The Conservation of Faithfulness

How to read this

This is a maturation pass over an earlier working paper. It does three things the prior draft did not. It states the conservation law formally, with an explicit scope condition and one named falsifier you can run. It separates the composition claim into the part that is provable at each layer and the part that is still a conjecture, instead of asserting both at once. And it audits the already-shipped claims against the real repositories, with one chain shown end to end (source, then test, then coverage) and the maturity gap stated plainly.

The claim labels are load-bearing. [established] is earned by runnable evidence. [designed] is built or specified but not yet demonstrated at the claimed strength. [reach] is a reasonable extension the experiments do not demonstrate. Where a claim could not be certified against a runnable artifact, it is marked UNVERIFIABLE and the reason is given, in the same discipline the principle itself demands. Read the labels.

Abstract

Information that crosses a boundary, between two minds, between two substrates, between two senses, is not conserved as bits. Almost all of the bits can be discarded. What can be conserved is faithfulness to a criterion: the recoverability of a specific readout of the subject. This paper states that as a conservation law with a precise scope condition. It tests the law on seven sensory substrates (sight, sound, shape, language, structure, quantity, identity), exercised as ten distinct criterion-readout instances (several substrates carry two instruments each), plus two structural tests (composition and an adversarial boundary probe). The counting is made re-derivable below, so a reader can confirm it from the table rather than take it on faith.

The law has a single named falsifier that would break it if it were false. It has exactly one boundary, and it is the important one: the law conserves faithfulness to the stated criterion and cannot certify that the stated criterion is the right one. That boundary is not a gap to engineer away. It is the permanent place a human stands, and it is the same fact that makes one shared center able to serve every domain at once. The paper also corrects two over-reaches from the earlier draft: the composition claim is split into a provable layer law and a separate open conjecture, and the shipped-system claims are re-counted against the actual repositories, where one number had drifted.

1. The question

Two minds cannot share their internal states. Each can only collapse a state into a stream the other reconstructs. The naive picture is a narrow tube: a bottleneck through which too little passes. The central question is whether the tube is really the limit, or whether the thing that has to cross is not the state at all but something far smaller and substrate-independent that the tube carries easily.

If the second reading is right, then shared understanding is not a bandwidth problem. It is a problem of what to conserve, and how to check it crossed faithfully. The thing to conserve is faithfulness to a named criterion. The way to check it is an external witness that holds that criterion. The law that ties them together is stated next.

2. The conservation law, stated formally [established]

The earlier draft gave definitions and five propositions but never wrote the law as a single statement with its scope condition and its falsifier in one place. Here it is.

2.1 The objects

• Definition 1 (transform). A substrate transform is a function that maps a subject to a representation. It may be lossy (the output has far fewer dimensions than the input) or bijective (the output has the same cardinality as the input).
• Definition 2 (criterion). A criterion is a readout: the property of the subject one cares to preserve, such as a label, an identity, a measurement, or an answer. Criteria are not authored by the transform. This externality is not decoration. It is load-bearing, and it is where the boundary in section 6 bites.
• Definition 3 (faithfulness, graded). A transform is faithful to a criterion at level one minus epsilon if there exists a recovery map whose readout error is at most epsilon over the relevant distribution of subjects, for a stated error metric and a stated distribution. Binary faithfulness is the special case where epsilon is zero. The recovery may be required to use an external secret key; that is the concealment case.

Three knobs must be named for the level to mean anything: the error metric, the distribution over subjects, and, if any, the secret key. A faithfulness number with these unstated is not a measurement. It is a mood.

2.2 The law

Conservation of Faithfulness. Fix a subject distribution, a criterion, an error metric, and a tolerance. If there exists a recovery map whose expected readout error is at most that tolerance, then the transform is faithful to the criterion at the corresponding level, and this fact is independent of the bit-rate of the transform. A transform may discard arbitrarily many bits of the subject and remain faithful, because faithfulness constrains the recoverability of the criterion, not the cardinality of the output.

Three corollaries hold under the same fixed setup:

• C1, bit-independence. Faithfulness does not track bit-loss. A transform can drop almost every bit and stay faithful; another can be bijective (drop no bits) and be unfaithful to the criterion without the external secret. That is the concealment case.
• C2, internal blindness. No statistic computed from the representation alone determines faithfulness to the criterion, because the criterion is external to the representation. Certification requires evaluating against the criterion itself, off-substrate.
• C3, criterion-relativity. Faithfulness is a property of the pair (transform, criterion), not of the transform. The same transform can be faithful to one criterion and lossy to another with nothing about the transform changed.

