The shape of self-reference
The collection's whole argument rests on a picture of self-reference it has never questioned: something local — a sentence that speaks of itself, a set that contains itself, a function applied to itself. Paradox is self-reference is the diagonal. One famous construction appears to break the chain at its first link. Yablo's paradox is an infinite list of sentences in which no sentence refers to itself — each speaks only of those below it — and the whole is paradoxical all the same. If that is genuinely self-reference-free, the collection's central identification is false: here would be paradox without the diagonal. The resolution is the deepest revision the collection's notion of self-reference has needed. Self-reference is not local but structural — a property of the whole web of reference, not of any sentence in it — and a paradox is a web that cannot be grounded, whether it loops back in one step like the Liar or descends forever like Yablo. The diagonal's essence was never the loop. It was the shape.
the challenge · Yablo's sequence · the fixed point in the making · the shape of paradox · local and distributed · where the corpus stands
The collection has spent its length showing one construction beneath the great limitative results, and it has pictured that construction as self-reference in the most direct sense: the Liar speaks of its own falsity, Russell's set asks after its own membership, the halting machine is run on its own description. Every paradox the collection has met loops back on itself in a single, visible step. This had come to seem not a feature of the examples but the definition of the thing — to be paradoxical was to refer to oneself. Stephen Yablo's construction of 1993, pointedly titled Paradox without self-reference, was built to show that this definition is too narrow. It exhibits a paradox with no loop anywhere in it — no sentence that mentions itself, directly or through any finite chain — and so poses, more sharply than anything the collection has faced, the question of whether the diagonal really reaches every paradox, or only the ones that wear their circularity on the surface. The honest course is to take the threat at full strength before answering it.
Every paradox the collection has catalogued refers to itself in one visible step. Yablo's does not — and if a paradox can do without self-reference, the collection's central equation fails at its first term.
Lay out exactly what the collection has claimed. Its thesis runs paradox = self-reference = the diagonal: the limitative results are paradoxes of self-reference, and self-reference is the diagonal's fixed-point structure. The first equality has always seemed the safest — the Liar, Russell, the Grelling, the Curry, every paradox in the canon turns on a thing that refers to itself. Indeed the collection's diagonal is built to capture exactly that: the construction g(x) = α(φ(x)(x)) feeds x to itself, the formal heart of self-application.
So a paradox with no self-application at all would strike the thesis at its foundation. It would be a limitative phenomenon — a genuine paradox, a real contradiction extractable from the premises — that the diagonal, the engine of self-application, could not be the source of, because there would be no self-application to engage. Yablo claims to have built precisely this. To see whether he has, one must look at the construction with care, because everything turns on a distinction the collection has not yet had to draw: between a sentence that refers to itself, and a list that refers to itself while no sentence in it does.
The construction is a single uniform list, each sentence denying the truth of all that follow. A contradiction falls out — and yet, read literally, nothing in it ever points back at itself.
The contradiction is swift, and it falls on every sentence at once. Take any Yₙ and suppose it true. Then, by what it says, every later sentence is untrue — in particular Yₙ₊₁ is untrue, and also every sentence after Yₙ₊₁ is untrue. But "every sentence after Yₙ₊₁ is untrue" is exactly what Yₙ₊₁ asserts — so Yₙ₊₁ is true. We have derived that Yₙ₊₁ is both untrue and true: contradiction. So the supposition fails, and Yₙ is untrue — and this holds for every n, since n was arbitrary.
But now run the consequence. Every Yₙ is untrue; so for any fixed n, every sentence after Yₙ is untrue; but that is precisely what Yₙ asserts, so Yₙ is true after all. Contradiction again, and again for every n. There is no assignment of truth and untruth to the sentences of the list that escapes the squeeze: suppose any one true and it forces a successor both-ways; suppose all untrue and each becomes true. The list admits no consistent valuation. It is paradoxical in the strongest sense.
Look for the loop and it is not there. Y₀ talks about Y₁, Y₂, Y₃, …; Y₁ about Y₂, Y₃, …; each sentence about a tail strictly beyond it. No sentence mentions itself. Trace any sentence's references forward and you descend the list forever, never returning. There is no cycle of any length — not a one-step loop like the Liar, not a two-step "each calls the other," nothing. By the collection's working picture of self-reference as a thing naming itself, this paradox has no self-reference. And yet it is as paradoxical as the Liar. Either the thesis is broken, or the picture of self-reference is too small.
