Section 21.8 asked for an entropy named on the corpus’s own phase space and not borrowed. The dm³ equations have been carrying one the whole time — it is what the contact form returns when you evaluate it on the corpus’s own vector field, and its zero set is the attractor
Chapter 21 §21.8 found that the corpus's contact form was Gibbs's with the heat term deleted, and closed with a precise demand: name a function $S$ on the corpus's own phase space, not borrowed from an analogy, or the $T\,dS$ term cannot be restored. Chapter 25 answered it for the closing field — an arithmetic entropy, $W(T) = d_1 - d_2$. That was one system, and not the one the framework actually runs on.
This chapter answers it for the dm³ system itself, and the answer required no new machinery at all. It is what happens when you take the contact form the corpus has written for a hundred and sixty-four files and evaluate it on the vector field the corpus has been integrating for just as long. Nobody had put the two objects side by side.
The dm³ toy model, as published (GTCT 2026, v4, equations 1–3), lives on $M = \mathbb{R}^2_{+} \times \mathbb{R}$ with the contact form $\alpha = dz - r^2 d\theta$ and the field
$$\dot r = r(1-r^2) + 2(r-1)e^{-z}, \qquad \dot\theta = 1, \qquad \dot z = r^2 - 2(r-1)^2 e^{-z}.$$
Evaluate the one on the other. It is a single subtraction:
identically on $M$, with equality precisely on $\Gamma = \{r = 1\}$. By §21.8, $\alpha(X)$ is $\delta Q - T\,dS$, so the entropy production of the dm³ flow is
$$\sigma(r,z) \;=\; -\frac{\alpha(X)}{T} \;=\; \frac{2(r-1)^2 e^{-z}}{T} \;\ge\; 0 .$$Three things follow at once, and none of them was put in by hand. COMPUTED
The dm³ flow satisfies Clausius by construction. $\alpha(X) \le 0$ everywhere, not as an assumption, a modelling choice or an imposed condition, but because that is what the equations say. The system could have been written so that this failed. It was not.
The attractor is the reversible locus. $\sigma$ vanishes exactly on $\Gamma = \{r=1\}$ and nowhere else. §21.8 established that Legendrian paths — those with $\alpha(\dot\gamma) = 0$ — are precisely the reversible, quasi-static ones. So the limit cycle is not merely where trajectories end up. It is the set on which the system stops paying.
The bill decays along the orbit. The $e^{-z}$ factor means dissipation dies out as the system climbs, so a trajectory pays most on approach and asymptotically nothing once settled. That is a dissipative structure in Prigogine's sense — order maintained by throughflow, with the throughflow falling to zero as the structure completes — expressed in the corpus's own coordinates rather than borrowed from his.
Version 4 of the source corrects versions 1–3, which misprinted the coupling as $e^{-r}$ where it should read $e^{-z}$. A result that depended on which of those was right would be worth very little. This one does not.
For any positive coupling $E$, writing $\dot z = r^2 - 2(r-1)^2 E$ gives
$$\alpha(X) = \dot z - r^2 \dot\theta = -2(r-1)^2 E ,$$
so the sign and the zero set come from the structure of the $\dot z$ equation — the $r^2$ that cancels against $r^2\dot\theta$, and the negative-definite remainder — and not from the choice of $E$. COMPUTED The identity holds for $e^{-z}$, for the mistaken $e^{-r}$, and for anything else in that slot. It was true in versions 1 through 4 alike, unnoticed.
As a reading check on the equations, the same field reproduces the source's own reported constants: $\partial\dot r/\partial r$ at $r = 1$ gives $\lambda(z) = -2(1-e^{-z})$, hence $\mu_{\max} = \lim_{z\to\infty}\lambda(z) = -2$; and the saddle cubic $r^3 - r^2 - 2r + 1$ is the minimal polynomial of $2\cos(3\pi/7) = 0.4450$, its root in $(0,1)$. CHECKED
That eigenvalue gives $\mu_{\max}$ and gives nothing else. It does not give $\tau = 2$. $\tau$ is the limit of the n-bonacci ladder — the root of $x^{n+1} - 2x^n + 1$ in $(1,2)$ as $n \to \infty$, established in The Ladder Polynomials — and it is arrived at from polynomial arithmetic with no dynamics in it at all. The two numbers agree, and agreeing is not the same as being the same claim. A reader who takes the transverse eigenvalue as a derivation of the embodiment threshold has merged an exponent with a root because both printed as 2, which is the failure this chapter's neighbours are about. OPEN whether there is a reason for the coincidence; none is offered here.
The natural way to want entropy in a framework is as another parameter: four dimensions, then time, then entropy on top of it. That is not available here, for two independent reasons, and the second is the interesting one.
The dimension is not free. A contact form must make $\alpha \wedge (d\alpha)^k$ a volume form, which forces $\dim M = 2k+1$. Contact manifolds are odd-dimensional. The corpus's $\alpha$ lives on 3 coordinates $(z; r, \theta)$; the Gibbs form of §21.8 on 5, $(U; T,S; \Omega,J)$; and the next slot is 7, not 6. Thermodynamic variables arrive in conjugate pairs — $S$ never travelled without $T$ — so the count goes $3 \to 5 \to 7$ and skips every even number. §21.8's restoration was the step from three to five, and it was taken before anyone counted it.
And on this system entropy is not even a coordinate. It is a function that the equations already determine, computed above. One does not add $\sigma$ to the model; one reads it off. That is a stronger position than having a parameter, because a parameter is a choice and a function is a consequence — and a consequence can be wrong, which is what makes it worth having.
GTCT's Axiom 3 makes time the parameter of the operator sequence, with the Reeb field as its geometric avatar, and Chapter 15 identifies the operator $T$ as the discretisation of the Reeb flow, $\xi_\alpha = \partial_z$. Since $\alpha(\xi_\alpha) = 1$ by definition, and §21.8 reads $\alpha(\dot\gamma)$ as $\delta Q - T\,dS$, one unit of Reeb flow carries one unit of $\delta Q - T\,dS$. A unit of generative time is a unit of dissipated energy.
Which is why nothing ever appeared to be spent. Under the adiabatic restriction $dS = 0$ the same calculation returns zero, so for a hundred and sixty-four files a turn of the chain cost nothing — not because time is free, but because the term that prices it had been deleted.
Part VII opened by asking what selects a structure rather than merely permitting it. It closes with three entropies named on three of the corpus's own objects, and with the honest size of each recorded.
| where | the entropy | zero on | how much it decides |
|---|---|---|---|
| Ch 25 | $S = k\log W$, $W = d_1 - d_2$ | the inert sizes | a tie-breaker worth $k\log 2$ |
| Ch 27 | $\langle S\rangle \sim \tfrac12 k\log\log T$ | — | doubly logarithmic against a linear cost: almost nothing |
| Ch 28 | $\sigma = 2(r-1)^2e^{-z}/T$ | the attractor $\Gamma$ | the whole approach; zero once settled |
The third is the only one that is not small. It is also the only one that is a rate rather than a count — which is the distinction §26.1 drew between a cost and a count, arriving here from the dynamics instead of the arithmetic.
This is the toy model. Result 28.1 concerns the three-equation dm³ system as published, and says nothing about the general operator chain $G = U \circ F \circ K \circ C \circ T$ acting on an arbitrary dm³ manifold. Whether $\sigma$ has this shape there is open, and the natural conjecture — that $C, K, F, U$ are flows of horizontal fields, $\alpha(X) = 0$, so that each alone is reversible and only their failure to commute reaches the Reeb direction — is stated here and not proved. OPEN
The source has now been checked, and it is current. Result 28.1 was first computed against the v4 manuscript, with a note that later revisions were reported and the two-line evaluation should be re-run before anything was built on it. That re-run has been done. The deposited record of record is 10.5281/zenodo.21708678 (GTCT 2026, v4.0, 30 July 2026; concept DOI 10.5281/zenodo.19498857), and it is the most recent version in the series. Its executable file dm3_simulation.py carries the right-hand side
character for character as §28.1 assumes — including the $e^{-z}$ coupling, so the $e^{-r}\!\to\!e^{-z}$ erratum of §28.2 is settled in the deposit rather than merely survivable. The accompanying FINDINGS.md reports $\mu_{\max} = -2$ recovered numerically to three decimals and an asymmetric inner basin boundary $r^\ast \approx 0.7732$, neither of which touches the identity: $\alpha(X)$ is computed pointwise and does not care which initial conditions reach $\Gamma$. CHECKED
What remains genuinely open is the paragraph above this one, not this one.
And §21.8's demand is met for this system only. It asked for an entropy on the corpus's own phase space; it now has one for the closing field and one for the dm³ toy model. It does not have one for the corpus.
What Part VII establishes, across four chapters, is narrower than it set out to prove and firmer for it: availability is arithmetic and cheap to compute; selection is thermodynamic and has to be paid for; and when you finally compute what a system pays, the number is usually smaller than the story around it. Twice out of three it was almost nothing. The third time it was the entire approach to the attractor, and zero thereafter.
28.1 Verify that $\Gamma = \{r=1\}$ is Legendrian for $\alpha$ and that the Reeb field is $\partial_z$. Why does the attractor being the zero set of $\sigma$ make $\Gamma$ a natural candidate for a reversible cycle?
28.2 Compute $\int_0^{T^*}\sigma\,dt$ along one turn of a trajectory at fixed $r \ne 1$, with $T^* = 2\pi$. What does the total dissipated per orbit tell you about how many turns the approach takes?
28.3 Suppose $C, K, F, U$ are flows of horizontal fields. Using $\alpha([X,Y]) = -d\alpha(X,Y)$ for horizontal $X, Y$, show that the chain can reach the Reeb direction only through non-commutativity, and relate this to §21.8's statement that one cannot gain $z$ without turning.
28.4 State what would have to be true of a general dm³ system for $\sigma \ge 0$ to hold identically, and say whether it is a condition on the manifold, the form, or the field.
dm3_simulation.py and FINDINGS.md.sorryAx. A clean axiom report is not a reading of the statement: per R20, a theorem can assume its conclusion and still report clean. Follow the link before citing one as evidence.