2.3 The scope condition

The law is only meaningful inside a declared scope. It is a precondition, not a footnote. The law holds relative to a fixed, externally-named distribution, criterion, metric, and tolerance. It says nothing about an un-named criterion, an un-stated distribution, or a transform evaluated against a criterion it was allowed to choose. If any of those four is authored by the transform or by the party operating it, the law does not apply and the result is UNVERIFIABLE by construction, not by accident. The law is sound because it refuses to speak outside its scope, not in spite of it.

2.4 The named falsifier, one and runnable

A conservation law you cannot break is not a law; it is a slogan. Here is the single experiment that would falsify the core claim.

Falsifier F0 (bit-rate predicts faithfulness). Construct a family of transforms indexed by output bit-rate, all evaluated against one fixed external criterion on one fixed distribution. The law is false if faithfulness is a monotone function of bit-rate across the family, that is, if lower bit-rate always means lower faithfulness and there is no transform that is simultaneously low-rate-high-faithfulness or high-rate-low-faithfulness. Concretely: if you cannot exhibit both a heavily lossy transform that stays faithful and a bijective transform that is unfaithful without an external key, on the same fixed setup, the central claim has failed.

The seven sensory substrates below are exactly attempts to trigger F0 across mechanically independent instruments. F0 stayed untriggered in every one: identity survived a 24,576-fold hash (lossy but faithful), and encryption was bijective yet unreadable without the key (bit-preserving but unfaithful). One clean trigger of F0 on any fixed setup falsifies the law. None occurred. That is the status: robust by independent replication, falsifier untriggered, not a theorem.

A second, sharper falsifier exists for corollary C2. Falsifier F1 (an internal statistic certifies faithfulness). C2 is false if there exists a statistic of the representation alone that reliably predicts faithfulness across transforms. If a single within-substrate number (residual energy, agreement, confidence, quantization error) tracked faithfulness with no false positives over an adversarial transform set, then no external witness would be needed and C2 would fail. The adversarial structural test (section 6) is the attempt to fire F1. It did not fire, but it located the precise edge where a proxy criterion can be gamed, which is a weaker and honest version of the same pressure.

2.5 Why the propositions are near-immediate, and what is not

Corollaries C1 to C3 fall almost directly out of Definition 3. That is on purpose: the framework is chosen so that what to conserve and why internal certification fails are consequences, not surprises. The empirical contribution is not the corollaries. It is that real transforms across every sensory substrate instantiate them, and that two phenomena the definitions do not force still appeared in the data: the concealment versus destruction split and graded faithfulness. Those two are findings, not theorems, and they are the parts a reviewer should push on hardest.

3. The empirical core [established]

3.1 Method

All simulations use the Python standard library (plus PIL for images), deterministic and seeded. Each substrate pre-registers its falsifier: the condition under which the claim would fail for that substrate, an instance of F0 or F1. The language and vision substrates use blind, fresh-context model subjects and arm-blind judges to avoid contamination. Code and per-substrate verdicts live in the substrate files and the principle document.

One process note, reported because it is the difference between a result and a rationalization. The pass/fail instruments mis-fired four times: a block-shuffle sized to the hash cell (perceptual hash), an internal-blindness test that compared two transforms instead of one transform across two criteria (quantize), a reused-keystream cipher that leaked plaintext, which is a genuine two-time-pad bug (encryption), and an over-strict absolute-area threshold for a spiky shape (geometry). In each of the four, fixing the measurement (not the claim) restored the pre-registered result. The status of this note is honest but not independently certified in this pass: it is taken from the build log and the current state of the substrate files, not from a cited commit diff showing the edit landed on the test and not on the claim. Its falsification condition is concrete and left as a next task: pull the four commits, and if any one of them moved the asserted threshold, label, or criterion rather than the measurement code, this note is DRIFT and must be rewritten. Until that diff is checked, read this as reported, UNVERIFIABLE against a commit, not established.

3.2 The seven sensory substrates, and how to count them

No shared math runs between any two rows, so agreement across them is independent replication, not one result re-skinned. Each row is an attempt to trigger F0 and a record that it did not.

Sight (perceptual hash)

Image to 64 bits, roughly 24,576-fold loss. Scene identity survives (drift 0.004); a different scene does not (0.484).

Sound (bandlimited audio)

Roughly 98% of spectral energy cut. Pitch and melody survive 4 of 4 notes; timbre, which is about 2% kept, does not. This is the phone-call effect.

Shape (polygon decimation)

To 10% of vertices. Gross area survives (4 to 6% error); 83% of fine corners are destroyed.

Language (lossy summary, model-read)

Roughly half of facts dropped. A question is answerable if and only if its supporting fact survived, with zero hallucination.

Structure (graph to spanning forest)

59% of edges dropped. Connectivity survives (1.000); distance (0.726) and triangle counts (0.492) do not.

Quantity (projection / quantization)

From 48 dimensions to 6, or down to 1 bit. The invariant survives to 1 bit when the transform keeps the criterion direction (faithful 1.00 versus unfaithful 0.46 at equal bits).

Identity (encryption)

Bijective, zero bit-loss. The criterion is at chance from ciphertext (0.510); the external key recovers it (1.000).

How the numbers reconcile, stated once and re-derivably. This paper uses three counts, and they are not in conflict once you see what each one counts.

• Seven sensory substrates: the seven rows above. This is the breadth of the replication.
• Ten criterion-readout instances: several substrates carry two distinct instruments or readouts, so the number of independent faithfulness measurements is larger than seven. In the canonical principle table, sight runs a perceptual hash and a vision arm (2), sound runs a bandlimit filter and an analog filter (2), quantity runs a linear projection and a one-bit quantization (2), and shape, language, structure, and identity run one each (4). That is 2 + 2 + 2 + 1 + 1 + 1 + 1 = ten readout instances. This is the count the abstract and conclusion mean by "ten."
• Two structural tests: beyond the sensory rows, two further substrate files probe structure rather than a sense: one for composition (section 4) and one for the adversarial boundary probe (section 6). They are tests of the law's shape, not of an additional sense, and are reported as such.

So on disk there are nine substrate files: seven sensory plus two structural. The seven-versus-ten distinction is breadth (seven substrates) versus measurement depth (ten readout instances). Wherever this paper says "ten," it means the ten readout instances; wherever it says "seven," it means the seven sensory substrates. If a reader counts the table and finds anything other than seven rows, or counts the canonical instrument column and finds anything other than ten readouts, that is a DRIFT to fix here, the same way the audit below fixes a stale organ count.

3.3 The two refinements the substrates forced

Loss is two things, not one. Destruction: the bits that determine the criterion are gone, and no key helps. Concealment: encryption discards nothing (it is bijective) yet the criterion is unreadable without an external secret. Information is conserved but scrambled, recoverable only through the keyed recovery. This is the modern reading of the black-hole information question made concrete: not erased, only put beyond reach. The model result is [established]; the physics analogy to the Page curve is [reach] and is labeled as analogy, not result, throughout.

Faithfulness is graded, not binary. In noiseless discrete substrates the criterion either factored through the transform or did not. In analog and noisy ones it survives smoothly down to a noise floor, bounded by signal-to-noise ratio. Definition 3 carries this as the epsilon knob; the substrates are where it earned its place.

4. Composition: what is provable, and what is still a conjecture [split, see labels]

The earlier draft asserted both "a pipeline is faithful if and only if every stage is" and "faithfulness composes across every layer of the architecture" as if they were the same established fact. They are not. One is provable at each layer under a stated criterion. The other is a conjecture that the visualization experiments have already partly falsified in their own domain. Honesty requires splitting them.

4.1 The layer composition law (provable, scoped) [established under its scope condition]

Layer Composition Law. Take a pipeline that is a chain of stages, and fix one criterion. Suppose a stagewise criterion chain is given such that each stage is faithful to the criterion handed to it and recovers into the criterion handed to the next stage. In other words, the criterion is carried forward stage by stage. Then:

• Binary case. If every stage is faithful (each stage error is zero) then the whole pipeline is faithful. If any one stage is non-preserving for its carried criterion, the whole pipeline is unfaithful. Faithfulness is the conjunction over stages: the absorbing meet. One broken stage collapses the whole.
• Graded case. For noisy stages the end-to-end error is bounded by the accumulation of the stage errors. In the worst case (correlated errors) the bound is the linear sum of the stage errors, from the triangle inequality. Under an additional independence assumption the standard error-propagation rate grows as the square root of the chain length rather than linearly, but that rate is a property of the independence assumption, not a measured result here.

