Pith. sign in

REVIEW 4 major objections 3 minor 1 cited by

A Mathematical Theory of Value: a synthesis on goal-directed agency under resource constraints

T0 review · 4 major / 3 minor · reviewed 2026-07-12 · grok-4.5

Pith's one-line read Value is the scale-invariant rate at which an agent turns a resource into goal-progress relative to a fixed goal frame, forced into the same logarithmic form as information.

desk verdict Abstract-only: ambitious Shannon-style packaging of value with strong claimed LM numbers, but the coding theorem and correlations cannot be audited and carry a real circularity risk until operational definitions are checked. read the letter →

arxiv 2606.12502 v2 pith:ASRIMKOR submitted 2026-06-10 physics.soc-ph cs.AI

classification physics.soc-phcs.AI
keywords valuetheorygoal-directedagencyresourceconstraintsscaleinvariancecodingtheoremmutualinformationalignmentlanguagemodels
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper argues that value—the quantity that goal-directed agents create, destroy, and exchange—is a lawful structural quantity of the same kind as Shannon information. By defining value strictly as the rate of converting a resource into goal-progress inside a frame fixed by the agent’s goal, a single scale-invariance axiom forces the measure V = ∑ k_i ln e_i; the same form also emerges from reinvestment and ergodicity. From this definition the paper derives a coding theorem ΔG ≤ I(X;Y) and a decomposition of realized value as a difference of Kullback–Leibler divergences. Population-level consequences follow: value remains frame-relative while price is frame-independent, and a resource-pooling fleet inherits a joint information ceiling. A dynamical layer yields an is/ought asymmetry from which alignment appears as ordinary control stability. Live language-model experiments are offered as empirical checks of the single-frame laws and of the population capacity-region predictions.

What carries the argument

The scale-invariance axiom applied to the resource-to-goal conversion rate relative to a goal-fixed frame; it selects the unique logarithmic measure V = ∑ k_i ln e_i and thereby licenses the coding theorem ΔG ≤ I(X;Y) and the KL decomposition of realized value.

What would settle it

A pre-registered experiment in which measured mutual information I(X;Y) and realized goal-progress ΔG on a new task family fail to track each other with slope near 1, or in which a resource-pooling multi-agent system systematically exceeds the predicted joint ceiling I(X;Y_{1:m}).

Watch

Extended reading notes

Core claim

Value is forced by scale invariance into the logarithmic rate V = ∑_i k_i ln e_i; the associated coding theorem states that the attainable goal-progress ΔG cannot exceed the mutual information I(X;Y) between resource state and perception, while realized value decomposes exactly as G = D(q∥r) − D(q∥p). The same structure extends to fleets under a joint ceiling G_fleet ≤ I(X;Y_{1:m}) ≤ H(X) and supplies a control-theoretic account of alignment.

Load-bearing premise

That the operational notions of resource, goal-progress and goal-fixed frame measured on live language models are faithful instances of the abstract quantities appearing in the coding theorem rather than merely correlated proxies.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 3 minor

Summary. The manuscript proposes that value is a lawful structural quantity in the same category as information: the rate at which a goal-directed agent converts a resource into goal-progress relative to a goal-fixed frame. A scale-invariance axiom is said to force the logarithmic form V = ∑_i k_i ln e_i (with a consistency check via Peters-style ergodicity). From this the authors derive a coding theorem ΔG ≤ I(X;Y) and a realized-value decomposition G = D(q∥r) − D(q∥p). Population corollaries (frame-relative value vs frame-independent price; fleet ceiling G_fleet ≤ I(X;Y_{1:m}) ≤ H(X)) and a dynamical is/ought asymmetry for alignment are sketched. Empirical claims on live language models are reported as pre-registered: perception mutual information tracks realized capability (Spearman ρ = 0.977, 30 points); out-of-sample ΔG tracks I(X;Y) (slope 0.953, n = 42); a frontier-model population continuation is said to confirm capacity-region predictions within frozen bands.

