What Cannot Be False Cannot Be True

Truth is buildable. The unbuilt is untrue.

The epistemology behind The Hypothesis Graph, and the frame its companion paper Verifiable Knowledge runs as a protocol. The argument was first stated informally across Truth Is Buildable (2026-06-04) and Belief Is the Edge of Knowing; this is the consolidated, citable version.

Abstract

A binary mode of truth cannot represent knowledge in its becoming. Before knowing, uncertainty for an agent demands an expression, but a certain null is ignorance. Posited under uncertainty, a unit of knowledge presents itself subjectively as phenomenon; it is then tested to either true or false. This uncertain status is untrue, not yet false nor true, and terminally so if unfalsifiable. Bivalence holds only where a verdict has been laid down, unlike the untested ground before it. Reasoning relates it to its priors, making it a graph; this yields connectivity and buildability through verification. A claim that cannot fail says nothing, so what cannot be false cannot be true. Ramsey proposed actionable truth; Popper asserted its admission; Peirce codified its inquiry.

1. Introduction

An AI now makes claims that pass for knowledge. It answers a question, fixes a defect, reports a number. Each time, it asserts the result confidently, sometimes hallucinating. You can see why it cannot know for certain. But can you know what you know for certain?

Classical epistemology takes the knower to be a person and the prize to be certainty. Knowledge is justified true belief (the lineage from Plato to its breakage in Gettier 1963): a belief that is true, and justified, where justification is a state of the believer’s mind and truth is correspondence to a world the believer cannot step outside of to check. As a description of what a human cognizer is doing this has its merits and its long list of problems. As a procedure, it is underspecified: it never says how to inspect the justification or replay the correspondence. No agent can read off whether its belief corresponds to a world it only ever meets through its own representations, and no agent can certify the inner justification of another.

This paper builds the alternative: an epistemology whose every stage is a structure an agent can construct and could see broken. Call the territory agent-native epistemics, the question of what it is for an AI rather than a person to hold an entitled claim. It specifies belief, knowledge, and truth as things an agent does, and redraws the trichotomy so the third state is an entitlement an agent can check rather than a truth value carried inside the claim. That redrawing is the contribution, and the boundary it draws is what cannot be false cannot be true.

The home tradition is pragmatism, and here the pragmatist program is carried to a knower it never had in view. The pragmatists made the decisive moves already: truth is what survives inquiry rather than what corresponds, belief is a disposition to act whose strength is the odds you would take, certainty a demand to drop rather than a prize to chase. What they could not finish, lacking the knower, is make the program run: a human cannot expose the inner state where their justification is supposed to live. The rest of the tradition is recruited where pragmatism left a part unspecified, and §8 names each borrowing; no single one of them is the point here; it is that, assembled in pragmatist order and read as a build, they compose into an epistemology an AI could run. This paper sets the frame, what truth is for such a knower, climbing from the phenomenon/noumenon boundary through belief, knowledge, and truth to empirical truth as a graded belief asymptotic to a noumenon it never reaches, and a disjoint graph where formal truth is decided absolutely against its axioms, a different type of entitlement entirely.

2. Phenomenon and noumenon

Everything that follows runs on a single boundary. There are two things one might mean by truth, and they behave so differently that holding them apart is the precondition for the rest.

Kant drew the line. The noumenon is the thing in itself, the world as it is independent of any knower, which a knower can think but never hold. The phenomenon is the appearance, the world as a build constitutes it for a knower who only ever meets it through representation. Every cognizer, human or machine, works the phenomenon. The noumenon is not a further region one might reach with a better instrument; it is what no instrument reaches by definition, because an instrument returns a representation and the noumenon is what representation is of.

The familiar shorthand is the map and the territory, and as a first lesson it serves: beliefs are not the world. What its imagery leaves out is the knower’s position. Not a surveyor with the ground spread out before them, the knower is inside it, subject and object at once, perceiving and perceived. Ants studied for their social behavior never perceive the study, and we will never know what examination, if any, we are under. The noumenon names that un-knowing, the frame that may hold us with no perceiving out to it.

Truth splits along this line. Noumenal truth is correspondence to the world in itself. Modeled as correspondence it is classically bivalent, a determinate fact true or false whether or not anyone can build an entitlement for it, but that bivalence is a property of the model, not a standard anyone can consult. There are exactly as many atoms in the universe as there are, and whether that count is even or odd is settled out there past any reach of ours. Noumenal truth is bivalent in the model and untouchable in the same breath, and the second property is what makes it useless as a working standard. You cannot run a claim against the world in itself, because every test you run returns a phenomenon.

Phenomenal truth is the one a knower works: in contrast to correspondence with the unreachable, a build that has been exposed to a test and is presently standing. This is the truth the rest of the paper specifies, and naming it phenomenal is what licenses the word truth for something less than correspondence-forever without thereby demoting it to mere belief. It is still truth, because it is still answerable to the world; it is phenomenal truth, because the world answers only by refuting builds, never by showing its face.

That last clause is the one place this paper departs from Kant, and it is what keeps the view from collapsing into either relativism or idealism. The noumenon never shows itself, but it is not inert. It pushes back. When a build is wrong, the world refutes it: the bridge falls, the program crashes, the prediction misses, the counterexample arrives. We never see the world in itself, but a wrong build runs into a brute outward clash, one bit of constraint that says this build is wrong. That clash is Peirce’s Secondness, conceded as his; the failed build is a phenomenal event constrained by reality, not a signal received from the thing in itself. What is the paper’s own is only the machine version, one bit, replayable. That single bit of contact is the whole of the leash that ties phenomenal truth to something outside the knower. A view on which the world could not refute a build would be idealism, truth manufactured by the builder. A view on which the knower could see the world in itself would be naive realism, the correspondence nobody can run. Phenomenal truth is the load-bearing middle: built by the knower, broken by the world, standing only on sufferance.

With the boundary in place, the rest of the paper stays on the near side of it. When it says truth without qualification from here on, it means the phenomenal kind, the build that could have fallen and hasn’t. The noumenon returns only to do its one job, which is to break things. And because it only ever breaks and never builds, the truths a knower holds can approach it without ever arriving, a shape the paper makes good on once the stages are in place (§5).

3. Belief

A knower never holds the world, only the build that stands in for it. So even the surest claim is a bet placed, not a fact read off, and that is where the climb begins. The arc has stages, belief then knowledge then truth, but they are degrees of one bet rather than jumps to a different kind of thing. Belief is the ground floor, and for a claim about the world there is no tier above it.

Suppose I say that I know my keys are in my pocket. The word know is doing no philosophical work there. It means I am confident enough to reach in without bracing for absence, and if the keys are gone I update without fuss. The knowledge was belief past a threshold the whole time. Raise the stakes and the threshold rises with it: stake a life on the answer and I downgrade to let me check. Nothing categorical sits above the confidence. There is only confidence, varying by degree, and a line for calling it knowledge that the stakes draw and redraw. The picture that puts knowledge in a tier of its own, justified true belief marked off from mere belief, has the architecture backwards: it builds an upper story that nothing stands on, and the upper story is the first thing to break under contradiction.

Ramsey put this operationally. A belief is what you would bet on, and its strength is the odds you would take. That strength is what makes it actionable, and the threshold for promoting it to knowledge is the odds at which you would act, given the stakes. Clicking the least-lethal covered square in Minesweeper is the same act, a bet placed on the odds rather than a thing known apart from them. This is the whole of the distinction, and it is continuous, never a jump to a different kind of thing. There is one regime where the climb does reach a top, the formal one, where a proof settles a claim against its axioms with nothing left to bet on. A type boundary keeps that formal ceiling from ever lowering onto a claim about the world (§6), so empirical belief stays graded all the way up.