Why it is provable and not just asserted. The proof is immediate from Definition 3 once the stagewise criterion chain exists: faithfulness of each stage to its incoming criterion gives a recovery map, and composing the recovery maps recovers the final criterion from the pipeline output with error bounded by the triangle inequality over the stage errors. That triangle bound is the linear worst case, and it is what the proof delivers. The binary collapse is the zero-error corner. This much is shown empirically in the composition substrate: a faithful-but-noisy chain degrades gracefully, and one unfaithful stage collapses the pipeline (0.955 to 0.501).

What is not established here, said plainly. The square-root-of-length rate in the graded case is the textbook independent-error-accumulation result, not something the composition substrate measures. The shipped simulation shows the binary collapse and graceful degradation, not a fitted square-root scaling curve over chain length. Status of the square-root rate claim: UNVERIFIABLE against the shipped artifact. Its falsification condition is a one-experiment check left for the next pass: build chains of length 1, 2, 4, 8, and so on with independent stage noise of fixed per-stage error, measure end-to-end error, and if it grows linearly in the length (or faster) rather than as the square root, the independence assumption did not hold for these substrates and the rate claim is DRIFT. Until that curve is run, only the linear triangle bound and the binary collapse are [established]; the square-root refinement is [designed, UNVERIFIABLE].

The scope condition that makes it true (and that the earlier draft hid). The law holds only when the stagewise criterion chain exists and is carried forward. It is not the claim "if each part is faithful to the final criterion, the whole is." That stronger claim is false, and the next subsection is the counterexample. The load-bearing object is the chain, not the single final criterion. Where someone can produce the chain, composition is a theorem. Where they cannot, it is a conjecture, and it can fail.

Falsifier F2 (composition without a carried chain). The Layer Composition Law is mis-applied, and therefore false in the form people want, if there is a pipeline where every stage is faithful to the final criterion in isolation yet the composed pipeline is not faithful. Such pipelines exist, shown next, so the law must be stated with the carried chain or not at all.

4.2 The whole-from-parts conjecture (open, partly falsified) [conjecture; counterexample on record]

The tempting stronger statement, "if every sub-part of a subject is rendered faithfully under a criterion, the whole is faithful under that criterion," is a conjecture, and in the one domain where it was tested adversarially it failed.

In the visualization program, a 3-cube under orthographic projection with identity rotation keeps every edge as a locally faithful straight segment, yet collapses four vertex pairs to the same 2D point, so the global vertex-and-edge graph is unrecoverable. Local fidelity did not entail global fidelity. The degeneracy is not rare: at dimension 4, 9.1% of views collide; at dimension 6, 69.1%. The fix was not to assume composition but to add a scope condition to the criterion: require the projection to be injection-preserving (no two distinct points collapse), enforced by a genericity check the renderer now runs. After enforcement the worst case went from 65.6% degenerate views to 0% residual.

So the honest statement is two-part.

• Provable: composition holds at each layer when a carried criterion chain exists (subsection 4.1).
• Conjecture, with a counterexample: parts-faithful-implies-whole-faithful does not hold in general. It holds only when the criterion is strengthened to forbid the collapse modes (injection-preserving projection, generic position, and, for global topological invariants, at least three orthogonal views). Whether a single scope strengthening makes whole-from-parts hold across all domains is UNVERIFIABLE from the experiments run so far. The visualization arc found a different strengthening for each failure mode (genericity for local collapse, three orthogonal views for global topology, coherence-weighted fusion for cross-channel). No experiment shows these reduce to one condition. Stated as one law it would be an over-claim.

This split is the correction. The earlier draft's "faithfulness composes" is true as subsection 4.1 and false as the stronger reading, and a working paper should not let the reader carry away the false one.

5. The neutral center [designed, grounded in section 3]

Here is the bridge from the experiment to what is being built. Because the criterion-relevant invariant is substrate-independent (the same criterion survives translation into sight, sound, shape, or language), two minds with different senses of perception can hold the same subject. The invariant is not in either mind and not in the channel. It is the readout both can recover and check.

So the move is not to widen the tube. It is to build a neutral center: a shared, perceptible form of the subject that both parties can render into their own senses, change, and witness, so the channel carries only deltas against a common referent and never whole state. In that center a subject is driven toward its telos: the form that is faithful to the criterion, the singular point of quality. Quality here is not taste. It is faithfulness made checkable. The reconcile (section 7) is the dynamics of the center; the certificate is its receipt; the shipped organs are its instances in each modality.

5.1 Corollary: quality is relational [established core, scoped]

If "best" means the telos, the criterion-faithful form, then because the criterion is necessarily external (section 6), "best" is not a property a subject can hold or certify alone. It cannot author its own criterion; it cannot certify from inside that it has reached it (internal confidence is blind, corollary C2); and a reconciliation across independent perspectives can reach a fixed point no single perspective holds. So a subject arrives at its best through being perceived and reconciled by another, with the experience shared. Quality is conferred in the meeting, not declared from within. The earlier program asserted this; here it is derived from the law.

Two honest bounds keep it from overreaching. First, "another" is exactly "an external criterion-holder." For a quality judged by minds (clarity, correctness-as-understood, beauty, meaning) that holder is another perceiving mind. For a telos fixed by an impersonal criterion (a physical law, a spec) the external thing may be impersonal: the subject still cannot self-certify, but it need not be perceived to be faithful, only checked. Second, shared perception is necessary, not sufficient: two minds can agree on a form faithful to a shared proxy criterion (section 6), so the meeting makes the best reachable and knowable without guaranteeing it. The criterion still has to be the right one.

5.2 Live demonstration: the center, inhabited [designed, shown, with the boundary]

The center was run once. A subject (propose the best new flagship from two or more projects) was reconciled by two minds with genuinely different senses: a visual mind perceiving only the relational shape (topology and roles, not function) and a symbolic mind perceiving only the descriptions (function, not shape). Solo, they diverged and each was half-right. At the meeting each corrected the other's blind spot and they converged on one reconciled result.

Judged blind by two external models, the same three artifacts ranked oppositely under two criteria:

Novelty-weighted

Visual solo 4.75 · Symbolic solo 4.75 · The meeting 4.0

Correctness / buildability

Visual solo 2.0 · Symbolic solo 2.5 · The meeting 5.0

Three things at once. The mechanism works (different senses combined, errors corrected, convergence). The corollary holds (under a criterion valuing wholeness, the meeting at 5.0 strictly beats either mind alone, and each solo's deficit is exactly its missing sense). And the section 6 boundary appears live: the ranking flips with the criterion, so "best" is criterion-relative and choosing the criterion is the human's seat.

The limits are real and stated: two model minds stand in for human-and-model, the sample is one subject, and the reconciliation introduced one ungrounded over-reach (meetings can over-build, not only complete). This is a single demonstration, not a study. It is UNVERIFIABLE at the strength of "the center works for human-and-model in deployment": that run has not happened, and the prior draft already conceded it. Nothing in this maturation pass changes that status.

6. Where it expires: the one boundary [established]

Driven to exhaustion, the law crosses every sense and every pipeline and stops at exactly one place. The adversarial structural test is the attempt to fire falsifier F1, and it locates the edge precisely. When the witness judges with a proxy criterion (the witness criterion aligns only partway with the true one, a cosine of 0.61, as any real witness does), a transform can move along the part of the true criterion orthogonal to the witness, reading as fully faithful to the witness while a large fraction of true labels silently flip. The external witness is necessary but not sufficient: it can be gamed if its criterion is a proxy for the intended one.

So the law is intact but scoped exactly as the scope condition says: it conserves faithfulness to the stated criterion and cannot certify the stated criterion equals the intended one. This is not a defect to engineer away. It is the same residue every layer returns to: a system cannot validate its own criterion from inside. That boundary is where accountability begins. "Which criterion is the right one" is an irreducibly external, human question, and the discipline of naming the criterion, externalizing it, and standing behind it is what this program has been calling accountability all along.

6.1 The boundary is the universality

Turned over, the boundary is the most valuable result, not the limit. If quality were criterion-absolute, each domain would need its own machine. Because quality is faithfulness to a named criterion (corollary C3), exactly one center suffices: substrate-neutral and criterion-neutral, it holds the criterion externally and lets each domain bring its own and weight it as that domain demands. The same fact that bounds the law (it cannot pick the criterion from inside) is what makes the center universal (it does not have to; it hosts whichever is named). The packages were never different tools. They are one center bound to different criteria: security to an origin criterion, novelty to a corpus criterion, correctness to a spec criterion, aesthetics to a fitness criterion.

