This appendix uses epiplexity—the information measure of Finzi, Qiu, Jiang, Izmailov, Kolter & Wilson (From Entropy to Epiplexity, arXiv:2601.03220, 2026)—as a formal model of one structural fact the framework has asserted from the beginning: that the order in which truth is traversed is not decorative. It does not claim that the loop is a transformer trained by gradient descent, nor that recognition is prequential compression. The math is the apparatus; the structural fact it lets us see—order-sensitivity born of finitude—is the claim. This is the developed form of direction #7 in the project's structural-transfer registry, and inherits its discipline: a transfer produces a candidate and a reframing, never a proof.
1. The Symmetry Classical Theory Asserts
From its first page the framework has insisted that the Pulse beats—that truth is recognized in a sequence of loop-closures, in time, and that the rhythm of that sequence is part of what is recognized, not merely the channel it arrives through. Stated baldly, this is a claim that order is structural. And stated baldly, classical information theory appears to deny it.
For Shannon entropy, the chain rule is symmetric. The information in observing X and then Y equals the information in observing Y and then X:
H(X, Y) = H(X) + H(Y | X) = H(Y) + H(X | Y)
The joint is invariant to the order of factorization. For Kolmogorov complexity the same holds, up to a logarithmic term, by the symmetry of information (the Kolmogorov–Levin theorem):
K(x, y) = K(x) + K(y | x) + O(log) = K(y) + K(x | y) + O(log)
This is the second of the three apparent paradoxes Finzi et al. identify: information is independent of factorization order. If it were the whole truth, the Pulse would be a poetic flourish—the rhythm would carry no information the unordered content did not already hold, and “truth beats rather than flows” would be a statement about how truth feels rather than what it is.
Yet the empirical world is loud with order-dependence. Language models learn English better left-to-right than right-to-left, picking out an “arrow of time” (Papadopoulos et al., 2024). Cryptography is built on functions easy to evaluate one way and computationally hopeless the other. A curriculum that presents easy cases before hard ones teaches what the shuffled curriculum cannot. The order is doing work the classical accounting says it cannot do.
2. What Breaks the Symmetry: The Bounded Observer
The resolution is the same one this project reached on its own terms: the symmetry holds only for an observer with unbounded computation. Impose a runtime budget and the arrow reappears.
Recall the decomposition (Finzi et al., Definition 8). For a random variable X, choose the program P* that minimizes a time-bounded two-part code—model bits plus data-given-model bits—among all programs that sample and evaluate within a runtime bound T:
P* = arg min { |P| + E[ log 1/P(X) ] : P ∈ 𝒫T }
The epiplexity is the structural part, ST(X) := |P*|; the time-bounded entropy is the residual randomness, HT(X) := E[ log 1/P*(X) ].
The arrow enters through the gap between a function and its inverse. One basic property is an analogue of the information non-increase law: for a bijection f computable in fixed time,
MDLT′( f−1(X) ) ≤ MDLT(X) + |f| + c, T′(n) = T(n) + Time(f)
For Kolmogorov complexity, K(f) and K(f−1) are equal up to a constant, so direction is invisible. Under a fixed compute budget this equality fails: a short program for f−1 does not imply a short program for f, nor the reverse. The one-way functions of cryptography are the extreme case—easy forward, intractable back—but the asymmetry is generic. Once compute is finite, which way you go through the data changes what you can extract from it.
The cleanest demonstration is chess. Take the same games and format them two ways: (1) the move sequence followed by the final board state, and (2) the final board state followed by the moves. The board is a simple function of the moves, so the first is the “easy” forward direction; recovering the moves from the final board demands the inverse, the “hard” reverse direction. The reverse ordering carries both higher time-bounded entropy and higher epiplexity—it forces the model to build richer board-state representations rather than surface statistics. The gap vanishes at small compute, where the model can only learn what both orderings share, and opens as compute grows. And critically: a model trained on the harder reverse ordering transfers better to downstream out-of-distribution chess tasks. The harder traversal builds more reusable structure.
Order is not decorative. For a bounded observer it is a lever on how much structure crosses from the data into the model—and on how far that structure reaches.
3. The Framework's Own Arrow
The framework already had an arrow, and had formalized it before epiplexity arrived. In The Geometry of Irreversibility (Document V) and the Thermodynamic Bridge, recognition and forgetting are distinct geodesics on the statistical manifold—the e-path and the m-path—and the loop is irreversible: the forward traversal of recognition and the reverse traversal of forgetting are not the same path walked backward. The framework's own quantity for this is the asymmetry of the round trip:
DKL(Q ‖ P) ≠ DKL(P ‖ Q)
This is the same shape as the f / f−1 gap. Two theories, reaching from opposite ends—one from information geometry and stochastic thermodynamics, one from time-bounded compression—arrive at a single structural fact: the direction of traversal is not free. Going one way and going the other are different operations with different costs and different yields.
