REVIEW 4 major objections 5 minor 26 references
Axioms for the Measure of Evidence
T0 review · 4 major / 5 minor · reviewed 2026-08-04 · deepseek-v4-flash
Pith's one-line read This paper argues that probability is a ratio of counts of possibilities, and that counting infinitesimal units of evidence makes the equal-probability principle exact even when the possibilities are infinite.
desk verdict A clearly written, philosophically bold axiom system for objective evidence, but the headline promise of a general measurement solution falls apart on the countability-of-the-continuum claim and the deferred continuous cases. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing object is the ratio identity P(A) ≡ E(A)/E(A∨¬A) (Theorem 4), where E is a hypernatural counting measure — a number that may be infinite, defined by Axiom 1 as the cardinality of a set, so each atomic possibility contributes exactly one unit of evidence. Axiom 4 fixes the size of the exhaustive possibility space 𝕌 at ℵ, making each atomic probability 1/ℵ, an infinitesimal whose infinite sum equals one; this is what reconciles uniformity with infinite sets. Hypernatural and hyperrational numbers from nonstandard analysis supply the arithmetic: they obey the usual rules of calculation while allowing infinite sums of infinitesimals to equal one, and the optional Axiom 4 only se
What would settle it
Work the paper's announced three-particle example: derive the possibility space for a triangle using uniformity over the three internal angles, then again using the three central angles. If both derivations satisfy the axioms and produce different probabilities for the same event, the claim that a model determines a unique possibility space fails. The paper defers this derivation to future work, so the calculation is open and concrete.
Extended reading notes
Core claim
The central claim is that all measures must ultimately be counting measures. Axiom 1 defines each hypernatural number as the cardinality of a set, so every element counts equally. Axiom 2 identifies a well-defined proposition with a disjunction of atomic possibilities. Axiom 3 sets evidence E(A) as the number of possibilities in A. Axiom 4 gives the possibility space 𝕌 a countably infinite size ℵ. Theorem 4 defines probability as P(A) = E(A)/E(A∨¬A). Each atomic possibility carries probability 1/ℵ, so the sum over all possibilities is exactly one, reconciling uniformity with infinity. From these it derives the usual rules of probability, leaving only the deduction of the possibility space.
Load-bearing premise
The load-bearing premise is that the continuum can be understood as a countable set of discrete atomic possibilities, which requires rejecting the Axiom of Choice for infinite sets; if the real line is genuinely uncountable, the counting measure cannot apply to continuous propositions and the proposed solution covers only finite or finitely described spaces.
Editorial extensions
If this is right
- Every adequately defined proposition receives a definite, uniquely correct probability from its definition alone, with no empirical input beyond the chosen model of reality.
- The K-axioms and C-axioms become theorems of the counting framework, so all standard probability calculations and Bayes's theorem remain valid while resting on simpler premises.
- Continuous parameters become countable: a continuous range is partitioned into ℵ atomic units, so uniform probability over an interval is exact and logically possible events receive probability 1/ℵ rather than zero.
- The standard non-informative prior over a scale parameter, P(r) = dr/r, is claimed to follow from uniformity over a correctly deduced possibility space rather than from invariance conventions, with the derivation promised in future work.
Reading between the lines
- A consequence the paper leaves implicit: adopting these axioms moves the hard work of probability into ontology — a model and a deduction of the possibility space must be specified before any number is assigned, and the paper's three-particle example shows that deduction is not yet a general method.
- If the rejection of the Axiom of Choice for infinite sets is taken seriously, it reaches beyond this paper: standard results of analysis and measure theory that depend on uncountable cardinalities or non-measurable sets would need re-derivation inside the countable framework, which the axioms do not attempt.
- The framework yields a concrete testable consequence: the earlier result that an unknown positive ratio has probability 1/2 of exceeding 1 should emerge as a theorem once the possibility space for ratios is derived from circular symmetry; deriving that space explicitly would confirm or refute the axioms.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proposes axioms for an objective measure of evidence, E(A), defined as the hypernatural cardinality of the set of atomic possibilities making A true. Uniformity is asserted as the basic principle: each atomic possibility has measure 1. Probability is then defined as the ratio P(A)=E(A)/E(A∨¬A), and conditional probability as E(A∧B)/E(B). The authors argue that all measures must ultimately be based on uniformity, that uniformity over infinite sets is made consistent by using hypernatural numbers, and that the continuum can be understood as discrete and countable, so that the axioms provide a general solution to the 'measurement problem' of assigning probabilities de novo. Theorems 1–6 derive additivity, sum rule, odds, probability, and conditional probability from the axioms; an optional axiom assigns an infinite hypernatural cardinality ℵ to the universal space U.