One distinction keeps this from collapsing, and it is load-bearing. The center is neutral; an act within it is not. Hosting every criterion is not "anything goes." Within a named criterion, faithfulness is objective, checkable, and still gameable by a proxy. To stand behind all of them is to welcome every criterion and require each to be named and owned: pluralism, not relativism. The neutrality is the place's; the accountability is the person's, per engagement.

7. The reconcile, and an honest audit of the shipped claims [established, with one DRIFT]

The earlier draft made several "shipped" claims in passing. A working paper that preaches proof-before-trust must hold its own claims to the same standard. Below is one chain shown end to end, then a re-count of the headline numbers against the actual repositories, with the gaps named.

7.1 The reconcile, in one sentence

The system reduces to one operation, the reconcile: perceive any artifact into a witnessed form; judge that form against a criterion it did not author; carry a re-checkable certificate of the judgment; return UNVERIFIABLE when you cannot. The shipped verifier organs are instances of it.

7.2 One chain shown: source, then test, then coverage [established, MATCH]

I audited the reconcile spine in the public coherence-membrane repository firsthand rather than citing it from memory.

• Source. The reconcile module exists and is substantive: 57 executable statements. It implements a reconcile function that is fail-closed (any exception in perceiving or judging yields an UNVERIFIABLE observation; it never raises), and it implements an independence tri-state (witnessed-independent, self-authored, unwitnessed). The "criterion it did not author" property, asserted by discipline in the prose, is recorded as a checkable field on the observation, with an opt-in strict mode that downgrades a self-graded decision to UNVERIFIABLE and a stronger require-independent mode that refuses any decision whose independence was not positively witnessed. The doctrine and the code agree.
• Test. The reconcile test file exists with 11 tests. Run on 2026-06-30, all 11 pass. They cover the verifier-organ equivalence (the organ is a reconcile, same verdict and oracle), refute and witness with a real SHA-256 digest, separate perceive and criterion, and the fail-closed path.
• Coverage. Running the same slice with coverage reports 88% line coverage of the reconcile module (7 of 57 statements uncovered). The uncovered lines are named: the safe-string exception fallback, the strict-mode self-authored downgrade branch, and the witnessed-independent exception path.

Verdict: MATCH. The reconcile is real, tested, and the core verdict-passthrough and fail-closed behavior are exercised. Maturity gap, stated plainly: the two strongest guards in the file, strict mode and require-independent mode, are exactly the under-covered lines. The default passthrough path is well tested; the opt-in anti-laundering downgrades are not fully exercised by this slice. The most security-relevant code is the least covered. That is the honest gap, and it is a concrete next task: add tests that drive a self-authored criterion through strict mode and an unwitnessed criterion through require-independent mode, and assert the downgrade to UNVERIFIABLE and the recorded reason.

7.3 Re-count of the headline numbers [one MATCH, one DRIFT, one UNVERIFIABLE]

• "Fifteen shipped organs" turns to DRIFT. The repository's organs directory contains 17 organ modules today. The prose number "fifteen" is stale relative to the current tree. This is a small drift, but it is exactly the kind of un-recounted number a proof-before-trust paper should catch. The fix in this version: say "the shipped verifier and modality organs (17 modules as of 2026-06-30)" and let the count be re-derivable, not asserted.
• "emet emits MATCH / DRIFT / UNVERIFIABLE over a transform" is MATCH on the vocabulary, partial on the round-trip claim. The public emet repository's verdict logic does emit the three-valued vocabulary, and its coverage map ties each check to MITRE ATLAS and OWASP-LLM identifiers. So "emet witnesses MATCH and DRIFT" is a match. The earlier draft's stronger open-program item, "a round-trip witness that measures invariant-survival through a transform, not identity," is a designed next step, not a shipped one, and remains UNVERIFIABLE at the strength the open-program list implied. I did not find a shipped emet check that measures criterion survival across a non-identity transform; emet's shipped checks are anchored to byte and view fidelity. Labeled accordingly.
• "157 tests" (visualization substrate) is UNVERIFIABLE in this pass. The visualization thesis cites 157 tests across its JavaScript organs. I did not re-run those suites in this maturation pass (they live outside the three Python repositories I audited), so I do not certify the number. It is carried as the source claims it, labeled UNVERIFIABLE here, with the honest note that re-running it is a one-command check left for the next pass.

