The Shape of Hard-to-Vary
In a recent thread, David Deutsch affirmed a sharp claim: the gold standard genuinely depends on a lack of progress in physics. He is right. Every hard-money system, Bitcoin included, rests on the current boundary of physical and computational knowledge holding. Break the boundary, dissolve the asymmetry, and the structure stops being hard-to-vary.
This is the failure mode that proves the framework. And it generalises. Every hard-to-vary structure I can identify lives in the same geometric shape, and dissolves through the same mechanism.
The conjecture is that knowledge has one shape, and that shape is Gabriel’s horn. The fallibilism is built into the geometry.
The two physical facts that force the shape
The first is Landauer’s principle. Every bit stored, computed, or erased pays at least k_B T ln 2 in heat. Information is physical. Verification is never free.
The second is verification asymmetry. For at least some operations, generating a valid configuration is super-polynomial while checking it is polynomial. The strong form is P ≠ NP. The operational form is the preimage-resistance class that powers cryptography, mathematical proof, and the error-correction protocols of biology.
Neither premise needs to be assumed in its strongest version for the geometry to follow.
The forcing argument
A persistent structure that survives by being checked needs Cf/Cp >> 1, which means the cost of faking it must exceed the cost of confirming it. Otherwise it is not operationally hard-to-vary. It dissolves under perturbation.
Let *I* be the bit-content of the structure’s specification, the smallest description a verifier accepts. Let *B(t)* be the integrated verification surface: every check, by every verifier, across all of time.
For *Cp* to stay small, *I* must be bounded. Otherwise verification cost scales in |I| times every use ever, and the structure becomes unsustainable.
For *Cf* to stay large, the structural difficulty must not live inside *I*. If it did, falsification would reduce to copying. The hardness has to be embedded in *B*, in the failed attempts and in the distributed asymmetry the verifier set enforces.
The consequence is that *I* stays bounded while *B* grows without bound, asymmetric in the same direction wherever the structure exists. The geometric form is Gabriel’s horn.
Where the geometry shows up
Bitcoin’s interior is about 1.5 MB per block, fixed. Its verification surface includes tens of thousands of nodes still running checks, every block ever built on the chain, and the cumulative hash work committed to its defense. The I/B trajectory collapses to zero. Measured Cf/Cp at 100-block depth runs about 95×.
Gold’s interior is 79 protons per atom, an arithmetic fact. Its verification surface is every assay and every refining operation in human history.
The human genome carries roughly 10⁸ functional bits of interior. Its verification surface is four billion years of cell division and reproduction.
Newton’s laws have an interior of a few pages of math. Their verification surface is three centuries of unbounded application.
Each lives in horn geometry. Each is hard-to-vary precisely because the asymmetry beneath it holds. The shape recurs because the forcing recurs.
The failure mode is in the geometry
Now the Deutschian point, made geometrically. The horn persists only as long as the asymmetry beneath it holds. When growing knowledge breaks the asymmetry, the structure dissolves.
Gold’s failure mode is cheap transmutation. If chemistry had cracked it in 1880, gold’s monetary horn would have collapsed by 1900.
Bitcoin’s failure mode is a constructive proof of P = NP. Quantum attacks on specific cryptographic primitives are real engineering problems, but they admit known patches because the underlying verification asymmetry still holds. A constructive P = NP is different. It dissolves the asymmetry itself, and the horn cannot be patched against the dissolution of its own load-bearing premise.
The human genome’s failure mode is sufficiently cheap synthetic gene editing, a falsifier that can fake any specification more cheaply than evolution can build one. We are approaching that boundary now.
Newton’s laws pinched at the edges. Relativity and quantum mechanics seeded new horns at scales where the original asymmetry weakened, while the Newtonian horn remained intact within its domain of applicability.
The fallibilism is in the shape. The horn has a built-in dissolution condition. It survives as long as growing knowledge respects the asymmetry it rests on, and dissolves the moment it does not.
This is what Deutsch is pointing at when he says the future growth of knowledge is unknowable, including its probabilities. You cannot bet on any specific horn surviving the unknowable. You can only note the geometric form they share while they last, and the universal failure mode they share when they do not.
The conjecture in constructor-theoretic form
If constructor theory carves possible tasks from impossible, and if “persist by being checked” is itself a class of task, then the possible-task manifold for persistence-by-verification has Gabriel’s-horn geometry as its attractor. Structures off the manifold dissipate immediately. Structures on it ratchet for as long as the asymmetry holds. When knowledge growth dissolves the asymmetry, the task moves from possible to impossible, and the structure dissolves.
This is a stronger claim than K = Ic², the measure of constraint quality I have been working with. K = Ic² counts the constraint quality of any candidate structure. The horn conjecture says the universe permits only one geometric channel for memory, and tells you exactly where it fails.
What I have, and what I don’t
I have the asymmetry forced: *I/B* → 0 as *t* → ∞ for any persistent verifier-checked structure. That is the floor, and I believe it is provable cleanly under Landauer plus persistence alone, without needing P ≠ NP in its strongest form.
I do not yet have Gabriel’s horn specifically (the slow-divergent surface, the 1/x form) derived from a clean variational principle. That is the work in progress. The weaker theorem on asymmetric geometry is provable. The stronger theorem on the specific horn shape is what I want falsified before I publish it.
The question
Does constructor theory predict this geometry? Or can you construct a counter-task, a persistent verifier-checked structure whose *I* and *B* grow together with the same hard-to-vary character, that breaks the shape?
If you can produce one, the conjecture dies and I learn something important.
If you can’t, then hard-money systems are not anti-fallibilist. They are the shape fallibilism takes when something persists. The universe has one channel for memory. We can finally see what it looks like, and where it breaks.
Post a comment: