Euclidean geometry is the worked example of what a formal system is: axioms chosen, theorems derived, objects idealized, reality approximated. Every move that broke it has an exact counterpart in logic — including the one nobody notices, that a proposition is as idealized as a point.
Links: Logic and Mathematics, The Absoluteness Claim, The Translation Problem, The Consistency Defeater, Aristotelian Logic, Non-Classical Logics, Gödel Against Himself, The SQL Argument, Theorems With Absurd Readings, Logics In Production, Where Concepts Live — uses this page’s always the approximation thesis as evidence that mathematics has a revision history, which an eternal structure should not have, What Is a Number?
Geometry is the best teaching instrument available for the claim that logic is a built system, and it beats every alternative for three reasons:
This page is the worked version. It exists to be used — the argument it supports lives in The Absoluteness Claim.
Four parts, and Euclid’s Elements (c. 300 BC) is the first complete instance in history:
| Part | In Euclid | In classical logic |
|---|---|---|
| Primitives | point, line, plane — undefined, taken as given | proposition, truth value, negation |
| Axioms/postulates | 5 postulates + 5 common notions | identity, non-contradiction, excluded middle (+ inference rules) |
| Inference rules | the constructions and deductions of Greek mathematics | modus ponens, substitution, etc. |
| Theorems | 465 propositions across 13 books | every tautology and valid form |
The critical property: the theorems are true of the system. They follow from the axioms with total rigour. Whether they are true of the world is a completely separate question that the system itself cannot answer — and that is the entire lesson.
The first four postulates are terse and obviously acceptable:
Then the fifth, which does not read like the others at all:
If a straight line falling on two straight lines makes the interior angles on the same side less than two right angles, the two straight lines, if produced indefinitely, meet on that side on which the angles are less than two right angles.
This is the detail that makes the story land, and it is not a modern reinterpretation:
The attempts continued unbroken until the 18th century, and the last of them is the best story in mathematics.
Girolamo Saccheri (1733), in Euclides ab omni naevo vindicatus (“Euclid Cleared of Every Flaw”), tried proof by contradiction: assume P5 is false, derive an absurdity, and P5 is vindicated. He assumed the negation and worked out theorem after theorem — and derived a large part of hyperbolic geometry. No contradiction ever came, because there isn’t one.
He had non-Euclidean geometry in his hands and rejected it, writing that the results were “repugnant to the nature of the straight line.” He could not find a contradiction, so he substituted an aesthetic judgment for one and published it as a proof.
This is the exact failure mode the absoluteness claim runs on. No contradiction is available, so intuitive repugnance is offered in its place. “I can’t imagine it being otherwise” gets promoted to “it cannot be otherwise.” Saccheri is what that looks like from the outside, with the verdict already in.
That last step is the one that matters logically, and it is usually skipped:
A model gives a relative consistency proof. If Euclidean geometry is consistent, then hyperbolic geometry is consistent — exactly as consistent, no more and no less. The alternative is not a sloppy approximation of the real thing. It is a peer.
This is the precise answer to “but the alternatives are just made-up.” So is Euclid. Both are axiom sets; each is consistent if the other is; the mathematics has no preference between them. Preference comes from application, which is empirical, which brings us to the verdict.
Kant made Euclidean geometry the centrepiece of the Critique of Pure Reason: space is the a priori form of outer intuition, and geometry is therefore synthetic a priori — substantive knowledge about the world, guaranteed in advance of any experience, unrevisable by observation. It is the strongest claim of absoluteness ever made for a formal system by a serious philosopher.
In 1915 General Relativity made spacetime curvature the mechanism of gravity, and in 1919 Eddington measured it. The geometry of the universe we live in is not Euclidean. Kant’s paradigm case of unrevisable a priori truth was revised by observation.
Note carefully what did not happen. Euclidean geometry was not refuted, discarded, or shown to be nonsense. Every theorem still follows from the axioms; every draughtsman and machinist still uses it; it remains one of the great achievements of the species. What was refuted was the claim that it is the geometry, absolute and unrevisable. It turned out to be an excellent local approximation whose axioms hold to superb precision at human scale and fail where gravity is strong.
That is the entire shape of the argument about logic, already completed once, in a case nobody disputes.
The geometry story has a second half that is usually treated as a separate point but is really the same one, and it is the half that transfers to logic in a way almost nobody notices.
Consider what geometry’s primitives actually are:
So geometry is not a description of physical space. It is a system of idealized objects that physical space approximates well enough to be useful. A pencil dot approximates a point. A taut string approximates a line. The theorems transfer to the world exactly as far as the approximation holds, and no further. (The vault’s Geometry Test uses this against Platonism; here it is doing different work.)
Nobody finds this troubling, and nobody says “the pencil dot proves geometry is false.” The relationship between idealization and instance is understood, accepted, and unremarkable.
Classical logic’s primitive is the proposition: an item bearing exactly one determinate truth value, context-independent, unambiguous, with sharp application conditions.
Nothing anybody ever says is one of those.
| Geometry | Logic | |
|---|---|---|
| Idealized primitive | point: zero extension | proposition: exactly one determinate, context-free truth value |
| Physical instance | pencil dot, taut string | a sentence uttered by a person |
| The gap | nothing has zero width | no sentence has one determinate context-free truth value |
| Sources of the gap | thickness, granularity, measurement error | ambiguity, homonymy, vagueness, indexicals, scope, tense, future contingents, speaker intent |
| What bridges it | measurement and tolerance | interpretation and judgment |
| Where it works | superbly, at human scale | superbly, for most arguments |
| Where it breaks | strong gravity, cosmological scale | vague predicates, unsettled futures, self-reference, translation across languages |
This is what the “loose connection between language and propositions” is. It is not a quirk of sloppy speech to be tidied up. It is the same category of gap as the one between a pencil dot and a point — the unavoidable distance between an idealized formal object and the physical situations it models. The Translation Problem catalogues the specific failures; this page names what kind of thing they are.