No sentence of the list refers to itself. But to bring the list into existence at all — to define what "Yₙ" means — one must reach for the diagonal lemma, the collection's own engine of self-reference, and apply it once to the whole.
The first answer is the one Graham Priest gave in 1997, and it relocates the self-reference without denying Yablo's observation. The Yablo sentences are not handed down from nowhere; to have them in arithmetic, where the paradox can be made precise, one must define a predicate that produces them. Write Y(x) for "the x-th Yablo sentence is true." Its meaning must satisfy
— the predicate Y appears on both sides, defined in terms of its own satisfaction. A formula with that property does not come for free; it is produced by Gödel's diagonal lemma, the very construction the collection has placed at the centre of every paradox, applied here once to manufacture the whole self-referential predicate. So a fixed point is used — not to make any single sentence loop, but to bring the entire list into being.
No sentence in the Yablo list refers to itself — but the list refers to itself. Its defining predicate is a fixed point of its own satisfaction condition, got by diagonalization. The circularity has not vanished; it has moved from the individual sentence up to the construction of the sequence as a whole. The self-reference is distributed across the list rather than localized in any member of it.
This is genuine, and it is the collection's first line of defence: the diagonal is present in Yablo after all, doing exactly its usual work, only at the level of the list rather than the sentence. But the collection's honesty discipline forbids resting here, because the fixed-point observation proves too much to be the whole story. As Leitgeb and Cook have pressed, every sentence of arithmetic is a weak fixed point of some predicate — apply the diagonal lemma to an arbitrary formula and you manufacture a fixed point on demand — yet almost no sentence is paradoxical. If being a fixed point were the mark of paradox, every sentence would be paradoxical. So "Yablo's predicate is a fixed point" cannot, by itself, be what makes Yablo paradoxical; the fixed point is necessary scaffolding but not the discriminating cause. To find what separates the paradoxical structures from the harmless ones, the collection must look past the fixed point to the shape the references make.
Draw an arrow from each sentence to the sentences it talks about and a reference graph appears. Paradox turns out to be a property of that graph's shape — and there are exactly two shapes known to produce it.
This is the move that resolves the matter, developed by Gaifman, Cook, Walicki, and Rabern, Rabern, and Macauley, and sharpened for arithmetic by Beringer and Schindler. Forget, for a moment, what the sentences say, and record only what they refer to: a directed graph with sentences for nodes and an arrow from each to those it speaks of. A truth-value assignment is an acceptable decoration of this graph — a labelling of each node true or false consistent with what its successors are labelled. And the definition of paradox becomes purely structural:
With paradox made a property of shape, the known results sort the shapes, and they vindicate the collection's instinct while widening it. Rabern, Rabern, and Macauley proved that an undirected graph admits a dangerous orientation if and only if it contains a cycle — so cyclic structure really is what underlies the ordinary, finite paradoxes, the Liar's one-step loop being the smallest. But they also proved that a dangerous graph with no cycle must have infinitely many nodes each referring to infinitely many others — which is exactly the shape of Yablo's graph. So there are two basic dangerous patterns, and only two are known:
A directed cycle in the reference graph — reference that loops back, in one step (the Liar) or several (the "no-no," the Curry chains). The self-reference is local: a closed loop you can point to. This is the shape of every finite paradox.
No cycle anywhere, but infinitely many sentences each referring to infinitely many later ones — reference that descends without end. The self-reference is distributed: no loop exists, yet the whole graph fails to ground. This is the shape of every known non-self-referential paradox.
The resolution is a revision of the collection's central concept. Self-reference is not a sentence pointing at itself; it is a reference structure that fails to ground. The loop was only the simplest way to fail.
Assemble the answer to Yablo's challenge. The collection pictured self-reference as local — a thing naming itself, a loop you can point to. Yablo shows that picture is too small: his paradox has no loop, yet it is a genuine paradox, because its reference graph fails to ground exactly as the Liar's does. What the two share is not a loop but ungroundedness — a reference structure with no consistent valuation, no resting place where truth values are fixed by fact. The Liar achieves this by looping back; Yablo achieves it by descending forever; both are ways of being non-well-founded. So the collection's first term was right but mis-described. Paradox is self-reference — where "self-reference" means not the local loop but the structural property of a reference web that refers back into itself, whether in one step or across an infinite descent.