Significance. If the derivation is non-circular and the LM operationalizations faithfully instantiate the abstract resource-conversion quantities, the paper would supply a Shannon-style foundation for value, a coding bound usable as a governance ceiling, and a control-theoretic reading of alignment. Pre-registration of the empirical protocols and explicit retirement of a failed mean-field residual law are methodological strengths that raise the evidential bar relative to typical theory-plus-proxy work in this area. The claimed unification of single-agent value, multi-agent capacity regions, and alignment-as-stability would be of high interest to physics-of-society, information theory, and AI-governance audiences.

major comments (4)
  1. [Abstract (coding theorem and LM tests)] The coding theorem ΔG ≤ I(X;Y) together with the decomposition G = D(q∥r) − D(q∥p) is load-bearing for both the mathematical and empirical halves of the central claim. From the abstract alone it is not possible to verify that q, r, p are defined independently of the channel variables used to compute I(X;Y). If G is extracted from the same logits, next-token distributions, or task scores that define the perception channel, the bound reduces (at least partly) to an information identity and the reported tracking (ρ = 0.977; slope 0.953) ceases to be a non-trivial test. The manuscript must supply independent operational definitions of resource, goal-progress, and the goal-fixed frame, and must show that ΔG is measured without using the same channel that yields I(X;Y).
  2. [Abstract (scale-invariance axiom / V = ∑ k_i ln e_i)] The scale-invariance route is said to force V = ∑_i k_i ln e_i. The factors k_i remain free parameters (per-resource scale factors). The abstract does not state how they are fixed, estimated, or shown to cancel in the reported ratios and slopes. Without that accounting, claims that the measure is forced by the axiom alone, or that the empirical slopes are parameter-free predictions, are not yet secured. The manuscript should either derive constraints that eliminate the k_i or report them explicitly as fitted quantities and re-evaluate the out-of-sample tests under that accounting.
  3. [Abstract (pre-registered LM tests; ρ = 0.977, slope 0.953)] The empirical half equates measured perception mutual information with realized capability and out-of-sample ΔG with I(X;Y) on live LMs. Faithfulness of those operationalizations to the abstract resource-conversion model is the weakest link in the strongest claim. The abstract supplies neither the measurement protocol nor a demonstration that the quantities are instances of the same objects appearing in the coding theorem rather than loosely correlated proxies. Pre-registration is noted but cannot be audited from the abstract; the full protocol, frozen analysis plan, and any deviations must be provided and checked before the reported figures can be treated as diagnostic of the theorem.
  4. [Abstract (fleet corollary / v5 correction)] An earlier sum-form population claim is acknowledged as wrong and corrected in v5 to the joint ceiling G_fleet ≤ I(X;Y_{1:m}) ≤ H(X). The abstract does not indicate whether any of the reported empirical confirmations (growth-gap law, coalition submodularity, Kelly selection) depended on the retracted sum-form statement. The manuscript should state which results, if any, were re-derived under the corrected corollary and whether any pre-registered bands were revised after the correction.
minor comments (3)
  1. [Abstract] The abstract packs mathematical claims, empirical numbers, a correction history, and a retired mean-field law into a single dense paragraph. A short roadmap (theory → single-frame tests → population continuation → retired claims) would help readers locate load-bearing results.
  2. [Abstract (frame-relative value)] Notation for the goal-fixed frame is introduced verbally but never symbolized; a consistent symbol (e.g., F_g) would clarify later frame-relativity statements about value versus price.
  3. [Abstract (Peters ergodicity route)] The phrase “kin routes, a consistency check, not an over-determination” is opaque without the full text; a one-sentence gloss of what is being checked against what would reduce ambiguity.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity identifiable from the abstract; scale-invariance forces the log form and the coding theorem is presented as derived, not definitional.

full rationale

Only the abstract is available, so no internal equation chain can be walked beyond the stated claims. The abstract defines value operationally as the rate of resource-to-goal-progress conversion relative to a goal-fixed frame, then states that a scale-invariance axiom forces V=∑_i k_i ln e_i (with Peters 2019 ergodicity as an independent consistency check, not a load-bearing self-citation). It separately claims to derive the coding theorem ΔG ≤ I(X;Y) and reports the decomposition G=D(q∥r)−D(q∥p) as a realized-value result, without equating the definition of G to that KL difference by construction in the provided text. Empirical claims are pre-registered correlations on live LMs (ρ=0.977, slope 0.953) rather than fitted parameters renamed as predictions. No uniqueness theorem is imported from the authors, no ansatz is smuggled via self-citation, and no self-definitional reduction (Eq. X ≡ Eq. Y) is quotable from the abstract. Absent full text, no circular step can be exhibited under the hard rules; the derivation is presented as self-contained from the axiom and abstraction. Score 0 is the honest finding.