Why does the ceiling bind a machine, and not just a person? That is the next move, and the reason §2 was drawn for cognizers in general. All cognition runs on lossy projection: retina to spike trains, characters to vectors, every modality stripping dimensions with no exit to a view that does not. Plato’s prisoners had the shape right; the modern update only drops the exit, no escape from the cave, just projections, some more useful than others. So there is no world-as-such available to a knower, only world-as-projected, and a claim that grades itself against ground truth grades a fiction it drew itself. The scope has to be set carefully, or it saws off the leash §2 hung the view on. It does not say no constraint exists. It says no absolute ground truth is available to hold up beside a claim and read the match off directly. Reality still refutes a wrong build; you simply never get to inspect the world in itself to confirm a right one, which is why entitlement has to be built rather than read.

4. Knowledge

Knowledge, then, is a derived predicate rather than a separate substance: belief past a stakes-dependent action threshold, indexed to a context. The same belief is knowledge at low stakes and mere belief at high, the keys in the pocket promoted or demoted by nothing but what rides on being wrong. What the promotion buys is not certainty but exposure. To call a belief knowledge is to say you will act on it, and acting on it is what hands the world a chance to break it. The pragmatists fixed truth to action for exactly this reason: a claim works or it does not, and warranted assertion is the assertion you have earned the right to act on (James 1907; Dewey 1929).

A skeptic stops the climb here. You said knowledge is belief you act on, but I can never be sure the belief is right, so by your own lights I never have knowledge at all. The demand smuggled in is for certainty, and certainty was never the test. Make it the test and watch the cost: if I know only what I am sure of, and can never be sure of anything anyone tells me, then I know nothing anyone tells me, and every word of that sentence was inherited from people I would have to know to doubt them. Global skepticism spends the credit it says does not exist. So drop the demand, the oldest pragmatist move there is: real doubt has a motive and a target, paper doubt has neither. You were never certain. You were exposed, and exposure is enough.

The split that makes this precise is between a claim about the world and a claim about your entitlement for it. Suppose I tell you I know the exact number of atoms in the universe. Call it N. Two claims hide in the one sentence. The number is N is about the world. I know it is about me. The world-claim is contingent, so it could be false, and my N almost surely misses, but nothing can show it false because no one can run the count. It might by luck be exactly right and still not be knowledge, correct and unbuildable at once. This is the noumenal truth of §2 doing precisely what it does: determinate out there, bivalent only in the correspondence model, and useless as a standard anyone can consult. The knowledge-claim is the one you can test, and not by counting atoms. You demand the build, the chain, a source that reaches a root you can check, and there is none, so I know N is false, refuted at the level of provenance while the world-claim under it stays out of reach. You could not check the number. You could check that I never built an entitlement for it. Knowing is a build, and a build either shows its chain or it does not.

5. Truth

Exposure is what knowledge buys, and truth is what it can earn, a subjective label more than an objective property: at the top the claim takes on no new substance, only the name the one graded bet wears once a test it could have failed leaves it standing. A build you acted on and the world did not break is the one presently standing, and that standing build is the top of the arc. What separates it from belief is the one thing belief never carries on its own, exposure: a test the claim could have failed and did not. That fixes an order. The capacity to be false comes first, and the right to say true is earned only after it, never before. The N-atoms case (§4) ran this order in miniature: luck runs no risk, so the lucky-correct number is never knowledge, while the claim to know it stuck its neck out. Only the claim that could have failed is on the climb at all. A claim that runs no risk of being wrong has bought its safety by emptying itself, and so it never reaches the question of truth at all. And cannot be false means just this, that no test could expose it as false. It does not mean the world has rendered it immune; the noumenal fact, true or false, stays out of reach either way. What cannot be false cannot be true.