The diagonal's essence was never locality. Its essence is the fixed point — the structural closure of reference upon itself — which can be realized locally, as the Liar's self-naming loop, or distributively, as Yablo's loopless infinite descent. Both are the diagonal: in the Liar, the fixed point is a single sentence; in Yablo, it is the predicate defining the whole list. The collection studies the fixed-point structure, and that structure does not require any one thing to name itself — only the web to close on itself. Self-reference is a property of the graph, not the node.
The collection's discipline requires marking where this defence stops, because the question is not closed by fiat. There is a residual, live dispute, and it should be stated at full strength. Roy Cook has constructed infinitary versions of the Yablo paradox — built from infinite conjunctions rather than arithmetic diagonalization — that contain, on careful analysis, no fixed points of any kind. If those count as genuine paradoxes, and if "self-reference" is read strictly enough that even the distributed, structural kind is absent from them, then a truly self-reference-free paradox exists, and the collection's thesis would face a real exception — exotic, infinitary, far from the canonical results, but real. Whether to grant this turns on contested choices: whether infinitary languages yield genuine paradoxes, and how strong a notion of self-reference one demands. The collection takes no triumphalist stand on a live research question. Its claim is the defensible one: for the paradoxes that matter to its argument — the Liar, Russell, Gödel, Tarski, the halting problem, Curry, and their kin, all formalizable in arithmetic — the diagonal's self-referential structure is present, locally or distributively, and is what makes them paradoxical. Whether some infinitary outlier escapes the net is open, and the collection says so.
Unlike the descents around it, this one does not shrink the diagonal. It enlarges it — showing the construction reaches even the paradox built to escape it, because its essence was always the structure and never the loop.
State the outcome in the collection's discipline. Yablo built a paradox to break the equation paradox = self-reference, and on the collection's own working picture — self-reference as a local loop — he succeeded: no sentence in his list names itself. The resolution is not to explain the loop away but to correct the picture. Self-reference is structural, not local: it is the closure of a reference web upon itself, the failure of its truth values to ground, and that closure can be realized as a one-step loop (the Liar) or a loopless infinite descent (Yablo). Both are the diagonal's fixed-point structure — localized in a sentence, or distributed across a list. So the equation holds, in the enlarged and correct reading, and Yablo's paradox is not a counterexample to it but its most instructive instance: the case that forced the collection to see that the diagonal was never about a thing naming itself, only about reference folding back into itself, by whatever route.
This sets the descent apart from those just before it. The Latitude, The Horizon, and The Coil each right-sized the diagonal — one of two roots, one of a family of forms, one of two faces — trading a claim of total reach for honesty. This descent moves the other way. It finds the diagonal larger than the collection's own local self-image supposed: present even where no sentence is self-referential, in the very paradox designed to exclude it, because its essence is the ungrounded structure and that structure does not need a loop. The collection had been picturing its own central object too narrowly. The web of reference, not the knot in any single thread, is where self-reference lives — and the diagonal is the theory of when that web fails to ground.
And the finding closes a circle with two earlier descents. With The Coil: a reference graph is paradoxical precisely when it is non-well-founded — looping or descending — and carries the falsity twist; circularity without the twist is grounded and harmless, circularity with it is paradox, exactly as the coil found at the level of sets, now found again at the level of reference. And with Kripke's grounding, which the collection can now place: to be grounded is for the reference web to bottom out in fact; to be paradoxical is for it to fail to, by cycling or by endless descent; and the diagonal is the structural signature of that failure. Local self-reference, distributed self-reference, non-well-founded sets, ungrounded truth — these are one phenomenon under four descriptions, and the collection is its theory.
So the collection stands, on this question, enlarged and exactly bounded at once. Its thesis survives the sharpest attack on its first term, in a form truer than the one attacked: paradox is self-reference, where self-reference is the structural closure of a reference web upon itself, realized locally or distributively, and the diagonal is the engine of that closure wherever it occurs. The honest boundary is drawn where the mathematics leaves it: the full characterization of which reference graphs are dangerous is an open problem, and whether some infinitary construction yields a paradox genuinely free of even distributed self-reference is a live and contested question. The collection claims neither the open theorem nor the contested outlier. It claims the core the attack was meant to break and could not — that the diagonal reaches every canonical paradox, the looping and the descending alike, because what it was always about was the shape of the web.
The loop was only the simplest way for reference to close on itself. Yablo found another — the endless descent — and it is the same closure, distributed. Self-reference is the shape of the web, not the knot in any thread.