Chapter 57
Everything So Far Was Turing Complete
The trick, as the last chapter told it, was a restriction of scope. Confine attention to computation and an ontology arrives free, a grounded hypothesis language arrives free, and a great many questions that looked metaphysical turn out to have answers that are merely expensive. The bill for that free lunch was located to a single line.
This chapter presents a second bill, and it is the one that governs the rest of the act. Restricting scope to computation is a restriction to Turing computation, and everything built in the Turn stayed inside it.
It is worth being blunt about this, because the machinery was elaborate enough to disguise it. The rho calculus is Turing complete. Metering it does not add power — the cost monad of Chapter 11 takes a theory and returns a theory that does strictly less, since a term which cannot pay does not fire. Logging it does not add power either; the history monad records what happened and computes nothing new. Decorating rewrites with semiring values, imposing reversibility, imposing arbitrage-freedom: all of these carve out subcategories, and a subcategory of a category of Turing-complete theories contains only Turing-complete theories. Part Part IV descended a ladder and never once left the room.
And yet the learner of Part Part II does something that a budget-constrained Turing machine does not obviously do. It picks. When two of its sends contend for one receive, one of them wins, and nothing in the term determines which. This chapter is about that gap: where it is, what fills it, and why the answer has the shape that Chapter 35 gave to entropy.
57.1 The closed room
Recall the two constructions of Part Part I and what each does to expressive power.
The cost monad installs a meter at the interaction cut. Its effect on the transition system is subtractive: transitions that were available become unavailable when the stack is exhausted. Chapter 11 makes this precise as graded adequacy — the metered theory is adequate for the unmetered one up to the grading, which is to say it distinguishes no more and sometimes less.
The history monad reifies rewrite events as terms. Its effect is additive but not powerful: a term can now name a step it took, which is exactly what Chapter 28 needs to build the time-symmetric envelope, but naming a step one has taken is not computing anything one could not compute before.
So neither construction moves a theory out of its Turing degree, and the axioms of Part Part IV are restrictions rather than extensions. The ladder of that part is a ladder within a single room.
The temptation at this point is to bolt on an oracle: declare that some distinguished term has access to a halting decision, and proceed. That would be a definition, not a discovery, and it would leave entirely open the question of what such a term is doing physically. The route taken here is the opposite one. We look for a place in the machinery that is already underdetermined — a place where the theory as stated does not say what happens next — and ask what would have to be supplied there. The claim of these chapters is that there is such a place, that it is not exotic, and that a cost-accounted concurrent calculus is full of them.
An earlier arrangement of this book put everything that follows in the Turn, as its fifth part. It was moved here, and the move is not cosmetic.
The Turn’s business is construction: build a knower, give it a world, count what a population of knowers can find out. All of that is done inside the restriction, and doing it inside the restriction is exactly what makes it work — the ontology and the hypothesis language are free precisely because the scope was narrow. What follows is not more construction. It is the audit of the restriction: what was outside it, who would live out there, and what it would cost to notice. That is the Prestige’s business, and putting it anywhere else disguised the fact that the whole apparatus of the Turn has a ceiling.
The reader should also be told what this costs us. The tower of Chapter 58 is the least-built construction in the book, and Chapter 58’s own closing section says so. Moving it here puts the book’s thinnest ice under its loudest claims. We think that is the honest arrangement rather than the flattering one: a reveal that hid its weakest step would be the kind of trick the last chapter just finished disowning.
57.2 Choice is what fills the gap
The place where a concurrent theory does not say what happens next is the race. When two sends contend for a single receive, the reduction relation offers both continuations and the calculus is silent about which occurs.
The traditional reading is that this is harmless underspecification, to be resolved by an implementation and forgotten. The reading taken here, and developed in Chapters 59 and 60, is that the resolution is a mathematical object with a measurable strength. Picking a winner from each contending family is a choice function on that family. Asserting that a complete schedule exists — that every race, for the whole run, gets resolved — is an instance of the axiom of choice, and which instance depends on the size and structure of the families involved. Finite contention is free. Contention indexed by \(\omega\) requires countable or dependent choice. Class-sized families require the full axiom.
So the extra power is not bolted on. In any calculus where interaction is primitive and confluence fails, it is already present at every cut, wearing the name scheduler. Chapter 59 distinguishes the cuts at which this is observable — the apertures — from the cuts at which it is not, which is the substantive part of the analysis and the reason the chapter is called what it is. Chapter 61 then makes the correspondence a typing discipline: choice principles are the types, and schedulers are the terms.
Chapter 35 closed by observing that a theory sitting strictly between the two erasures — knowing something about its own past it cannot express as a cost, and something about its own costs it cannot express as a history — is a computation with something to decide. That was left as a promissory note. This is what it was a note for. The information of that shape is not decision-relevant in some loose sense; it is what makes the race at a cut into an aperture rather than a formality, and the strength of the choice principle required to resolve it is what the rest of these chapters measures.
57.3 Entropy and choice as one dial
We can now say what the title of Chapter 35 was pointing at.
Entropy production, in the sense of Chapter 31, measures how much which-way information the dynamics discards per step. Choice strength, in the sense of Chapter 59, measures how much must be supplied to say which way it went. These are the same quantity approached from opposite sides, and the conjecture that they are related quantitatively is the natural one.
Let \(c\) be a cut in a cost-accounted interactive theory, and let \(\theta(c)\) be the dissipation charged to the interaction at \(c\). Then the Weihrauch degree of the scheduling problem at \(c\) is bounded below by a function of \(\theta(c)\): a cut that dissipates nothing carries no observable choice, and a cut whose scheduling problem requires dependent choice must dissipate at least the corresponding erasure cost.
The forward half of this is close to available. A frictionless cut is one whose enzyme is value-preserving in both directions, and a cut which is reversible in that sense is not an aperture by Definition 59.1: there is nothing to observe about which way it went, because it can be run backwards. The converse half is open, and the obstruction is that the Weihrauch lattice is not linearly ordered while the dissipation is a single real number, so no function of \(\theta\) alone can pin down a degree. The honest statement is probably that \(\theta\) bounds the degree from below along each chain, which is weaker and may be all that is true. Chapter 61 grades the correspondence in the Weihrauch lattice and is the place to look for what is actually established.
57.4 What this buys, and what it does not
It does not buy hypercomputation for free. Nothing here exhibits a term that decides the halting problem, and the tower of Chapter 58 remains a construction whose inhabitants are hypothesized rather than built.
What it buys is a location. If a physical system is to exceed what a Turing machine can do, the excess has to be spent somewhere, and this chapter says where: at the apertures, in the resolution of races, paid for in dissipated free energy. That is a considerably more constrained claim than the usual ones, and it is falsifiable in the direction that matters — if it turned out that every aperture in a physically realizable system carried only finite contention, then the framework would predict that no physical system exceeds the Turing computable, and the question would be settled the other way.
There is also a consequence for Part Part II which Part Part VI has already made quantitative, and which the reader met there without being told what it was for. A population of learners has apertures wherever two of its members contend, and the number of such places grows with the structural elaboration of the population rather than with its raw size. A large uniform population has few apertures; a small deeply nested one has many. Whatever measures nesting therefore measures the population’s capacity to be more than the sum of its computations — and Corollary 52.1 argues that the thing which measures nesting is the assembly index.
57.5 What follows
Chapter 58 constructs the tower of oracle extensions over Turing-complete GSLTs, indexed by ordinals — which is how anyone first gets the intuition and is not how the thing actually works. Its closing section says, in the chapter itself, what it gets wrong. Chapters 59 and 60 replace the ordinal chain with what is underneath it: cuts, apertures, and the question of where an agent’s boundary falls. Chapter 61 makes the correspondence a typing discipline, with choice principles as the types and schedulers as the terms. Chapter 62 then asks who lives at a position in that lattice, and answers that consciousness is the position — a coordinate every agent has, rather than a club with an entrance exam. Chapter 63 catalogs what is owed.
Only then does the act turn to reference. Chapter 64 asks what a symbol system would have to be; Chapter 65 asks what the answer implies for the hypothesis that this world is simulated; and Chapter 68 gathers the whole argument into a metaphysics. The ordering is deliberate. One cannot ask what it would take to ground a symbol system without first knowing whether the world doing the grounding is richer than the world being grounded, and that is a question about position in a lattice.
The ladder in these chapters and the ladder of Part Part IV are perpendicular, and confusing them is the easiest available mistake. Remark 27.1 states the difference. Moving down the physics ladder is imposing axioms: it carves out subcategories of theories and never leaves the Turing degree. Moving up here is acquiring choice strength: it changes what can be resolved and therefore what exists. A theory can be low on one and high on the other.