Significance. If the central claims were established, the paper would offer a striking resolution of the objective Bayesian measurement problem: a single counting principle would determine probabilities for arbitrary propositions, including continuous parameter spaces, via infinitesimal atomic probabilities. The paper has some genuine strengths: it clearly identifies the gap between calculation axioms and measurement axioms, it explicitly recognizes that standard measure theory assigns probability zero to possible atomic events, and it is candid about where the derivation is incomplete. The axiomatic system itself is simple and internally consistent for discrete/hyperfinite possibility spaces, and the use of nonstandard counting measures is an interesting proposal. However, the claimed generality is not delivered. The paper's treatment of continuous spaces rests on a nonstandard set-theoretic premise—countability of the continuum—that is asserted rather than proved and that is false in the standard foundations the paper otherwise uses. Moreover, the key derivation of the U-space for continuous parameters is explicitly deferred to future work. Thus the contribution, as it stands, is a philosophic
major comments (4)
- [§10.1.3 and Theorem 4] The load-bearing premise that 'the continuum can be understood as discrete and countable' is unsupported and, in standard ZF set theory, false. Cantor's diagonal theorem proves uncountability of ℝ without the Axiom of Choice; rejecting Choice does not make ℝ countable. The paper rejects uncountable sets and the Axiom of Choice for infinite sets, but it does not specify an alternative set theory or provide a proof that continuous possibility spaces are countable or hyperfinite. Yet Theorem 4, P(A)=E(A)/E(A∨¬A), applies only if every proposition is a set of atomic elements with hypernatural cardinality. If the possible values of a continuous parameter form an uncountable set, no such cardinality exists and the axioms do not apply. This is a central, load-bearing gap rather than a peripheral philosophical aside.
- [§10.2, §10.3, §10.5.1] The paper's claim to provide a general solution to the measurement problem is not supported by the actual content. The crucial step—deriving the U-space, i.e., the correct atomic possibility space, from an ontological model for continuous parameters—is deferred in all continuous examples: §10.2 says 'We will show in future work that there is a necessary symmetry that determines the U-space,' §10.3 says 'In future work we will show how space can instead be deduced,' and §10.5.1 says 'We intend in future work to derive the possibilities 𝕌.' Without these derivations, the axioms apply only to already-specified discrete or hyperfinite sets, which is a substantial restriction. The abstract's claim of a general and relatively simple solution is therefore premature.
- [§6, Axioms 1–3, Theorem 4] Uniformity is fixed by construction rather than derived. Axiom 1 assigns measure 1 to every atomic element; Axiom 3 identifies evidence with that measure; Theorem 4 then defines probability as |A|/|U|. The argument in §6 that 'all numbers are based on uniformity' is a philosophical appeal to the nature of counting, not a mathematical derivation, and it does not establish that any measure of evidence must be uniformly distributed over the chosen atomic elements. The paper's own acknowledgment in §8.1 that the absolute scale is 'only a convenient assignment' further undercuts the claim that uniformity is 'inevitable if all propositions are adequately defined.' The axioms are internally consistent, but the central claim of inevitability is not substantiated.
- [§10.2] The analogy between U-space and Euclidean distance, and the assertion that 'for any model O, there must be a U-space analogous to a Euclidean space,' is an unproved conjecture. The example of three angles in a triangle is used to motivate the existence of a unique U-space, but no formal definition of U-space is given that would allow one to check existence, uniqueness, or the claimed rotational invariance. This is not a minor omission: the paper's program reduces the measurement problem to 'deducing possibilities 𝕌 from model O,' and the paper provides no general method or theorem for doing so.
minor comments (5)
- [§8.4, footnote 2] The notation is confusing: 'ℵ=2ℵ0, where ℵ0=|ℕ|' is asserted, but then 'ℵ is not a transfinite number as defined by Cantor' is stated without explanation. If ℵ is a hypernatural nonstandard integer, its cardinality in the model needs clarification.
- [§6, Eq. (6)] The example '{X,X,X}' is not a set with three elements; set notation suppresses duplicates. This undermines the intended point about counting three objects.
- [§9] Theorems 1–6 are direct consequences of the definitions and are more naturally called propositions or corollaries. The label 'Theorem' overstates the mathematical content.
- [§4] The derivation of ¬(A∨¬A)=A∧¬A relies on classical logic; this is fine, but the text presents it as a general logical identity without stating the underlying classical framework.
- [§10.1.3] The phrase 'countably infinite continuum' is an oxymoron in standard set theory; the paper should state explicitly what alternative set theory is assumed.
Circularity Check
Uniformity is built into the definition of number and into Axiom 1; the claimed 'inevitability' of uniformity is therefore self-definitional, and P(u_i)=1/ℵ is fixed by construction. The continuous-space generality is deferred/unproved.
-
self definitional
[Section 6, Eq. (6)]
"We define a number as the measure (size or cardinality) of a set. Thus 3 is the measure μ of any set that contains three elements, 3≡μ{∙,∙,∙}=... Since the natural numbers are based on uniformity, and all numbers are derived from the natural numbers, all numbers are based on uniformity. Therefore any numbers that measure evidence must also be based on uniformity."
The conclusion that all measures must be based on uniformity is made true by the definition of 'number' introduced in this sentence. The paper defines a number as the result of counting equally weighted elements; it then uses this definition to conclude that any number measuring evidence must be uniform. This is not an independent demonstration: the uniformity is an input to the definition, not a derived property of numbers or evidence. The paper later concedes 'Uniformity is too obvious to prove' and 'we assert uniformity in our first axiom', confirming that the claimed 'show that all measures must ultimately be based on uniformity' is a restatement of the definitional choice, not a derivation.
-
self definitional
[Axiom 1 (Eq. 7), Theorem 4 (Eq. 14), Section 10.1.2]
"Axiom 1: 𝔫≡μ{x1,x2,…x𝔫} ... so that ∀i,j, μ(xi)=μ(xj)=1. ... Theorem 4: P(A)≡E(A)/E(A∨¬A). ... The minimal units of evidence will be E(ui)=1 and P(ui)=1/ℵ (∀i)."
Uniformity over atomic possibilities is stipulated directly by Axiom 1: every element xi has measure 1. Axiom 3 identifies evidence with this measure, and Theorem 4 defines probability as the ratio of two such equal-count measures. Consequently P(ui)=1/ℵ is the original axiom rewritten in probability notation, not a predicted consequence. The paper presents this as the key outcome of its solution to the measurement problem, but the equal atomic probabilities are fixed by construction at the level of Axiom 1, before any theorem is stated.
1 more flagged steps
-
self citation load bearing
[Section 10.5.2, citing reference [26]]
"We demonstrated previously that P(q<1)=P(q>1), so that q=1 is the median of the distribution, not just in this case but for a large set of statistical models that includes all those famous enough to have names [26]. The present axioms indicate that it is not only probability, but space itself, that is symmetrically distributed around q=1."
The claim that 'space itself is symmetrically distributed around q=1' is justified only by the authors' own prior result [26], which is cited as a demonstrated fact but not proven or independently verified in the present paper. The sentence 'The present axioms indicate...' does not supply a derivation from Axioms 1-4; it imports the conclusion from the same group's earlier work. This is not the central derivation of the paper, but it is load-bearing for the scale-parameter example, which is offered as an application of the axioms.
full rationale
The paper's central mathematical machinery is an axiomatic system: Axiom 1 defines each element of a set as unit measure, Axiom 2 identifies propositions with sets of atomic possibilities, Axiom 3 defines evidence as that counting measure, and Theorem 4 defines probability as a ratio of two such counts. Derivations of additivity, the sum rule, and conditional probability from these axioms are straightforward and internally valid. However, the paper's headline claim that it 'first show[s] that all measures must ultimately be based on uniformity' is not an independent result. Eq. (6) defines 'number' as the size of a set, with each element counted equally, so the conclusion that numbers and hence evidence measures are based on uniformity is true by definition. Section 6 then asserts uniformity as an axiom ('we assert uniformity in our first axiom'), confirming that the uniformity conclusion is an input. The equal atomic probabilities P(ui)=1/ℵ, presented as the key solution to the measurement problem, are literally Axiom 1 plus Axiom 3 plus Theorem 4: they are fixed by construction. The paper also defers the hard cases for continuous parameters: §10.2 says 'We will show in future work', §10.3 says 'In future work we will show how space can instead be deduced', and §10.5.1 says 'We intend in future work to derive the possibilities U'. Without a proof that continuous possibility spaces are countable or hyperfinite, the axioms as stated apply only to discrete/hyperfinite U-spaces, and the abstract's claim of a general solution is not delivered. The unsupported countability of the continuum in §10.1.3 (rejection of the Axiom of Choice is not sufficient; Cantor's diagonal argument is ZF-provable) is a correctness risk rather than a circularity step. One self-citation, reference [26], is used to support the q<1 vs q>1 symmetry example; it is not the core of the axiomatic derivation but it is load-bearing for that example. Overall, the derivation chain is transparent and axiomatic, but the central uniformity conclusion and the uniform probabilities are definitionally equivalent to the axioms, giving a circularity score of 6.
