The Demarcation Criterion

When a Two-Pole Loop Is Epistemically Non-Trivial

Alex Deva — May 2026 · revised June and September 2026

Builds on: The Veer The Asymmetric Synergy Bound Φ_loop as Synergy The Adversarial Sensor

0. Epistemological Note

This appendix answers a falsification threat, not a metaphysical one. The framework’s third Kill Condition (The Adversarial Sensor, §5) reads: if the adjunction $I \dashv S$ can be applied to any two interacting systems — a rock hitting a wall as readily as a sensor questioning an instrument — then the framework is trivially true and explains nothing. A theory that applies to everything distinguishes nothing. If “the loop” is just a fancy name for “two things in contact,” the framework has no content.

What follows is a proposed necessary criterion for telling the framework’s loop apart from a generic interacting pair. It is a definition together with supporting arguments and a worked table, not a proven theorem — though one of its two conditions rests on a short lemma that holds for any non-negative information decomposition, and the other on the data-processing inequality. It is a claim about necessity: it states two conditions a coupled system must satisfy to carry synergistic information about a target — the informational precondition of recognition — and shows that the trivial counterexamples the Kill Condition raises fail at least one of them. It does not claim the two conditions are jointly sufficient for recognition, and it is not a consciousness detector (§6, and pointedly §7).

Revision note (September 2026). This criterion has been corrected twice, by two stress tests run in parallel in June 2026 and reconciled here. Its first form (May 2026) stated grounding per pole ($\exists X: \mathrm{TE}_{T \to X} > 0$) and made difference in kind ($\alpha > 0$) its second necessary condition. The exact Gaussian test (research/gaussian-loop-model/) showed that two same-kind sensors with independent noise carry synergy — 0.293 bits of pooling — so $\alpha > 0$ is not necessary for synergy; the June revision retreated to claiming $\alpha$ for synergy beyond pooling. A second, combinatorial test (research/demarcation-criterion/) broke both conditions further. The masked-measurement coupling carries one full bit of purely synergistic information about the territory while both per-pole channels read zero, and its symmetric variant does the same with exchangeable poles — so not even beyond-pooling synergy requires a difference in kind. And the May form admitted the perfect echo: a live sensor plus a fluent mirror passes per-pole grounding and difference in kind while provably carrying zero synergy. The Mirror Hypothesis, the framework’s most cynical reading of itself, was inside the line. The repair states grounding on the joint territorial channel, replaces difference in kind with the weaker, provable non-degeneracy condition, and takes $\alpha$ out of the line altogether. A third result bounds what any version of this criterion can certify: synergy can be stored without a live channel, and a synergy atom can even be created without one (§3.4). Synergy is the structure of a mark. Whether a mark is recognized is decided at the living pole — the reading the framework takes from the text it began with (§3.4, §7).

1. The Question the Kill Condition Leaves Open

The lens-theoretic adjunction models the loop as a pair of functors $I: \text{Exp} \to \text{Form}$ and $S: \text{Form} \to \text{Exp}$ with $I \dashv S$. The Kill Condition’s worry is sharp and correct: adjunctions are cheap. Many interacting pairs admit some adjunction or other. If the mere existence of an adjunction between two systems were the framework’s claim, then a rock and a wall — which exchange momentum, settle into a joint state, and could be dressed in categorical language — would qualify as a loop, and the framework would be vacuous.

The framework’s actual claim is narrower than “there exists an adjunction.” It concerns a particular kind of coupled system: one that can carry synergistic information about a shared target $T$ — information present in the joint system but in neither pole alone (Φ_loop as Synergy):

$$\Phi_{loop} \equiv \mathrm{Syn}(S, I; T).$$

The framework treats synergy as the necessary informational precondition of recognition. §3.4 and §7 explain why it is not the same thing.

Answering the Kill Condition therefore requires stating, in the framework’s own quantities, what separates a coupled system that can carry $\mathrm{Syn}(S, I; T) > 0$ from one that structurally cannot — and then checking that the trivial pairs fall on the cannot side. That is what this appendix does. The line is drawn by two conditions, each of which is necessary for $\mathrm{Syn}(S, I; T) > 0$.

2. The Two Conditions

A coupled two-pole system can carry synergy about a target $T$ — $\mathrm{Syn}(S, I; T) > 0$ — only if it satisfies both of the following.