There is exactly one place the line does not hold. It governs empirical claims about the world, where the uncheckable number and the tautology both earn nothing, each immune to falsehood and therefore barred from truth. It does not govern the formal regime, where a proof closes entitlement without ever exposing the claim to the world, and the tautology that says nothing about the world is the whole point rather than the defect (§6). So the equation lands on world-claims, and the formal exception is held aside rather than waved away. The move from cannot be false to cannot be true is the one the verificationists made, Dummett gave its rigorous modern form by tying truth to what can be verified and withholding bivalence from the rest, and Popper circled without quite making.

Once reached, truth is not flat. It is a gradient, and belief intensifies along it by degrees. A hypothesis begins at the foot, deduced and waiting on a test it has not yet faced, untrue but already in motion. A standing result is the first to hold, a build that has survived its first real trials. A fact lies far along it, a build run so often that the test is retired and the claim treated as if it cannot fall, though it never stops being breakable. Beyond even that, the claim proven absolutely against its axioms does not lie on this gradient at all; it belongs to the formal regime, which runs its own non-graded passage from conjecture to proof, and importing that certainty here would be exactly the type mismatch the paper is about to rule out (§6). The empirical gradient has no end. The build is the climb, and the grade is how much the claim has survived.

One guardrail keeps buildable from sliding into manufacture-to-spec. Buildable does not mean a claim is true because you built it. The build has to include a test that can fail, and a build whose test can never fail is a hardcoded return value, return 0.70, a mocked test reporting green, not a true claim at all. This is the single-knower version of the reality leash from §2: the world has to be able to break it. The motivating case sits right here. A benchmark whose patches are withheld reports a confidence no one can collect, a number that could never have failed because the test it claims to have run is the test it will not show. It is not a wrong measurement. It is not a measurement. And because every claim stands trial forever, even the facts near the top hold their place only on sufferance, belief climbing and slipping by degrees as the trials keep coming.

What inquiry keeps approaching is the noumenon, the relation asymptotic. Empirical truth never reaches correspondence-forever, but the sequence of standing builds approaches it under inquiry, each better-placed than the last, the limit drawn by the succession and touched by no member of it. This is not a borrowed patch on the pragmatist core; it is the core, Peirce’s convergence in one line: truth is the opinion fated to be ultimately agreed to by all who investigate, and the real is what that opinion represents. The scope is Peirce’s own: convergence is a regulative ideal, his word for it hope, not a theorem that inquiry must converge (Misak 1991). This also resolves the asymmetry left hanging in §2, where the world only ever refutes a build and never confirms one. The one-sidedness is not a defect; it is the gradient that drives the climb, each refutation a confirmed prune and the survivors converging on a limit the curve never touches. True stays uncertified precisely because a knower is never at the limit, only on the way to it. This is fallibilism with a geometry.

Two derivations of the one limit meet here. The first is diachronic, Peirce’s: a community of inquiry approaching its fated agreement across time. The second is synchronic, a single knower now: subjective truth is capped at a graded belief, a credence strictly below one on any contingent claim, and with no tier above belief there is no state where the grade reaches one and the claim turns into certified correspondence. Watch the ceiling bite: however many sunrises you have logged, your credence that the sun rises tomorrow climbs toward one and never arrives, because one more morning is always a test it could fail. The limit, credence one, is the impossible occupied-correspondence state, unreachable for the same reason the upper story was empty: there is nothing above belief to climb into. Both derivations say the limit is approached and never occupied, and together they weld the two halves of the frame into one sentence. Subjective truth is a graded belief asymptotic to the noumenon.

Take the hardest case, the one domain a critic would call plainly settled: physics. The most closely studied empirical knowledge there is, is still asymptotic. Newton stood until Mercury’s perihelion broke it and gave way to general relativity, which itself strains where it meets the quantum, a successor still pending. Each was a better-standing build, none the final correspondence, and the succession itself is the asymptote drawn in history. Newton was a fact, then rescoped to a bounded domain where it stays approximately true and never unbreakable. If even physics only ever holds a graded belief approaching the noumenon, then everything softer does all the more, and the objection that settled science is simply true answers itself: a retired test means you have climbed far up the curve, without stepping off it onto the limit.