And it makes the pre-logical layer respectable rather than merely awkward. Measurement is not a defect of geometry; it is the discipline that connects geometry to the world, and engineers who ignore tolerances kill people. Interpretation is not a defect of logic; it is the discipline that connects logic to the world, and debaters who ignore it argue past each other all night.
Here is why the second lesson is not a side note. Put the two halves together and the absoluteness claim has to choose:
Either logic is an idealization that reality approximates — in which case it is a human construction of exactly the kind geometry is, its objects do not exist in nature, and the interesting question is how good the approximation is in a given domain. That is the vault’s position, and it needs no divine ground, any more than Euclid’s postulates do.
Or logic is a set of absolute laws that reality obeys — in which case vague predicates, unsettled futures, and the permutation-invariance of identical particles are not approximation gaps. They are counterexamples, and they have to be answered rather than tolerated.
The absoluteness claim habitually takes the first horn to explain away the hard cases (“of course natural language is loose, that’s not logic’s fault”) and the second horn to reach God (“the laws are universal, invariant, immaterial”). Those are different claims and they are not compatible. The looseness is either logic’s problem or reality’s. Pick one.
A law that reality obeys has no need of tolerances. A system with tolerances is not a law reality obeys.
Expect this, because it is the best defence of the absoluteness claim available and it is not a dodge:
“Even if we get the language wrong, there is still an underlying truth, and it’s driven by logic. Our sloppiness in encoding a proposition doesn’t touch the proposition, or the logical relations it stands in. You’ve found a flaw in us, not in logic.”
Concede the true part immediately, and concede it generously, because the page overclaims if it doesn’t: the translation problem does not show that logic is invalid. Every inference rule still holds. Nobody’s misuse of the word “bat” refutes modus ponens. If the argument is heard as “language is vague, therefore logic is broken,” it deserves to lose.
What the reply does is take a specific, expensive position. Read what it commits him to: a determinate realm of propositions, each with a determinate truth value, standing in determinate logical relations, existing independently of any language that expresses them. That is propositional realism — and it has to be paid for.
The reply models the gap as encoding noise: reality is sharp, our words are a lossy channel, clean up the channel and the sharpness is recovered. That is exactly right for ambiguity. “She saw a bat” really does express one of two perfectly sharp propositions, and context picks it out. Noise, recoverable.
But the hard cases are not noise. They are cases where there is no sharp scene to photograph, and a noiseless channel does not help when the source itself is indeterminate:
So the reply must either accept that some propositions lack determinate truth values — which is the concession — or insist that all three cases secretly have sharp answers nobody can access. That second option is epistemicism, and taken across the board it is a very large bill: an unknowable precise boundary for every vague predicate in every language, a settled fact about every unmade choice, and a determinate answer to CH. None of that is a claim about language. It is a claim about reality, offered without evidence, to protect a rule.
This is the part worth having ready, because it is structural rather than a matter of intuition.
“There is still underlying truth driven by logic” is the second horn of the dilemma above: logic as a law reality obeys, not an idealization reality approximates. He is entitled to that horn. But taking it costs him the move he was using two minutes earlier:
He can no longer explain away the hard cases as “language being loose.” Vagueness, unsettled futures, and independent propositions are no longer tolerable approximation gaps between an ideal system and a messy world. On the horn he just chose, they are counterexamples to a law — and they have to be answered, not excused.
The absoluteness claim needs both horns and it can only stand on one. It uses “logic is an idealization, of course reality is messier” to absorb every awkward case, and “logic is universal, invariant, immaterial” to reach God. Once you make the choice explicit, one of those two sentences has to go.
The last move is his own criterion, applied where he is now standing. He cannot exhibit a single language-independent proposition — every one he can point at arrives dressed in a sentence, in some language, including the formal ones. The determinate realm is posited to protect the rule, then the rule is used to prove the realm’s ground.
So: what accounts for the determinacy? If the answer is God, then the argument’s engine has become its own conclusion, and we are at the circle presuppositionalism openly accepts. That is not a knockdown — but it is parity, and parity is fatal to an argument whose distinctive claim was that only theism can account for its preconditions. See The Consistency Defeater, Reply A.
Note the pattern, because it is the most useful thing on this page. Every strong reply available to him — “consistency is guaranteed by God’s nature,” “the underlying propositions are determinate,” “the deep structure all logics converge on is what needs grounding” — funnels into the same place: God grounds it. That answer is coherent and he will not be embarrassed by it. But once every road leads there, the argument has stopped being a proof and become a statement of the commitment it started from. The work is then to show the road is open to everyone, not to close it.
Not “logic is wrong.” This:
Logic is a formalism over idealized objects. Getting from the world to those objects is an interpretive act that logic cannot perform and does not govern. Therefore logic does not directly describe reality — it describes reality as we have rendered it, and the rendering is ours.
That is a limit on premise 1’s reach, not on logic’s validity. Premise 1 needs the laws to be absolute features of reality itself. What the idealization argument shows is that they are absolute features of a system we built to model reality — which is exactly the status of Euclid’s postulates, and exactly as much divine grounding as those require.
The theological form of the question is worth stating, because it is the same question as which logic is God’s and it is much harder to wriggle out of when it is about geometry.
If Euclidean geometry were divinely grounded — universal, invariant, immaterial, the very structure of the divine mind — then God built a universe that does not satisfy it. Spacetime is curved. The available replies map one-to-one onto the replies about logic: