The Regress

Make the recourse precise. We want a metatheory M that grounds the object theory T — that settles what T's intended model is and certifies that T speaks truly of it. At a minimum, this means M proves that every theorem of T is true: the soundness of T. Formalized, soundness is a reflection principle, RFN(T), the schema Prov_T(⌜φ⌝) → φ ranging over all sentences φ. To ground T is to assert that what it proves holds — and that assertion is exactly reflection.

This is the exact twin of The Ascent, turned over. There the question was completeness, and the ladder rose because no consistent system proves its own soundness, so there is no roof. Here the question is grounding, and the same theorem — T ⊬ RFN(T), Gödel's second — says there is no floor: no theory certifies itself, so the certifying must be passed upward forever. One ladder, read in two directions; one theorem forbidding both the self-proving summit and the self-grounding base. What remains is to see the ladder's structure — for it is not a vague infinity but a precise transfinite object, and its precision is where the next surprise lives.

Give the ladder its rungs. Start with T₀ = T. At each successor step add the reflection principle for the stage below: T_α+1 = T_α + RFN(T_α) — or, in the leaner version, + Con(T_α). At a limit stage, take the union of everything below. This is a transfinite recursive progression, the object Turing introduced in 1939 as an "ordinal logic" and Feferman developed in 1962. But there is a catch already in the indexing. To climb to a transfinite stage α, the construction must refer to α — and a theory can only refer to an ordinal it can name. So the progression is indexed not by ordinals but by ordinal notations: concrete, recursive descriptions of ordinals, the members of Kleene's O. The ladder rises exactly as high as the climber can write down where it is going.

The stepper has previewed the whole drama. A precise transfinite ladder, forced rung by rung; a finite climb that never closes; a transfinite reach that recovers Gentzen's measure of PA and presses past it; and a limit at which completeness is attained — but only along the right path through the notations, and choosing that path is harder than anything the path proves. The two halves of that limit are two theorems, one of Turing's and one of Feferman's, and they are the heart of this panel.

Turing's 1939 thesis asked exactly the right question of the consistency progression: how much of arithmetic does climbing consistency settle? His answer was the first sign that the regress, though endless, is not powerless. Restrict attention to the Π⁰₁ sentences — those of the form "for all n, D(n)" with D decidable, the class that contains every consistency statement and every "this program never halts." Turing proved that the progression based on iterated consistency is complete for Π⁰₁: every true Π⁰₁ sentence is proved at some stage. The finitary core of arithmetic is reachable from below, by climbing consistency alone.

Two boundaries frame the result. Below, it is a genuine triumph: the consistency statements that Gödel's theorem keeps out of any single theory are all recovered, somewhere up the ladder — the finitary floor is reachable by ascent. But the result reaches only Π⁰₁, and the floor the descent was after — a full grounding, true of every arithmetic sentence — lies higher in the hierarchy. To climb past Π⁰₁ to all of arithmetic, mere consistency is too weak; one needs the full strength of reflection. That is the step Feferman took, and it reaches the summit — at a price exact enough to name.

Feferman replaced consistency with the full uniform reflection principle and let the progression climb along all of Kleene's O, every notation at once. The result is the most vertiginous theorem in the neighbourhood: the progression proves every true sentence of arithmetic, of whatever quantifier complexity. Completeness from below, in full — the summit The Ascent declared unreachable by any single effective theory is reached, by the transfinite iteration of reflection. For a moment it looks as though the regress has, against all expectation, terminated in bedrock. It has not, and seeing why is the whole point.

So this is the precise content of "the ground recedes." It does not mean the floor is unreachable in principle — Feferman shows it is reachable, at the transfinite limit. It means the reaching cannot be carried out: no effective procedure climbs to the grounding, because the procedure would have to solve a Π¹₁-complete problem to know where to climb. Grounding is available only to a reasoner who can already survey the whole of O — which is to say, to no reasoner that computes. The bedrock exists at the top of the ladder, and the ladder's rungs, past a point, can only be located by an oracle the descent has shown no theory to possess.

Feferman's completeness leaned on an outside oracle: the right notations were handed in, chosen non-effectively from the whole of O. Take the oracle away and impose the honest discipline instead — you may climb to a stage α only once you have proved that the notation for α really is a well-ordering. This is an autonomous progression: each new height must be earned by the heights already secured, with nothing assumed that has not been justified. Now the climb cannot leap; it can only grow by what it can certify. And a climb that grows only by self-certification does not rise forever. It closes at a definite ordinal — the exact altitude reachable without borrowing.

the regress, mapped  This is what the endless ladder becomes once measured. To every theory there belongs a single ordinal — its proof-theoretic ordinal — recording how far the consistency climb runs before that theory's strength is spent: ε₀ for PA, Γ₀ at the predicative limit, the Bachmann–Howard ordinal and beyond for the impredicative theories. The ordinal is a precise gauge of grounding power: it says exactly how much of the climb a theory can perform on its own, and exactly where it must borrow. The regress is not a fog. It is an ordered tower, each storey a theory, each height an ordinal, and the discipline of ordinal analysis is the science of reading those heights. The floor a theory can stand on is its ordinal; the floor it cannot reach is everything above.

Gather the ascent. To ground a theory is to prove it sound, and proving it sound forces a strictly stronger theory — so the search for a floor is a climb, and the climb is a precise transfinite progression indexed by ordinal notations. Turing's climb reaches every true Π⁰₁ sentence; Feferman's reaches every truth of arithmetic — but only non-effectively, the incompleteness conserved as the Π¹₁-complexity of recognizing the path. And the honestly reachable ground, the height a self-justifying reasoner can earn, is measured exactly by a proof-theoretic ordinal. The regress that The Descent named is now fully mechanized: no self-grounding base, a transfinite completion that no machine can compute, and each theory's reach pinned to an ordinal.

So this is the proof-theoretic twin of the descent the other panels traced in semantics. The Impostors, The Relativity, and The Original showed the floor cannot be fixed from inside — no theory pins its own model. The Regress shows it cannot be reached from outside either, except by an ascent that never lands at any computable stage and conserves Gödel's obstruction at every height. The two halves meet in one verdict: the ground is real — there is a true arithmetic, an intended model, a bedrock of fact — and it is reachable from no position the reasoner can occupy, neither below by definition nor above by computation. The diagonal that opened the corpus is carried, undissolved, up the entire ladder. It is the fixed point with no fixed point, and the descent's last word is that you cannot get beneath it: every floor you build rests on a higher one, forever, and the only footing is at infinity.