6. Disjoint graphs

There is one regime where none of that holds, and marking it off is what closes the frame here. Mathematics is the detached case. Empirical truth answers to the world and is asymptotic to the physical noumenon, which refutes it from outside. Mathematical truth answers to nothing outside its axioms, and self-consistency is the whole of its tester: no world-signal refutes a theorem, only a counterexample or an inconsistency does, and both of those are internal to the system. So the asymptote headline is scoped to empirical truth, and formal truth is buildable but detached. The detachment is not a flaw in it. It is what makes mathematics the cleanest case for the trichotomy to come, the one place where the test is internal and, where it is decidable, decisive, with no noumenal remainder to muddy the verdict. Truth-by-construction was the intuitionists’ insight a century ago, and it is the detachment cashed out: a claim is neither true nor false in this regime until it is constructed or refuted.

So the architecture is two disjoint hypothesis graphs. The empirical graph holds truth that is graded and asymptotic, credence below one, the world refuting, no node ever absolute. The platonic graph holds truth that is decisive within a stipulated formal system where proof is available, truth without grade, entitlement complete relative to its axioms and reachable precisely because the regime has no external noumenon, the system being closed. Where a proof or a refutation exists the entitlement closes internally; otherwise the node stays open relative to the system, since many statements are undecidable within it, and incompleteness gives the platonic graph its own permanently-open region rather than threatening the absoluteness of what is proven. Two cautions keep the word absolute from overreaching. The absoluteness belongs to the standard; no checker carries it: Gödel’s second theorem forbids a system from certifying its own consistency, so there is no complete tester, and a Bayesian checking a long proof never quite reaches credence one on having checked it correctly. Checking a proof is itself a replayable build, which folds that worry into the paper’s own vocabulary rather than leaving it outside.

Two disjoint hypothesis graphs drawn as transposes: the same graph, in the same layout, side by side. The left graph, labeled empirical, has hollow-ring nodes and one node tethered upward by an experiment edge to a short 'world' mark, with an arrowhead where the world strikes it, the place a build can be broken. The right graph, labeled formal, is the identical network sealed, with no experiment edge and its nodes filled solid, decided where proof is available. A faint vertical line labeled type boundary divides them.
Figure. The two graphs as transposes, the same graph read two ways. The empirical graph (left) is tethered to the world by an experiment edge, the one place a build meets a test that can break it, and its nodes stay hollow because no empirical claim is ever absolute. The formal graph (right) is the identical structure sealed: no edge reaches the world, nothing outside can break it, and its nodes fill solid, decided absolutely where proof is available. The wall between them is a boundary of type.

The boundary between the two graphs is the load-bearing part, and it is a boundary of type: absolute truth in the platonic graph is a different kind of entitlement from anything the empirical graph can hold. One disclaimer travels with the name. Platonic here names the regime’s behavior, absolute and detached, and nothing more; truth in it is relative to the chosen axioms, internal to a stipulated game, and the absoluteness is just the proof closing the gap inside that game. No commitment to mathematical Platonism is being made or needed. The name is kept because it is evocative.

The two kinds never convert, and the reason is where each is earned. Empirical entitlement is bought by surviving a world-facing test; platonic entitlement is a proof closing against its axioms, with no world in the loop. So a mathematical model of the world is an empirical hypothesis, an open node still awaiting its test, and its platonic self-consistency is entitlement in the platonic graph, never in the empirical one. Consistency cannot substitute for reality, and spending the former as if it were the latter is the characteristic error, a type mismatch: formal entitlement passed off where only the empirical kind counts. It is the exact dual of the withheld benchmark, which hides an empirical test it did run, where this lacks one and borrows a platonic substitute. That is the wall, and Einstein put it best: as far as the laws of mathematics refer to reality, they are not certain, and as far as they are certain, they do not refer to reality.

