer we program it through obeys none. That is a gap in software, addressable now, on hardware that already exists — and testable in simulation before any quantum machine is involved.

**Source basis:** §8

**Source status:** Conjecture

The common shape is worth naming, because it is what makes the list one claim rather than six. In every case something was treated as a property of the thing when it was a property of the arrangement the thing was measured in. A score, a confidence, a valuation, an apparent instability: each read as belonging to the object, each in fact belonging to the frame, and each becoming tractable the moment the frame is declared and sized.

What the work implies for itself should close it. The applications above are not established by this paper. Its cross-science claim is that the six bodies of work exhibit a common structure whose independently justified agreement would be stronger than a single account. Everything else here is an experiment with a stated way to fail: the gap should narrow when independent references are added and not when capacity is; the drop should be larger for unauthored references than for authored ones; the frame term should be separable in a system known not to be chaotic. If those hold, the implications above are consequences. If they do not, the structure stands and the programme built on it does not, and it will be the independent references that say so rather than the author.

## References retained from the source

1. Maxwell, J. C. A Treatise on Electricity and Magnetism. Clarendon Press, 1873.

2. Whewell, W. The Philosophy of the Inductive Sciences, Founded upon their History. John W. Parker, 1840.

3. Wilson, E. O. Consilience: The Unity of Knowledge. Knopf, 1998.

4. Rovelli, C. Relational quantum mechanics. International Journal of Theoretical Physics 35, 1637–1678, 1996.

5. Friston, K. Life as we know it. Journal of the Royal Society Interface 10(86), 2013.

6. Maturana, H. and Varela, F. Autopoiesis and Cognition: The Realization of the Living. Reidel, 1980.

7. Frege, G. Über Sinn und Bedeutung. Zeitschrift für Philosophie und philosophische Kritik 100, 25–50, 1892.

8. Kripke, S. Naming and Necessity. Harvard University Press, 1980 (lectures delivered 1970).

9. Gibson, J. J. The Ecological Approach to Visual Perception. Houghton Mifflin, 1979.

10. Hoel, E., Albantakis, L. and Tononi, G. Quantifying causal emergence shows that macro can beat micro. PNAS 110(49), 19790–19795, 2013.

11. Gödel, K. Über formal unentscheidbare Sätze der Principia Mathematica und verwandter Systeme I. Monatshefte für Mathematik und Physik 38, 173–198, 1931.

12. Lee, T. D. and Yang, C. N. Statistical theory of equations of state and phase transitions II. Physical Review 87, 410–419, 1952.

13. Goodman, S. N. Toward evidence-based medical statistics 2: the Bayes factor. Annals of Internal Medicine 130(12), 1005–1013, 1999.

14. Euler, L. Theoria motus corporum solidorum seu rigidorum, 1765; instability of rotation about the intermediate principal axis. See also Ashbaugh, M. S., Chicone, C. C. and Cushman, R. H. The twisting tennis racket. Journal of Dynamics and Differential Equations 3(1), 67–85, 1991.

15. Poincaré, H. Sur le problème des trois corps et les équations de la dynamique. Acta Mathematica 13, 1–270, 1890.

Author statement. The research programme, the four constraints, the three-scale account of cognition mechanics, the observer theorem and its application, the staged enrichment programme of §7.5, the BRAINK architecture and the conjecture of §8 are the author's own work, developed independently and prior to the literature search reported here. The correspondences in Tables 2 and 5 were identified afterwards; in no case was a constraint derived from a cited source. Notation throughout is a chosen formalisation of that prior work, marked as such wherever the choice is load-bearing.

Declaration. All system figures in Section 7 are as reported and unverified pending the checks in Table 4. Citations in Tables 2 and 5 record structural correspondence only, and imply no endorsement by the cited authors. No funding, and no competing interests.

## Edition references and custody

E1. Aboudy Keddeh. Current author interpretation directions, retained in *Observation and Identity Review R3*, 17 October 2026; current at 18 October 2026. Literal .1, ordered observation roles and preserved context.

E2. Aboudy Keddeh. *Shared-Origin Research — Source Interpretation Corrections*, E1, 16 October 2026. [Corrected source](https://docs.google.com/document/d/1Zy6UFShKONxd6EHzq3afc3UgaTfhkILJAVeOg3oHhYM/edit). Transported arithmetic is a comparison construction.

E3. Keddeh Systems. *KEX-BRAINK: Mathematical Foundations and Proof Obligations*, v1.0, 13 July 2026, §§2, 4–8 and 11. [Source document](https://docs.google.com/document/d/1x3m7DwJbqSFMK4skvzf0pWUXauM-R_gX/edit). The cited definitions and proofs are used with their declared conditions.

E4. Keddeh Systems. *Native Algebra and Artifact Verification — Full Research Consolidation*, v4, 14 July 2026, §§2–5. [Source document](https://docs.google.com/document/d/1sz_cntJj0g1eqbSBrEGnC9gjjR1Jqp9p/edit). Typed-state, ledger, conditional-loss and phase-coordinate constructions.

E5. Bureau International des Poids et Mesures. *The International System of Units*, ninth edition, §2.3.1, “The kelvin.” [Official publication](https://www.bipm.org/en/publications/si-brochure). Current scale relation; reviewed in the accompanying relational architecture study.

E6. Harvard Natural Sciences Lecture Demonstrations. *Tennis Racquet Flip*. [Demonstration and physical explanation](https://sciencedemonstrations.fas.harvard.edu/presentations/tennis-racquet-flip). The explanation identifies intermediate-axis instability and growth of small perturbations; consulted 18 October 2026 alongside source reference 14.

E7. Paul E. Black and Andrew W. Lane. *Modeling quantum information systems*, NISTIR 7103b, 2004, §§1–2. [Official report](https://nvlpubs.nist.gov/nistpubs/Legacy/IR/nistir7103b.pdf). Symbolic and numerical quantum representation; consulted 18 October 2026.

E8. Aboudy Keddeh. *From Observation to Governed Computation*, R1, 18 October 2026. Relational architecture study, source register and primary-reference review. Used as the current interpretive and editorial guide.

Source identity: [September preprint](https://docs.google.com/document/d/11IjwDz_VPRWyA0samFCrOI7jLjnT-cwFH_HD2Z02izc/edit). Source access and distribution remain governed by existing permissions. Bibliographic inclusion does not broaden them.

The original text, byte digest, native re-read, structural coverage and editorial changes accompany this edition. The source text's content is unchanged in its custody carrier. Edition scope notes are distinct from the author’s axioms and historical claims.