7.4 What this audit establishes and what it does not

It establishes that the central object, the reconcile, is not vaporware: it is a small, real, tested module whose documented behavior matches its code, with a named coverage gap on its most important guards. It does not establish that the whole 15-to-17-organ ecosystem, the emet round-trip witness, or the visualization substrate are all at the same maturity. They are not, and the labels above say so. The discipline the paper argues for (perceive the artifact, check it against a criterion it did not author, carry a re-checkable result, say UNVERIFIABLE when you cannot) is the discipline this section applied to the paper's own claims. One number drifted; one chain matched cleanly with a named gap; one remains unverified. That is the receipt.

8. The ethical corollary: what you put in lands somewhere [reach, with a grounded core]

A parallel thought, that positive and negative energy put into the world lands and has an effect, has a grounded core in the commons results and a reach beyond them. The grounded core, with its source named: in the aperture and commons simulations (the same runs labeled the wrong-attractor result) a contribution to a shared center is never neutral. It shifts the fixed point everyone settles to, and the shift is not proportional to how right the contribution is but to how loud and how coordinated it is. The wrong-attractor run is the concrete instance: a tight, confident consensus formed around the wrong attractor, with confidence tracking coordination rather than truth. A shallow, loud, aligned voice moves the center more than a deep, quiet, sincere one. [established in model, sourced to the aperture and commons run; falsification condition: if re-running those simulations shows the settled fixed point tracks contribution correctness rather than loudness and alignment, this claim is DRIFT.] That is a precise, unsentimental version of "words carry energy": in any commons, what you emit propagates and lands, weighted by force and alignment, not by truth, which is exactly why a center needs a gate that caps how loud any one voice can be at the seam. The reach: extending this from the simulated commons to human action in the world is an analogy, not a result. It is a reasonable ethic; it is not something these experiments demonstrate, and the paper does not pretend otherwise. [reach.]

9. Related work [moderate confidence]

Since this program began, the field has appeared to converge toward the same spine from several directions: proof-carrying verification motivated formally by incompleteness arguments; demonstrations that ungrounded self-critique fails; creative render-then-critique loops that collapse without a grounded external critic; provenance work fragmenting for lack of a composition-sound criterion. No one, to my reading, has assembled the cross-domain claim that faithfulness is the conserved quantity across substrates and senses, and the criterion must be external. That assembly is the contribution; the individual pieces are not all novel.

The primary citations for the frontier-convergence claims are summarized from an internal H1-2026 landscape survey that this paper does not reproduce, and are UNVERIFIABLE here as primary sources. Their falsification condition is explicit: each of the four convergence items is a claim about a specific external literature, and a citation pass that fails to locate a primary source for any one item turns that item from "claimed convergence" to "unsupported," and it must be struck. They are listed as claims pending a citation pass, not as sourced facts. This honesty about section 9 is itself an instance of the section 6 boundary.

10. The horizon, and a fenced coda

What is earned. A falsifiable conservation law with a stated scope condition and one runnable falsifier (section 2), replicated across seven sensory substrates exercised as ten criterion-readout instances (section 3); a layer composition law that is provable under its carried-chain scope condition, cleanly separated from the whole-from-parts conjecture that its own sibling experiments partly falsified (section 4); a boundary that locates accountability precisely (section 6); and one shipped-claim chain audited end to end with its gap named (section 7).

What remains. The demonstration with a human in a perceptual seat (here two model minds stood in). A full rate-distortion and Landauer derivation of the reversible-to-conservation hinge (the propositions give the structure, not yet the thermodynamic bound). The square-root composition-rate curve (section 4.1), still UNVERIFIABLE against the shipped simulation. A cited commit diff confirming the four method-note instrument fixes landed on the tests and not the claims. Primary citations for section 9. Tests that close the coverage gap on the strict and require-independent guards. A single answer to whether the section 4.2 scope strengthenings reduce to one condition, which the experiments so far leave UNVERIFIABLE.

Coda, strictly [reach], offered as a question, not a claim. We ask where thoughts come from, whether from some lossy, imperceptible place that has always been around us, that is us and that we are of. The law gives this intuition a vocabulary it did not have: concealment shows a substrate can hold everything and still present almost nothing to a given criterion, so the imperceptible-but-conserved is a real category, not a mystical one. A mind, on this view, would be a neutral center of the kind being architected: a place where signals from a vast substrate we do not perceive are reconciled into the small, checkable forms we call thoughts. But the experiments do not establish this and could not. They establish how information crosses and is checked, not where it originates. The honest position is that the law fits the intuition without confirming it, and that fit is itself an instance of the section 6 boundary: an elegant criterion the work cannot certify from inside. The paper names it and leaves the certifying to a place outside the paper.

Conclusion

Between two minds, the thing that crosses is not state and not bits. It is faithfulness to a criterion, and it is substrate-independent enough that minds with different senses can meet over it. Stated as a law with its scope condition and its falsifier, the claim survived every attempt to break it: seven sensory substrates, exercised as ten criterion-readout instances, plus two structural tests, and the falsifier stayed untriggered across all of them. Stated honestly, its composition holds at each layer under a carried criterion chain and fails as a naive whole-from-parts claim, which is a correction this draft makes explicit. Build the minds a neutral center where the subject is made perceptible to both and checked against an unauthored criterion, and the subject can be reconciled toward its telos: its faithful, quality form. The one thing the center cannot do for them is choose the criterion. That choice, and the standing-behind-it, is the human's. It is what accountability is, and it is the reason the work has a person at its center and not only a machine.

Provenance and labels

Evidence (this program): the principle document (substrate results and the seven-substrate, ten-instance table); the aperture and commons simulation records (including the wrong-attractor result cited in section 8); the vision-arm report; nine substrate files (seven sensory substrates plus the composition and adversarial structural tests); the live-center demonstration; the reconcile spine in coherence-membrane; the visualization thesis (the composition counterexample and its fix).

Audited firsthand on 2026-06-30 (section 7): the coherence-membrane reconcile module (57 statements), its test file (11 tests, all pass), reconcile coverage 88% with named gaps; the coherence-membrane organs directory (17 organ modules, correcting "fifteen"); emet's verdict logic and coverage map (MATCH, DRIFT, UNVERIFIABLE vocabulary present, round-trip-over-transform witness not found as shipped).

Carried but UNVERIFIABLE in this pass (falsification conditions stated in line): the square-root composition rate (section 4.1, not measured by the composition substrate); the four method-note instrument fixes being test-side (not certified against a commit diff); the section 9 frontier-convergence items (no primary citations reproduced here); the "157 tests" visualization count (section 7.3, not re-run).

Label key: [established] earned by runnable evidence; [designed] built or specified but not yet demonstrated at the claimed strength; [reach] a reasonable extension the experiments do not demonstrate; MATCH / DRIFT / UNVERIFIABLE the audited status of a shipped claim against the real repository. The labels are load-bearing. Read them.

References

Foundational works, well-established; exact bibliographic details to be finalized in copy-edit.

• Shannon, C. E. (1948). A Mathematical Theory of Communication. Bell System Technical Journal. Channel capacity; the basis of rate-distortion: preserve the relevant information, not the bits.
• Landauer, R. (1961). Irreversibility and Heat Generation in the Computing Process. IBM J. Res. Dev. Erasing a bit costs at least kT ln 2 (the floor that makes destruction dissipative).
• Bennett, C. H. (1973). Logical Reversibility of Computation. IBM J. Res. Dev. Reversible computation approaches zero dissipation (transforming is not erasing; the basis of concealment).
• Bateson, G. (1972). Steps to an Ecology of Mind. Information as "the difference that makes a difference"; a criterion-relative readout, not a substrate property.
• Bekenstein, J. D. (1981). Universal Upper Bound on the Entropy-to-Energy Ratio. Phys. Rev. D. Maximal information density (the black hole as the limit case).
• Page, D. N. (1993). Information in Black Hole Radiation. Phys. Rev. Lett. The Page curve: information conserved and scrambled, not destroyed (the physics of concealment).
• Lloyd, S. (2000). Ultimate Physical Limits to Computation. Nature 406. Energy bounds on operations and memory.
• Cover, T. M., and Thomas, J. A. Elements of Information Theory. Rate-distortion, sufficient statistics.
• Ashby, W. R. (1956). An Introduction to Cybernetics. The Law of Requisite Variety (the quantitative shadow of the criterion bound; load-bearing in the visualization sibling thesis).

Frontier convergence: claims summarized from an internal H1-2026 landscape survey; primary citations pending and marked UNVERIFIABLE in section 9, with the per-item falsification condition stated there. They cover proof-carrying verification motivated by incompleteness arguments; the empirical failure of ungrounded self-critique; the collapse of render-then-critique creative loops without a grounded external critic; and the fragmentation of provenance for lack of a composition-sound criterion.