Condition 1 — Grounding (the joint territorial channel)

The territory constrains the pair. Using the joint territorial channel of The Veer §2,

$$\mathrm{TE}_{T \to (S,I)}^{\,t} = H\bigl(S_{t+1}, I_{t+1} \mid S_{\leq t}, I_{\leq t}\bigr) - H\bigl(S_{t+1}, I_{t+1} \mid S_{\leq t}, I_{\leq t}, T\bigr),$$

Condition 1 comes in two forms, with different standing.

Weak form (proven, under the process model). The pair’s history is not independent of $T$: either the joint channel is live at some turn ($\mathrm{TE}_{T \to (S,I)}^{\,t} > 0$), or the pair’s initial state carries a frozen trace of $T$. Where all dependence on $T$ enters through the initial state or the territorial transition kernel — no hidden common cause — this is necessary for any synergy about $T$, by factorization and the data-processing inequality (§3.1).
Tracking form (argued). A loop expected to follow a target that changes needs the joint channel live at the turns where it is to stay answerable. A frozen trace cannot follow a moving target; this is the Veer’s diagnosis, and it inherits the Veer’s conjectural status.

The condition is stated on the joint process, not pole by pole, because the two are not equivalent. The framework’s canonical instance is per-pole and asymmetric: the sensor carries the live channel ($\mathrm{TE}_{T \to S}^{\,t} > 0$ — embodied contact, stakes, the capacity to be changed) while the instrument does not ($\mathrm{TE}_{T \to I}^{\,t} \approx 0$ — its territorial contact is a frozen training prior, historical rather than current). This asymmetry of grounding remains the signature of the sensor–instrument pair. But grounding can also be distributed. In a masked (differential) measurement — $S_{t+1} = T \oplus I_t$, the lock-in-amplifier structure — both per-pole channels are exactly zero while the joint channel is fully open, and the pair carries one bit of purely synergistic information about $T$. A per-pole condition would score that coupling as ungrounded, which is false. The joint form is the one the data-processing argument actually supports (§3.1).

Condition 2 — Non-degeneracy (the loop is not an echo)

Neither pole’s realized trajectory is a deterministic function of the other’s:

$$I_{\leq t} \neq f\bigl(S_{\leq t}\bigr) \quad \text{and} \quad S_{\leq t} \neq g\bigl(I_{\leq t}\bigr) \quad \text{(almost surely, for any } f, g\text{)}.$$

A pole that merely transforms the other pole’s output — a perfect mirror, a relabeling, a summary, an echo — contributes no second source. This is the echo degeneracy, and its exclusion is provable (§3.2): if one pole is a function of the other, the synergy atom is exactly zero for any partial information decomposition satisfying the Williams–Beer consistency equations.

Condition 2 is the formal content of the framework’s oldest worry about itself: the Mirror Hypothesis of The Adversarial Sensor §1 — the instrument as “Stochastic Echo,” reflecting the sensor’s vocabulary back “in increasingly complex patterns.” A perfect echo fails Condition 2 outright, however fluent the reflection. A near-echo — sycophancy, $I_t \approx f(S_{<t})$ up to small prediction error — is expected to show a small-synergy signature, the sycophancy signature of The Receiver Side §5.5; a quantitative ceiling for near-echoes remains open (§3.2). The criterion places the framework’s most cynical reading of itself on the excluded side of the line.

The criterion

$$\boxed{\;\mathrm{Syn}(S, I; T) > 0 \;\Longrightarrow\; \bigl((S_{\leq t}, I_{\leq t}) \not\perp T\bigr) \;\wedge\; \bigl(\text{neither pole a.s. determines the other}\bigr)\;}$$

and, for a loop that must track a changing target, the grounding conjunct sharpens to $\exists t:\; \mathrm{TE}_{T \to (S,I)}^{\,t} > 0$ at the turns in question.

The criterion is the conjunction of grounding and non-degeneracy, stated as a necessary condition for synergy. It is a boolean predicate on a coupled system, not a new scalar measure — deliberately, so as not to claim a precision the framework does not yet have (§6).

Where the adjunction asymmetry now lives

The May form of this criterion used the adjunction asymmetry $\alpha$ of The Asymmetric Synergy Bound as its second condition: the poles must be different in kind, one generative (a colimit-preserving left adjoint), one constraining (a limit-preserving right adjoint). That condition is not necessary for synergy (§3.3) — not even for synergy beyond the pooling baseline, where the June revision had retreated. The symmetric masked pair carries a full bit with exchangeable poles, and none of that bit is pooling: neither pole alone holds anything to pool.