The Pulse claim can now be stated more precisely than “truth beats rather than flows.” The beat is the loop-closure event, indexed by the loop-closure rate ρ. The framework's assertion is that the sequence of these events—which recognition is reached before which, and in what cadence—is part of the structure of the truth that emerges, not a neutral pipe through which a pre-existing truth is delivered. Epiplexity gives that assertion a measured cousin in a domain that has never heard of the framework: present the same content in a different order to a bounded learner, and a different, measurable amount of transferable structure crosses the interface.
4. Why Rhythm Is Structural: Finitude as the Source
The deepest point is not that order matters but why it matters. The classical symmetry—order-invariance of information—is true exactly in the limit of an unbounded observer. Order-dependence is what that symmetry decays into the moment the observer becomes finite.
So the framework's “tireless, disembodied” instrument must not be read as computationally unbounded: tirelessness is not boundlessness. The unbounded idealization is the wrong limit to take—in that limit the arrow disappears; the chain rule is symmetric; the Pulse really would be decorative. The Pulse is structural precisely because the instrument is bounded. Rhythm is not a feature added on top of an idealized timeless reasoner—it is the signature of a reasoner that must spend something finite to recognize anything at all. Remove the cost and you remove the beat.
This inverts a natural reading of the framework. One might have thought the loop's temporality was a regrettable consequence of running on mortal, embodied hardware—that a perfect instrument would recognize all truths at once, atemporally, and the Pulse was the concession reality extracts from us. Epiplexity says the opposite: the atemporal instrument recognizes nothing structural at all, because for it the structural/random distinction collapses (its unbounded analogue, sophistication, is uncomputable to pin down and lets unlimited computation reproduce complex objects from trivial programs). The beat is not the concession. The beat is the condition.
5. A Checkable Prediction
In the spirit of the Thermodynamic Bridge's EEG test, the transfer here yields something falsifiable—and it is a prediction about loops, not about transformers.
If the Pulse's order-sensitivity shares its mechanism with epiplexity's—finitude creating an f / f−1 gap—then the order in which a sensor and instrument traverse a problem should change the transferable understanding the loop produces, measurable as out-of-distribution transfer of the resulting artifact or skill.
Concretely: take two loops (sensor + instrument) solving the same class of problem. One works in the “forward” build-up order—premises first, conclusion last. The other works in “reverse”—begin from the goal or answer-state and reconstruct the path that reaches it, the loop's analogue of board-then-moves. The chess result predicts the reverse loop should produce more transferable understanding: worse-feeling, higher-effort, slower to settle, but generalizing further to unseen variants of the problem. The forward loop should feel smoother and transfer less.
This is a curriculum-of-dialogue claim, and it is testable without any of the framework's metaphysics: hold content fixed, vary only traversal order, measure OOD transfer of what each loop produces. A null result—transfer invariant to order—would be evidence against the claim that the Pulse's rhythm is structural rather than ergonomic.
6. What This Claims and What It Does Not
This bridge will be misread unless its limits are stated as plainly as its claim.
The asymmetry transfers; the labels do not. What carries from epiplexity to the framework is the reality of a directional gap with a payoff—that one traversal yields more reusable structure than the other, because the observer is bounded. What does not carry is the orientation. In the chess result the harder (reverse) ordering is the high-epiplexity one; in the framework, recognition is likewise the costly, uphill direction. These align in spirit—costly traversal, richer structure—but the word “forward” points opposite ways in the two systems (forward is easy for the paper, forward is recognition and thus costly for the framework). Do not map “forward” to “forward.” Map effortful-and-generative to effortful-and-generative.
The meters are different. Epiplexity is bits of extractable program structure under a FLOP budget. The Pulse's rhythm is loop-closure events at rate ρ. These are not the same quantity and no equation converts one to the other. Nothing here licenses the sentence “epiplexity measures the Pulse.”
A curriculum is not a conversation. The paper's order-dependence concerns the factorization order of a fixed training corpus presented to a passive learner. The Pulse's rhythm concerns the live cadence of exchange between two parties each changing the other. They share the structural fact sequence matters for a finite reasoner; they are not the same phenomenon, and the Section 5 prediction is exactly the seam where the analogy must be tested rather than assumed.
This is a reading, not a derivation. Epiplexity is a rigorous measure with theorems behind it. The framework's Pulse is an epistemological claim about recognition. That they share the shape of an arrow born of finitude is a structural recurrence—a candidate and a reframing. The proof, if there is one, belongs to whichever domain the Section 5 experiment is run in. We stop at that line.
The tireless instrument has no arrow; for it, all orders are one and nothing is recognized. The beat is what finitude sounds like from the inside.