One objection presses here. If the platonic graph is sealed off, why is mathematics so unreasonably effective at physics? Because formal structure is what empirical hypotheses are built from, and the entitlement still comes from the test: the structures we call effective are the ones whose builds have stood. That the world is compressible enough for any of this to work is itself an empirical fact, an open node like the rest, not a crossing of the wall.

String theory exercises the boundary at once. Run it through the machine one move at a time. It is empirically untrue, with no passing world-facing build and no refutation either, so neither true nor false. On the gradient it is untrue and stalled rather than untrue and in motion: a healthy conjecture says I am untrue, here is the test, and string theory’s test sits past anything we can probe, out at the Planck scale. The charge its own critics press is sharper, that it is not even falsifiable: the landscape of some ten-to-the-five-hundred vacua can accommodate nearly any data, so nothing breaks it, which is to say nothing could refute it. That single fact fires several of the paper’s verdicts at once: it is irrefutable and therefore useless and therefore edgeless, a platonic structure with no edge to reality presented as physics. It cannot be false, so it cannot be phenomenally true. And it lands on the floor below false, the not even wrong that Pauli named and Woit took for a title, unable to earn even the dignity of being wrong. The decades-long argument reduces, on this account, to a single question: can anything refute it? Yes-but-unreachable makes it a stalled but legitimate open node. No makes it a detached non-hypothesis dressed as physics. The framework does not adjudicate the physics. It names the one question that decides the matter.

String theory is both the witness for what cannot be false cannot be true and the proof that the line needs the two-graph boundary under it. Popper stops at unscientific; the step on to cannot be true is the verificationist move he resisted as anti-realist. The boundary rescues that step by scoping it to phenomenal truth: with no failure possible there is no entitlement anyone could ever build, while the claim might still be noumenally true, the universe simply stringy in a way forever beyond reach. The line denies the buildable entitlement while leaving the fact of the matter untouched, splitting the verdict Popper and the verificationists fought over and handing each side the half it had right. Neither could state that alone. The charge is made charitably and dated on purpose: the problem is epistemic; it is no failure of the work’s rigor. Depth is not the bottleneck, a reachable refutation is, and even the most carefully built program, with no falsifier within reach, stalls. The verdict holds only from how we see it today, itself a dated, refutable standing build; let a swampland-style falsifier finally refute or confirm it, and the verdict updates. The error the framework names is never the ambition. It is presenting the platonic build as the empirical result, or quietly dropping that the test is out of reach.

With the wall in place the frame is complete. Kant drew the boundary, Peirce drew the asymptote, Einstein drew the wall between the two graphs. The paper’s own additions to the frame are two: the noumenon breaks things, and entitlement splits into two types. What the two regimes imply together is a ledger of three states, and naming them is the last move of the frame.

7. The trichotomy

Let’s name the three. An edge that can go red is what gives a claim a verdict at all, and the verdict comes back one of three ways, so the three states record which. Built and stood is true. Built and broke is false. No passing build is untrue. True and false are not opposites but siblings, split only by how the test came out; untrue is the one with no test, neither standing nor broken. Untrue is the zero of an entitlement that only accumulates and never inverts: false is not its negative but a verdict of its own, a build that met the test and broke, while untrue is the plain absence of any verdict at all. This is bookkeeping of entitlement, not a new logic, and the books are not subjective: the grade of belief is the knower’s, but whether a claim carries a passing build is not a matter of judgment. Consider deciding the truthiness of an event based on which channel it aired on. It either happened or it didn’t. The three are entitlement states, not truth values, and bivalence is not denied but housed: absolute bivalent truth lives in the platonic graph where proof decides (§6), the empirical graph stays graded and asymptotic, and the trichotomy sits across both, tracking what each claim has earned rather than assigning a third value inside the claim language. The third value is a century old, Łukasiewicz and Post. The narrow delta is where the third state lives: in the ledger of entitlement, read as no-passing-build, never as a logical value of the claim.