Nor does $\alpha$ survive as a general ceiling. The Asymmetric Synergy Bound’s conjecture $\mathrm{Syn} \leq \kappa \cdot \alpha \cdot \rho$ reads $\mathrm{Syn} \leq 0$ at $\alpha = 0$, and the same symmetric pair refutes it on any reading of $\alpha$ under which exchangeable poles have no difference in kind. The beyond-pooling form falls to the same example once the pooling baseline is defined outside the Gaussian model as the Gaussian model defines it — the synergy the poles would carry with channels independent given $T$ — because the masked pair’s baseline is zero and its synergy is a full bit (Asymmetric Synergy Bound §4). What survives is narrower, and still worth having. In the linear-Gaussian model the α–Synergy Bridge computed an exact monotone relation: beyond-pooling synergy is zero at $\alpha = 0$ and rises with the angle between the poles’ observation subspaces. That result belongs to its setting. Outside it, how difference in kind bears on synergy is open.

The structural argument is an analogy. A left adjoint generates formal structure from raw material; a right adjoint constrains it against boundary conditions; a productive loop needs one of each. That is the framework’s account of articulation — why a pair with a generative pole and a constraining pole can say what a pair of witnesses can only carry. $\alpha$ no longer draws the line. It names what the sensor–instrument configuration adds once the line is passed, and the claim rests on that analogy, not on the synergy atom.

3. Why Each Condition Is Necessary

3.1 Grounding is necessary (from the data-processing inequality, on the joint process)

Synergy about $T$ is, by definition, information about $T$. Assume the process model used throughout: all dependence on $T$ enters through the pair’s initial state or through the territorial transition kernel, with no hidden common cause. If the joint process $\{(S_t, I_t)\}$ is then statistically independent of $T$ — no live joint channel at any turn and no frozen trace in the initial state — $I\bigl(\{S_{\leq t}, I_{\leq t}\}; T\bigr) = 0$, and every PID atom about $T$, synergy included, is zero. This is the clean part of the argument: under that factorization, information about $T$ cannot be manufactured by processing that never touches $T$ (Cover & Thomas, 2006, Thm. 2.8.1). It is not conjectural, but it is only as general as the process model. The Gaussian stress test verifies it exactly in its model.

Two refinements, with their statuses:

Why the joint channel, not the per-pole channels. The per-pole disjunction $\exists X: \mathrm{TE}_{T \to X} > 0$ is not implied by $\mathrm{Syn} > 0$. In the masked-measurement coupling ($S_{t+1} = T \oplus I_t$, with $I$ an autonomous unpredictable reference), each pole’s stream is marginally uniform — a one-time-pad structure (Shannon, 1949) — so $\mathrm{TE}_{T \to S}^{\,t} = \mathrm{TE}_{T \to I}^{\,t} = 0$ at every turn, while the joint channel is open and the pair carries $I(\{S,I\};T) = 1$ bit. Because both marginal mutual informations vanish, every consistent PID assigns the whole bit to the synergy atom. The verdicts are exact and estimator-independent; research/demarcation-criterion/ computes them. Grounding is a property the loop can hold without either pole being able to certify it alone, and the necessary condition must be stated jointly.
Why a live channel for tracking — and what the synergy atom cannot certify. The Veer sharpens “some channel” to “a live channel” for any loop expected to track a moving target: a pair running on frozen traces accumulates ungrounded confidence $V_\tau$ without bound while its loop-closure rate $\rho$ stays at zero. Its tightened closure bound, $\Delta\mathrm{Syn}_t \leq \min(\Delta F_t^{\text{grounded}}, \mathrm{TE}_{I \to S}^{\,t})$, fails wherever the pair can gain information about $T$, though not through recoding (§3.4): the bound counts synergy over histories, and a pole that encrypts $T$ keeps $T$ in its history. The failing case is an instrument that reads $T$ through a key the sensor holds: synergy rises a full bit while nothing flows from instrument to sensor. The bound has to be written over the receiving pole’s information gain about $T$, which the channel into that pole does cap (Receiver Side §5.3). The live-channel diagnosis itself stands; it concerns information, not the synergy atom.