Assumptions & free parameters
free parameters (2)
- ℵ: infinite scale of the universal possibility space =
unspecified; suggested ℵ = 2^ℵ0 (Section 8.4)
- Ontological model O (choice of U-space) =
unconstrained
assumptions (6)
- ad hoc to paper Axiom 1: every hypernatural number n is the measure of any set with n elements, assigning each element measure 1 (uniform counting).
- domain assumption Axiom 2: a proposition is a subset of U, a disjunction of atomic possibilities.
- ad hoc to paper Axiom 3: evidence E(A) is the measure of set A, a hypernatural number.
- ad hoc to paper Axiom 4: the universal set U has infinite hypernatural size ℵ.
- ad hoc to paper Rejection of uncountable sets and of the Axiom of Choice for infinite sets; the continuum is countable and discrete.
- ad hoc to paper Symmetry of ignorance: reason alone cannot justify unequal evidence for indivisible, mutually exclusive propositions.
invented entities (3)
-
E(A): hypernatural measure of evidence
-
ℵ: infinite hypernatural cardinality of U
-
U-space / S-space distinction
Cite this review
Pith. "Pith review of Axioms for the Measure of Evidence." pith.science (2026). https://pith.science/paper/GG7TB3KQ
@misc{pith2026250907033,
author = {Pith},
title = {Pith review of: Axioms for the Measure of Evidence},
year = {2026},
howpublished = {\url{https://pith.science/paper/GG7TB3KQ}},
note = {Machine review of arXiv:2509.07033}
}
read the original abstract
There has not been an established mathematical measure of evidence. Some Bayesians have argued that probability can be an objectively correct measure of ``rational degrees of belief,'' which we do not distinguish from evidence. However, support for the objectivist view has been limited due to the lack of a general method for assigning probabilities to evidence (belief) de novo. The standard axioms of Kolmogorov and Cox solve only the calculation problem, specifying how probabilities can be calculated from other probabilities. They do not solve the measurement problem of how to determine the uniquely correct value of P(A) given only A. The prototypical solution has always been Laplace's principle of indifference, which assigns equal (uniform) probabilities to all possibilities. However, uniformity is well known to be inconsistent with the standard axioms when there are infinite possibilities. Here we introduce new axioms that resolve this inconsistency. We first show that all measures must ultimately be based on uniformity, and that uniformity is inevitable if all propositions are adequately defined. We then reconcile uniformity with infinite possibilities by using hyperrational numbers, so that an infinite sum of infinitesimal probabilities can equal one. Our axioms thereby provide a general and relatively simple solution to the measurement problem. We discuss a variety of conceptual obstacles that have made this solution difficult to recognize.