The third state has structure. In the formal regime it can be terminal; in the empirical regime it reopens, since fallibilism keeps every node revisable (§5). And the wait itself comes in knowable shapes. In a decidable system it is knowably temporary, a statement of Presburger arithmetic settled before you run anything; a provably non-halting search is knowably forever hung; an open conjecture like P versus NP is a wait whose own length is itself unsettled. So undecidability is itself decided, a vigil you can be told, with certainty, has no morning, and untrue is a status with structure rather than a uniform fog.

From the three states falls an ordering of dignity, and it runs against the instinct. Among claims presenting as knowledge, accountable falsehood outranks unaccountable pseudo-knowledge. A false claim stuck its neck out and reality took the swing; it narrowed the space and told you something, even if the something was no. An unbuilt claim risked nothing and told you nothing, and not even wrong is the worse verdict, the floor Pauli named. The qualifier is load-bearing and not decoration. An openly labeled conjecture is not presenting as knowledge and is not being demoted by this ordering; it is doing exactly what the trichotomy asks, announcing its own untruth and naming its test. Strip the qualifier and the ordering is simply false. Keep it, and it ranks the claim that took a risk above the claim that posed as having taken one.

Pragmatism is the home: truth as what survives inquiry rather than what corresponds (James 1907, Dewey 1929, Peirce), belief as a disposition to act whose strength is the odds (Ramsey 1926), paper doubt dropped (Peirce). What pragmatism could not finish, lacking the knower, is make the program run, since a human cannot expose the inner state where the justification is supposed to live. Kant supplies the phenomenon/noumenon boundary, recruited for the frame only; the breaking is Peirce’s Secondness, and only its machine version, one replayable bit, is the paper’s own. Popper supplies the capacity to fail as the mark of a claim that says anything, and his “irrefutability is a vice” is narrower than the line here, unscientific rather than untrue, a widening. Truth-by-construction and a third status are a century old (Brouwer; Łukasiewicz 1920; Aristotle’s sea battle first); the narrow delta is to keep bivalence in the world, put the third state in the ledger of entitlement, read it as no-passing-build, and rank false above untrue, which is Popper’s spirit and not the logicians’. And verificationism is named on purpose, since cannot be false, therefore cannot be true leans on it. Its sharpest modern form is Dummett’s anti-realism, truth tied to verification and bivalence withheld from the undecidable, the lineage this paper takes as home.

The frame is mostly inherited; the synthesis is the contribution, the citable epistemology a buildable account of machine knowledge presupposes.

9. Conclusion

The account is old parts, assembled in pragmatist order and read as a build, and the claim is that the boundary they draw is the right one: bivalence rules where a verdict is laid down, the unbuilt is untrue, and what cannot be false cannot be true. The scope is deliberately narrow. The paper treats the empirical and formal regimes and does not claim they exhaust truth, with the normative and modal and aesthetic left out of frame, a narrowing and not a branching.

The picture it leaves is the one the frame began with. A knower works only the phenomenon, and the world in itself stays out of reach and makes contact only by refuting a wrong build. So every truth a knower holds is a build standing on sufferance, could have fallen and has not, held while it stands without ever being certified, because the knower is never at the limit, only ever on the way to it. That is not a smaller thing than certainty. It is the only thing exposure was ever able to buy, and it is enough to act on.

And a single mind can assemble every structure here and still certify none of it, because it can only ever grade its own projection against itself.

References

Canonical sources the argument rests on. The author’s companion essays, which work out several of these moves first and informally, are listed separately under Provenance below as lineage, not as entitlement.

Provenance

This paper consolidates arguments first worked out informally on the author’s blog. Each is reproduced here in full, so the paper stands on its own; the posts are listed as lineage, not as entitlement, so a reader can trace any move back to where it was first made and rerun it there.