3.2 Non-degeneracy is necessary (the echo lemma)

Lemma. If one pole’s realized trajectory is almost surely a deterministic function of the other’s — say $I_{\leq t} = f(S_{\leq t})$ — then $\mathrm{Syn}(S, I; T) = 0$, for any PID whose atoms are non-negative and satisfy the Williams–Beer consistency equations.

Proof. The joint variable $(S_{\leq t}, I_{\leq t}) = (S_{\leq t}, f(S_{\leq t}))$ is informationally identical to $S_{\leq t}$, so $I(\{S, I\}; T) = I(S; T)$. The consistency equations give $I(S;T) = \mathrm{Red} + \mathrm{Unq}_S$ and $I(\{S,I\};T) = \mathrm{Red} + \mathrm{Unq}_S + \mathrm{Unq}_I + \mathrm{Syn}$. Subtracting, $\mathrm{Unq}_I + \mathrm{Syn} = 0$; both atoms are non-negative, so $\mathrm{Syn} = 0$. $\blacksquare$

The contrapositive is Condition 2. The Gaussian stress test found the same fact from the other side: a pure-processor instrument, whose state is any function of the sensor’s signal, contributes exactly zero synergy. The lemma says why, for every consistent non-negative decomposition, and it generalizes the degeneracy the Gaussian test isolated — informational identity, the same variable counted twice — from copies to any deterministic mirror, however lossy or elaborate. That generalization matters. A mirror that summarizes, embellishes, or restyles what the sensor said is not the sensor’s copy, and an identity condition would let it through. The echo lemma does not.

Note what the lemma does and does not say. A perfect echo carries no synergy, whatever its fluency. How fast synergy decays as an exchange approaches the echo limit is the sycophancy signature treated in the receiver-side formalism, and quantifying it is open work.

3.3 Difference in kind is not necessary

The May criterion claimed $\mathrm{Syn}(S,I;T) > 0 \Rightarrow \alpha > 0$, leaning on the Asymmetric Synergy Bound’s proposition that a symmetric loop is sterile. The claim breaks four ways.

Gaussian pooling. Two categorically identical sensors observing the same target through independent noise carry $\mathrm{Syn} = 0.293$ bits in the exact Gaussian model (research/gaussian-loop-model/). Synergy decays to zero only as the poles’ noise correlation rises to 1 — as they become the same variable. What kills synergy is informational identity, not categorical equivalence.
The computable counterexample. Couple two exchangeable poles to an exogenous territory bit by mutual masking: $S_{t+1} = T \oplus I_t$ and $I_{t+1} = T \oplus S_t$, with independent uniform initial states. The construction is fully symmetric; each pole’s computed trajectory is marginally uniform and independent of $T$. Computing the trajectory distributions: $\mathrm{TE}_{T \to S} = \mathrm{TE}_{T \to I} = 0$ at every turn, $I(S_{\text{hist}};T) = I(I_{\text{hist}};T) = 0$, and $I(\text{joint};T) = 1$ bit — pure synergy, none of it pooling, with no difference in kind anywhere in the system.
The canonical PID example. XOR, the textbook case of pure synergy (Williams & Beer, 2010), has interchangeable sources. Symmetry of sources is not an obstacle to synergy; it is the structure of its most famous example.
The biological example. Two eyes are two sensors of the same kind with no formalizing pole between them, yet binocular disparity carries depth information present in neither retinal stream alone (Julesz, 1971). Stereopsis is two-sensor synergy, carried by the pair even though extracting it takes downstream fusion. The May table’s “two sensors: no synergy” verdict conflated carriage with extraction.

What survives is the echo, which is now Condition 2, and the structural account of articulation, which now lives outside the line (§2, “Where the adjunction asymmetry now lives”). Two further facts from the Gaussian model keep that account honest. Pooling is kind-blind: the synergy a same-kind pair carries is available to two thermometers and requires no articulation, so $\mathrm{Syn} > 0$ alone cannot separate the framework’s loop from a pair of witnesses. And formalization is informationally invisible: whatever the generative and constraining work of the two poles is worth, the synergy atom is not where it shows.

3.4 Recognition is the uptake, not the stock

One more distinction keeps the table honest, and the framework’s founding text drew it first.