Reference graph
Works this paper leans on
-
[1]
Springer, New York, U.S., 1985
James O Berger.Statistical Decision Theory and Bayesian Analysis. Springer, New York, U.S., 1985
work page 1985
-
[2]
J. M. Bernardo and A. F. Smith.Bayesian Theory. John Wiley & Sons, Chichester, U.K., 1994
work page 1994
-
[3]
Cambridge University Press, 2003
Edwin T Jaynes and G Larry Bretthorst.Probability Theory: The Logic of Science. Cambridge University Press, 2003
work page 2003
-
[4]
Claus Beisbart and Stephan Hartmann, editors.Probabilities in Physics. Oxford University Press, 2011
work page 2011
-
[5]
Towards a general theory of neural computation based on prediction by single neurons
Christopher D Fiorillo. Towards a general theory of neural computation based on prediction by single neurons. PLoS ONE, 3:e3298, 2008
work page 2008
-
[6]
Christopher D Fiorillo. Beyond Bayes: On the need for a unified and Jaynesian definition of probability and information within neuroscience.Information, 3(2):175–203, 2012
work page 2012
-
[7]
Oxford University Press, Oxford, England, 1961
Harold Jeffreys.The Theory of Probability, 3rd Edition. Oxford University Press, Oxford, England, 1961
work page 1961
-
[8]
Cox.The Algebra of Probable Inference
Richard T. Cox.The Algebra of Probable Inference. The Johns Hopkins Press, Baltimore, MD, USA, 1961. 12 APREPRINT- SEPTEMBER24, 2025
work page 1961
Show all 26 references
-
[9]
Dover Publications, New York, 1951 (original French, 1814
Pierre Simon Laplace.A Philosophical Essay on Probabilities. Dover Publications, New York, 1951 (original French, 1814
1951
-
[10]
Chelsea Publishing Company, New York, USA, 1950
Andrey Kolmogorov.Foundations of the Theory of Probability, 2nd Edition. Chelsea Publishing Company, New York, USA, 1950
1950
-
[11]
Theselectionofpriordistributionsbyformalrules.JournaloftheAmerican Statistical Association, 91(435):1343–1370, 1996
RobertE.KassandLarryWasserman. Theselectionofpriordistributionsbyformalrules.JournaloftheAmerican Statistical Association, 91(435):1343–1370, 1996
1996
-
[12]
Dennis V. Lindley. The philosophy of statistics.The Statistician, 49(3):293–337, 2000
2000
-
[13]
Probability and logic.Journal of Applied Logic, 1:151–165, 2003
Colin Howson. Probability and logic.Journal of Applied Logic, 1:151–165, 2003
2003
-
[14]
Robert, Nicolas Chopin, and Judith Rousseau
Christian P. Robert, Nicolas Chopin, and Judith Rousseau. Harold Jefferys’s Theory of Probability revisited. Statistical Science, 24(2):141–172, 2009
2009
-
[15]
Springer-Verlag, Berlin, 2004
Marek Kapinkski and Ekkehard Kopp.Measure, Integral, and Probability; 2nd Edition. Springer-Verlag, Berlin, 2004
2004
-
[16]
On the theory of errors and least squares.Proceedings of the Royal Society of London, 138:48–55, 1932
Harold Jeffreys. On the theory of errors and least squares.Proceedings of the Royal Society of London, 138:48–55, 1932
1932
-
[17]
Non-informativepriorsdonotexist: AdialoguewithJoseM.Bernardo
TelbaZ.IronyandNozerD.Singpurwalla. Non-informativepriorsdonotexist: AdialoguewithJoseM.Bernardo. Journal of Statistical Planning and Inference, 65:159–189, 1997
1997
-
[18]
Oxford University Press, New York, 1986
Thomas Nagel.The View from Nowhere. Oxford University Press, New York, 1986
1986
-
[19]
North-Holland Publishing, Amsterdam, 1966
Abraham Robinson.Non-standard Analysis. North-Holland Publishing, Amsterdam, 1966
1966
-
[20]
Finite arithmetic axiomatization for the basis of hyperrational non- standard analysis.Axioms, 10(4):263, 2021
Yuri N Lovyagin and Nikita Yu Lovyagin. Finite arithmetic axiomatization for the basis of hyperrational non- standard analysis.Axioms, 10(4):263, 2021
2021
-
[21]
Addison-Wesley, Reading, MA, USA, 1974
Donald Ervin Knuth.Surreal Numbers: How two ex-students turned on to pure mathematics and found total happiness: a mathematical novelette. Addison-Wesley, Reading, MA, USA, 1974
1974
-
[22]
General covariance and the foundations of general relativity: eight decades of dispute.Reports on progress in physics, 56(7):791, 1993
John D Norton. General covariance and the foundations of general relativity: eight decades of dispute.Reports on progress in physics, 56(7):791, 1993
1993
-
[23]
Princeton University Press, Princeton, NJ, USA, 2012
Tim Maudlin.Philosophy of Physics: Space and Time. Princeton University Press, Princeton, NJ, USA, 2012
2012
-
[24]
Basic Books, New York, USA, 2020
Julian Barbour.The Janus Point: A New Theory of Time. Basic Books, New York, USA, 2020
2020
-
[25]
Dover Publications, New York, 1958
Hans Reichenbach.The Philosophy of Space and Time. Dover Publications, New York, 1958
1958
-
[26]
TheobjectiveBayesianprobabilitythatanunknownpositiverealvariable is greater than a known is 1/2.Philosophies, 6(24):1–24, 2021
ChristopherDFiorilloandSunilLKim. TheobjectiveBayesianprobabilitythatanunknownpositiverealvariable is greater than a known is 1/2.Philosophies, 6(24):1–24, 2021. 13
2021
Reviewed August 4, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.