Assumptions & free parameters 2 free parameters · 4 assumptions · 2 invented entities

The central claim rests on identifying value with a resource-to-goal-progress conversion rate relative to a goal-fixed frame, plus a scale-invariance axiom that forces the log form; Peters (2019) is used as a consistency check, not an independent free parameter. Free modeling degrees of freedom include the k_i weights and frame choice. No new physical particles; the invented structure is the value measure and frame. Standard information-theoretic identities are assumed.

free parameters (2)
  • k_i (per-resource scale factors)
    Appear in V=∑_i k_i ln e_i; abstract does not state whether they are fixed by uniqueness or fit to data.
  • goal-fixed frame
    Value is defined relative to a frame fixed by the agent’s goal; frame choice is a modeling degree of freedom that sets measured value and can absorb empirical degrees of freedom.
assumptions (4)
  • ad hoc to paper Scale-invariance forces the logarithmic measure V=∑_i k_i ln e_i
    Stated as the ruthless abstraction’s forcing axiom; standard functional-equation move, but chosen here to define value.
  • ad hoc to paper Value is the rate at which an agent converts a resource into goal-progress relative to a goal-fixed frame
    The single abstraction on which the theory is built; not derived from prior physics or economics in the abstract.
  • domain assumption Compounding of a reinvested resource yields the same log form via Peters (2019) ergodicity
    Cited as a consistency check, not an over-determination; relies on the external ergodicity argument.
  • standard math Standard Shannon mutual information, entropy, and KL-divergence inequalities
    Used for the coding theorem, decomposition, and fleet ceiling.
invented entities (2)
  • value as Shannon-like structural quantity independent evidence
    purpose: Place goal-directed agency under the same mathematical category as information and derive coding/fleet bounds.
    Core postulated object. Abstract claims falsifiable handles via ΔG ≤ I(X;Y) and LM capability tracking; independent_evidence set true only insofar as those tests are real and non-definitional.
  • goal-fixed frame
    purpose: Relativize value while keeping price frame-independent across agents.
    Modeling construct introduced to separate value from price; no external collider-style handle beyond internal consistency of the theory.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Mathematical Theory of Value: a synthesis on goal-directed agency under resource constraints." pith.science (2026). https://pith.science/paper/ASRIMKOR

@misc{pith2026260612502,
  author       = {Pith},
  title        = {Pith review of: A Mathematical Theory of Value: a synthesis on goal-directed agency under resource constraints},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ASRIMKOR}},
  note         = {Machine review of arXiv:2606.12502}
}
abstract

We propose that value -- the quantity goal-directed agents create, destroy, and exchange -- is a lawful structural quantity in the same category as information. Following Shannon's method, we make one ruthless abstraction: value is the rate at which an agent converts a resource into goal-progress, relative to a frame fixed by its goal. A scale-invariance axiom forces a logarithmic measure, $V=\sum_i k_i\ln e_i$; compounding of a reinvested resource forces the same form via the ergodicity argument of Peters (2019) -- kin routes, a consistency check, not an over-determination. We derive a coding theorem of value, $\Delta G \le I(X;Y)$; realized value decomposes as $G=D(q\|r)-D(q\|p)$. For populations, value is frame-relative while price is frame-independent; a fleet that pools its resource and fuses its perception inherits the ceiling $G_{\rm fleet}\le I(X;Y_{1:m})\le H(X)$ (a corollary; an earlier sum-form claim was wrong and is corrected in v5). A dynamical layer yields an is/ought asymmetry from which alignment emerges as a control-stability condition. We test the single-frame laws on live language models, pre-registered: perception mutual information tracks realized capability (Spearman $\rho=0.977$ over 30 model$\times$domain points); out-of-sample $\Delta G$ tracks $I(X;Y)$, shape-invariant across four task shapes ($n=42$, slope $0.953$); over-confidence is measurable dissipation. The stated continuation gate has since been run (pre-registered, frontier-model population): the coupled capacity-region prediction -- growth-gap law, coalition submodularity with an XOR synergy control, joint ceiling, Kelly selection -- is confirmed within its frozen bands on real agents; the mean-field residual law $\|Vg\|/\gamma$ found no domain (populations hold no goal dispersion) and is retired to its mathematical scope. The contribution is the unification and the governance mapping that follows.

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Information Limits and Attractor Dynamics in Economies of Frontier LLM Agents: A Pre-Registered Test

    cs.AI 2026-07 conditional novelty 6.0 of 10

    A pre-registered experiment on Claude Opus 4.8 agents confirms an information-theoretic capacity region for coupled LLM-agent markets (with one key result being an algebraic identity) and finds that LLM populations ex...

Pith tools

Reviewed July 12, 2026 · model on record in the stance chip above.