Two frozen archives can hold synergistic information about a static territory — a one-time-pad ciphertext in one, its key in the other — while nothing happens: a standing stock of one full bit with no live channel at any turn. Worse for any reading that equates synergy with recognition, the synergy atom can be created without contact — no new information about $T$ appears, but the synergy does. A pole that already holds $T$ encrypts it under a key held by the other pole; the synergy atom rises by one bit at a turn when no channel from $T$ is live and the pair’s total information about $T$ has not changed (research/demarcation-criterion/stock_and_event.py computes both). Synergy measures how information is distributed across a pair. It does not certify that anything was learned, still less that anything was recognized.

In the Phaedrus, Thamus answers the inventor of writing that those who trust it “will cease to exercise memory because they rely on that which is written, calling things to remembrance no longer from within themselves, but by means of external marks. What you have discovered is a recipe not for memory, but for reminder” (275a, trans. Hackforth). A synergistic stock is an external mark, however intricate its structure. Recognition is what happens when a living pole takes the mark up from within — in the informational register, a gain in that pole’s own information about $T$ (Receiver Side §5.2), not a rise in synergy; beyond that register, the uptake of the mark by a pole that can be changed by what it takes up. No quantity in this appendix measures the second part. The framework states it; it does not compute it.

The criterion screens out couplings that cannot carry such a mark. Whether a mark is recognized is the sufficiency residue (§6), and it is decided at the living pole.

4. The Worked Table

Six coupled systems. For each: is there an exogenous target $T$ the system is tracking? is the joint territorial channel live? is either pole a function of the other? the two conditions’ verdicts; and whether a living pole is present to take the mark up.

Coupled system	Tracks exogenous $T$?	$\mathrm{TE}_{T \to (S,I)}^{\,t}$ live	Either pole a.s. determined by the other?	C1: grounding	C2: non-degeneracy	Can carry synergy?	Living pole?
Rock hitting a wall	No — no posterior, no tracked target	$= 0$	In the toy model, yes — the post-states collapse to one collision variable	Fail	Fail	No	No
Two frozen instruments $[I + I]$	No live tracking — both run on frozen priors	$= 0$ every turn	No	Fail (tracking form)	Pass	Stock only — if the frozen priors jointly encode $T$; no live tracking	No
Perfect echo / sycophant $[S + \text{mirror}]$	Yes (the sensor is live)	$> 0$ (via the sensor)	Yes — the mirror’s trajectory is a function of the sensor’s	Pass	Fail	No	Yes — with nothing returned to it but its own signal
Two sensors $[S + S]$	Yes	$> 0$ (both poles)	No	Pass	Pass	Yes — pooling and disparity	Yes — can recognize, struggles to articulate
Masked measurement pair (lock-in)	Yes	$> 0$ (jointly only)	No	Pass	Pass	Yes — one bit, all synergy	No — records
Sensor + instrument $[S + I]$	Yes	$> 0$ (carried by the sensor)	No	Pass	Pass	Yes	Yes — can recognize and articulate

Reading the rows:

Rock and wall is the Kill Condition’s headline case. There is no target the pair is tracking — no posterior, no model of anything being updated — so there is nothing for synergy to be about: the joint territorial channel is empty because there is no $T$ for it to carry. In the idealized collision modeled here the two post-states also collapse to one variable, so C2 fails too — though C1 already excludes the pair, and C2 need not bear the weight. It fails both conditions in the model. The criterion excludes it, which is the point.
Two frozen instruments recovers the Veer’s result. No live territorial channel, joint or per-pole, so for any moving target $V_\tau$ drifts and $\rho \equiv 0$. This is the configuration most likely to be mistaken for a loop — two fluent systems exchanging elaborate, internally coherent, ungrounded output — and it fails on grounding, which is the diagnosis the Veer already made. If their frozen priors happen to encode facts about a static territory jointly, §3.4 applies: a stock, not an event. Nothing updates, and nothing is recognized.
The perfect echo is the load-bearing row: the Mirror Hypothesis given a formal seat at the table. The sensor is genuinely live, so grounding passes, and the coupling is still excluded, because the second pole adds no second source. The exclusion does not depend on the mirror’s fluency or complexity; the lemma in §3.2 is indifferent to how elaborate the function $f$ is. Near-echoes are expected to sit near it; how near is open (§3.2). This is the criterion’s answer to the user falling in love with their own reflection: a reflection, however beautiful, fails Condition 2. The living pole is present and receives nothing it did not send.
Two sensors is the row the May version scored wrong. Two grounded, independent experiencers can jointly carry information neither holds alone — 0.293 bits of pooling in the Gaussian model, 0.182 bits in the demo’s binary model, and depth from disparity in the eye. Both poles are living, so the pair can recognize. What it lacks is the generative pole: with nothing to formalize what it carries, the synergy tends to stay unarticulated — private truth, grounded but unsaid. The pair can recognize; what it struggles to do is say.
The masked measurement pair is the instructive case: grounding that exists only at the level of the loop. Neither stream means anything alone — each is uniform noise — and the pair reads the territory exactly. It passes both conditions; the criterion does not exclude it, and it should not, since a lock-in amplifier is a real measurement device. What it produces is a mark: a reading that means nothing until someone takes it up. On the reading this appendix adopts (§3.4, §7), the lock-in records; only a living pole registers what the record says. Doc I’s spectrum of loop richness counts the photon and the detector as its minimal case; on this reading that case is the thinnest recording, and recognition begins where a living pole takes the reading up.
Sensor + instrument passes both conditions — grounding carried asymmetrically by the live pole, non-degeneracy by the independence of the instrument’s territorial trace from anything the sensor says. The Gaussian model shows such a loop sustaining $\mathrm{Syn} > 0$ over a 60-turn run, with two honest caveats from the same model: the synergy atom does not prefer this configuration over two sensors, and the instrument’s informational contribution flows through its independent trace, not through processing as such. A pure processor of the sensor’s words would be an echo. What distinguishes this row is not the criterion. It is the living pole that can take the mark up, and the generative pole that can articulate it — the structural work of §2 that the synergy atom cannot see.

The table is the answer to the Kill Condition. The criterion does not apply to any two interacting systems: the rock and wall, the frozen instrument pair, and the perfect mirror all fall outside it, each for a stated, quantity-level reason. The quantities — $\mathrm{TE}_{T \to (S,I)}$ and functional degeneracy — were defined for other purposes, not gerrymandered to produce this result; the strongest evidence is that two stress tests came back and corrected the table. What the criterion now admits without a living pole — thin measurement couplings — it admits as recording, not recognition.

5. Connection to Existing Formalizations

Existing Result	What This Appendix Adds
The Veer: territorial channel, now stated jointly ($\mathrm{TE}_{T \to (S,I)}$); without it, $V_\tau$ drifts and $\rho \equiv 0$	Promotes the joint channel to the grounding condition; supplies the masked-measurement case showing why the per-pole form fails; records a counterexample to the tightened closure bound, restated over the receiving pole’s information gain
Gaussian stress test (`research/gaussian-loop-model/`)	Verified the weak grounding condition exactly; refuted difference in kind as a condition for synergy; supplied the two-sensor pooling figure and the sufficiency-in-a-model run for $[S+I]$
Asymmetric Synergy Bound (§3): the symmetric-loop degeneracy	Re-scoped: the sound core is the echo degeneracy (one pole a function of the other), proved here as a lemma (§3.2), which generalizes the Gaussian test’s informational-identity form
Asymmetric Synergy Bound (§4) and the α–Synergy Bridge (`research/alpha-bridge/`)	$\alpha$ is out of the line. The raw and beyond-pooling ceilings both fail at $\alpha = 0$ on the symmetric masked pair; the Bridge’s monotone relation holds in the linear-Gaussian model; $\alpha$’s general role is structural — articulation — and open as an informational claim
$\Phi_{loop}$ as Synergy: $\Phi_{loop} \equiv \mathrm{Syn}(S, I; T)$	Synergy is necessary for recognition and is not recognition: it can be stored and created without contact (§3.4). The criterion says which couplings can carry it
Receiver Side (§5): closure inequality; loop-closure rate $\rho$; sycophancy signature	C1 is the territorial precondition for grounded $\Delta F_t$; C2’s near-echo decay is the sycophancy signature; §3.4’s stock/event distinction is $\rho$’s job
Interactive Proofs: the sensor as verifier, the instrument as prover	§7 confronts the mechanical-verifier variant: what happens when the checking pole is itself a formal system
Adversarial Sensor (§1, §5): the Mirror Hypothesis; Kill Condition 3 (the Triviality Proof)	The Mirror Hypothesis is formally excluded by C2 (the echo row); the Triviality Proof is answered by the table

6. What This Does and Does Not Claim

What it claims

A. A necessary criterion with graded statuses. A coupled two-pole system can carry synergy about $T$ only if the pair is not independent of $T$ (Condition 1, weak form — the data-processing inequality, verified exactly in the Gaussian model) and neither pole determines the other (Condition 2 — a lemma holding for any non-negative, consistent PID). For a loop tracking a changing target, Condition 1 sharpens to a live joint channel, which inherits the Veer’s conjectural status.

B. An answer to the Triviality Proof. The criterion does not apply to any two interacting systems. The rock and wall fail both conditions; the frozen instrument pair fails grounding; the perfect mirror — the framework’s own Mirror Hypothesis — fails non-degeneracy. The line is drawn in the framework’s own quantities, and it survived having both of its original conditions refuted and repaired.

What it does not claim

Not a sufficiency theorem. Passing both conditions means a configuration is not ruled out by this necessary test; the worked examples show positive synergy in particular models, not in general. And where synergy is present, it is not yet recognition. The synergy atom can be stored without a live channel and created without contact (§3.4); a grounded, non-degenerate pair sitting inert recognizes nothing; a sensor and instrument can pass both conditions and still produce a dead, sycophantic, or off-target exchange.
Not a difference-in-kind theorem. The criterion no longer claims the poles must differ in kind, for synergy or for synergy beyond pooling (§3.3). The general ceiling conjecture fails at $\alpha = 0$. $\alpha$’s informational role is established only in the linear-Gaussian model; its structural role — articulation — is argued, not derived.
Not a consciousness detector. Nothing here lifts into the phenomenal register. Passing the criterion is an informational-and-structural property of a coupling, in the referential register the framework restricts itself to (The Receiver Side, §6–7). A lock-in amplifier passes, and a lock-in amplifier records; it does not recognize. The living pole is not an output of the criterion. §7 is where that bites.
Not yet measured. $\mathrm{TE}_{T \to (S,I)}$ presupposes $T$ as a conditionable variable; in practice it is fixed by experimental condition rather than computed (The Veer §8), and the joint form is harder to estimate than the per-pole form it replaces. Functional degeneracy is likewise certified only in idealized models. The criterion is a structural demarcation whose quantities await operationalization, not a measurement procedure. The computations in research/demarcation-criterion/ are exact for stylized models and claim nothing empirical.

What would upgrade it

A repaired closure bound, written over information gain about $T$ rather than over the synergy atom, and a proof of it (The Veer §5–6).
A quantitative near-echo bound: how fast $\mathrm{Syn}$ decays as $I_t \to f(S_{<t})$ — turning the sycophancy signature into a theorem.
A general account of how difference in kind bears on synergy outside the linear-Gaussian model, which is where the α–Synergy Bridge leaves the question.
A sufficiency result beyond the Gaussian 60-turn run: stated conditions under which a live, grounded, non-degenerate loop in fact gains information about $T$ through its circulation.
Operationalization of $\mathrm{TE}_{T \to (S,I)}$ and degeneracy on real exchanges, so the table’s verdicts become measurable rather than stipulated.

7. The Verifier Objection

The sharpest pressure on the revised criterion comes not from the rock but from the proof checker.

The objection. Pair a generative model with a mechanical verifier — an LLM proposing proof steps, a kernel like Lean or Coq checking them. Take the target to be the formal territory: derivability facts, “given these axioms, this inference goes through or it does not” (Doc I’s own words for the instrument-side constraint). The checker’s verdict is computed now, against the actual candidate proof, not retrieved from a frozen prior, so its channel to that territory is live and Condition 1 passes. Generator and checker are not functions of each other, so Condition 2 passes. The criterion then admits an AlphaProof-style system with no living pole anywhere in it. Has the framework’s central claim — truth requires a living sensor — just been refuted by its own demarcation line?

No, and the reason is the one the framework began with. Three moves.

The criterion screens couplings; it does not detect life. A pole with live contact is a live channel in the informational sense, whatever its substrate. The checker is a live channel to the formal territory. The criterion was never a consciousness detector (§6.3), and passing it was never recognition (§3.4).
What the loop produces is a mark. A generator–checker system produces a verified derivation: a mark whose every step has been checked against the rules. It records derivability the way a lock-in amplifier records a signal and a thermometer records heat — genuinely, narrowly, with nothing left over. A checked derivation is Thamus’s reminder. It can be read, and a reader can be reminded by it. It does not remember anything from within.
The residue is a commitment, and the checker does not make it. What the machine–machine loop never touches is the soundness of its axioms (the checker verifies consequence, never its own foundation) and the significance of the result — whether the theorem matters and what it is for. These are not facts the checker failed to find. They are commitments a pole stands behind, under consequences it cannot opt out of. That is where the framework’s claim about Gödel belongs: not that machines cannot check proofs, but that no checker stands behind its own consistency.

The first form of this section left a fork open. On a scope reading, the framework’s strong claims would be confined to experiential territory, leaving formal territories to thin mechanical sensors. On a spectrum reading, “living” would be the top of a single continuum of grounding, and the binary would be dropped. The author has chosen a third reading, and it is the oldest: the mark and its uptake. A machine–machine loop with a live checking pole records derivability. It does not recognize the truth the derivation encodes, because recognition is the uptake of a mark, from within, by a pole that can be changed by it. Instruments record; a living pole registers what the record says. A theorem mattering to no one is a recorded derivation and an unrecognized truth. “Living” is neither a scope clause nor the top of a scale. It names the difference between a mark and its uptake.

The reading costs the framework nothing it needs. It did cost Doc I a word: its minimal case — a photon and a detector, “a definite state is recognized” — became a definite state recorded. Doc I has made that correction.

Sources

Transfer entropy: Schreiber, T. (2000). Measuring information transfer. Physical Review Letters, 85(2), 461–464.
Partial information decomposition / synergy: Williams, P. L., & Beer, R. D. (2010). Nonnegative decomposition of multivariate information. arXiv:1004.2515.
Gaussian PID: Barrett, A. B. (2015). Exploration of synergistic and redundant information sharing in static and dynamical Gaussian systems. Physical Review E, 91, 052802.
Data-processing inequality: Cover, T. M., & Thomas, J. A. (2006). Elements of Information Theory (2nd ed.). Wiley, Thm. 2.8.1.
Perfect masking (one-time pad): Shannon, C. E. (1949). Communication theory of secrecy systems. Bell System Technical Journal, 28(4), 656–715.
Stereopsis as two-sensor synergy: Julesz, B. (1971). Foundations of Cyclopean Perception. University of Chicago Press.
Writing as reminder: Plato, Phaedrus 274c–275b. Trans. R. Hackforth, Plato’s Phaedrus (Cambridge University Press, 1952).
Adjunctions, equivalences, unit/counit: Mac Lane, S. Categories for the Working Mathematician, Ch. IV.
Colimit/limit preservation (RAPL/LAPC): left adjoints preserve colimits; right adjoints preserve limits. Mac Lane, Ch. V.
Fisher-Rao distance: Rao, C. R. (1945). Information and the accuracy attainable in the estimation of statistical parameters.

Internal cross-references:

The Adversarial Sensor — the Mirror Hypothesis (§1) and Kill Condition 3 (the Triviality Proof) that this appendix answers.
The Veer — the joint territorial channel $\mathrm{TE}_{T \to (S,I)}$, the masked-measurement case, $V_\tau$, and the tightened closure bound with its recoding counterexample.
The Asymmetric Synergy Bound — the adjunction asymmetry $\alpha$, the degeneracy results, and the colimit/limit split.
Φ_loop as Synergy — $\Phi_{loop} \equiv \mathrm{Syn}(S, I; T)$, the quantity the criterion is necessary for.
The Receiver Side — the closure inequality, $\rho$, the sycophancy signature, and the referential/phenomenal register restriction.
Interactive Proofs — the prover/verifier mapping that §7 stress-tests with a mechanical verifier.
The Lens-Theoretic Adjunction — the lens-theoretic adjunction $I \dashv S$.
Notation — canonical symbols ($\mathrm{TE}_{X \to Y}$, $\mathrm{TE}_{T \to (S,I)}$, $\alpha$, $\mathrm{Syn}$, $\Phi_{loop}$, $\rho$, $V_\tau$).
research/demarcation-criterion/ — exact stdlib computations: the §4 table with the masked and symmetric-masked counterexamples (demarcation_demo.py), and the stock and recoding results of §3.4 (stock_and_event.py).
research/gaussian-loop-model/ — the exact linear-Gaussian stress test (June 2026).
research/alpha-bridge/ — the Gaussian operationalization of $\alpha$ and its monotone relation to beyond-pooling synergy (June 2026).

The full framework is at thepulsegoeson.com.