{"id":"5907d3eb-3063-4e19-9cbd-c1633757b5e4","arxiv_id":"2507.02028","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":3.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A preliminary formal model represents capabilities as transformations of worlds and defines well-being as the maximum value reachable through them, while highlighting negative externalities of capability expansion.","lead":"This paper proposes a simple mathematical model of Sen's capability approach, where each person's capabilities are functions that change the world, and well-being is the best value reachable through those functions. It argues that one person's capability gain can hurt others, and that public infrastructure should be treated as part of people's real wealth.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Capabilities are modeled as total functions from worlds to worlds, so the closure under composition contains infeasible transformations and V can overstate real capability.","rationale":"I read the paper as a deliberately preliminary sketch whose central deliverable is the equation V(i,w)=max_{f∈Ci} v(i,f(w)), with Ci generated by compositional closure. The most load-bearing weakness is not interpersonal comparability. The paper explicitly defers social-welfare aggregation to future work, so the individual-level formalization stands or falls on whether Ci contains exactly the feasible capabilities. The total-function formalization fails this test. Because capabilities such as 'ride the bus' or 'order ice cream' are conditional on the agent's world, representing them as total functions W→W makes them executable from every world, and closure under composition generates transformations that are not real options. This is an internal modeling flaw, not merely a disagreement with a school of thought: the paper's own examples require conditional availability. The repair is straightforward—use partial functions or an availability relation—so the note remains conditionally salvageable rather than fundamentally unsound. For this reason I keep the reader's CONDITIONAL verdict unchanged, but I disagree that the weakest assumption is interpersonal comparability; the total-function problem infects V itself even in single-agent examples.","tokens_in":6510,"tokens_out":13197,"duration_ms":171527,"concrete_test":"Take W={home, city} and let 'take the bus' be a basic capability b with b(home)=city and b(city)=city, although there is no bus stop at home. Under the paper's total-function definition, b is executable at home, so V(i,home)≥v(i,city). Re-model the same situation with b as a partial function defined only on bus-stop worlds, adding a walk-to-stop capability f if present, and recompute V(i,home) with composition restricted to matching domains. If the total-function value strictly exceeds the partial-function value, then the formalization includes an infeasible transformation and the central equation must be revised. This single check settles whether the paper's closure operation is a faithful model of feasible capability.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 2 defines each capability as a total function f: W→W and takes Ci to be the closure of the basic capabilities under composition. Because every f is total, a capability such as 'take the bus' is formally executable in every world, including worlds where the agent is not at a bus stop. The paper's own bus-line example requires walking to the bus stop followed by the bus procedure, but in the formal model the single total function b already maps home to the city, making the bus ride executable from home. Thus V(i,w)=max_{f∈Ci} v(i,f(w)) can be realized by a transformation that is not a feasible capability. The same problem affects every conditional capability: 'order ice cream' is available only after reaching the shop, yet as a total function it is available from the starting world. Composition therefore does not model feasible sequences; it models an idealized semigroup of global world-to-world transformations. This is not a minor technicality. The central claim is that the equation models the capability approach, and the capability approach is about real, feasible opportunities. The monotonicity claim, that enlarging Ci increases V, is then vacuous for infeasible enlargements. A faithful formalization needs partial functions or, better, an availability relation A(i,w,f) restricting which capabilities can be applied in world w, with closure defined only over feasible compositions.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a mathematical formalization of the capability approach. Each agent i is given a set of basic capabilities, modeled as functions from a world set W to itself; closing this set under composition yields the capability set C_i. The agent's capability net value in a world w is defined as V(i,w) = max_{f in C_i} v(i,f(w)). The paper claims that enlarging C_i weakly increases V, that one agent's capability gain can harm another, and illustrates these points with a gun-threat game, a voting example, Sen's liberal paradox, and an independence condition formalized through product worlds W = product_i W_i.","tokens_in":6758,"tokens_out":5768,"duration_ms":64442,"significance":"The paper is a preliminary and self-contained contribution that gives a simple vocabulary for talking about capabilities, capability gains, and capability externalities. It does not rely on fitted parameters or external data, and the definition of V is explicit. The product-world formalization of independence in Section 5.1 is a useful step, as is the game-theoretic example showing that one agent's capability increase can reduce another's payoff. If the technical problems identified below are fixed, the framework could serve as a starting point for more quantitative capability-based policy analysis. However, the current version is not yet a fully developed theory, and the main formal claim is largely definitional.","major_comments":[{"comment":"The definition of capabilities as total functions f: W → W and the closure C_i under composition is too permissive: every capability is formally executable from every world, including worlds where its preconditions fail. In the bus-line example, the total function b already maps Aditi's home directly to the city, so V(i,w,C_i+b) is attained by applying b from home; the walking-to-the-bus-stop step f is unnecessary. The same problem affects any conditional capability, such as ordering ice cream only after reaching the shop. Thus the closure under composition does not model feasible sequences of actions; it models an idealized semigroup of global world-to-world transformations, and V can overstate real capability. I suggest modeling capabilities as partial functions, or adding an availability relation A(i,w,f) that restricts when f can be applied, and defining closure only over feasible compositions.","section":"Section 2"},{"comment":"The probability calculation for a tied election is incorrect. For 44,000 fair-coin votes, the probability of an exact 22,000-22,000 split is the binomial probability C(44000,22000)/2^44000 = 44000!/(22000! * 22000! * 2^44000). The expression (22000! * 22000!)/44000! is the reciprocal of the binomial coefficient, not a probability under this model. The subsequent bound involving '0.75(11,000)' also needs to be re-derived from the corrected formula. The qualitative point that a single vote is very unlikely to be pivotal may survive, but the quantitative claim as printed is wrong and should be corrected.","section":"Section 3"},{"comment":"The model assumes a real-valued, interpersonally comparable utility v(i,w) and uses the gain G(i,w,b) = V(i,w,C_i+b) − V(i,w,C_i) to compare benefits across different individuals, as in the swimming-pool-versus-bus-line example. The paper does not explain how v is calibrated or whether interpersonally comparable cardinal utility is presupposed. The text acknowledges that such issues are left for another paper, but this assumption is load-bearing for the social-choice interpretation of G. I recommend stating explicitly that interpersonal comparability is assumed, and discussing its limitations, or restricting the quantitative claims to an individual-level comparison.","section":"Section 2"}],"minor_comments":[{"comment":"The notation C_{i,k} is introduced but never used; the subscript k appears extraneous.","section":"Section 2"},{"comment":"The composition order is ambiguous: the text says that if f takes w to x and g takes x to z, then the composite f∘g takes w to z, but standard function composition writes f∘g as applying g first. Please clarify the convention or write g∘f.","section":"Section 2"},{"comment":"In the second payoff matrix, the row labels are not entirely consistent: the first game has rows 'buy' and 'no buy', while the second game adds a row 'threaten Sona'; the labeling of the columns and the payoffs should be checked for clarity.","section":"Section 3"},{"comment":"The expression '0.75(11,000)' should be written as 0.75^{11,000} or in an otherwise unambiguous exponential notation.","section":"Section 3"},{"comment":"There are typographical errors, including 'deccrease', 'intrroduce', 'equiibrium', 'Indpendence' in the Section 5.1 heading, and 'besife'. A careful proofreading pass is needed.","section":"Throughout"},{"comment":"The van Hees (2020) reference is listed but is not cited in the main text; please either cite it where relevant or remove it from the list.","section":"References"},{"comment":"The 'controversial example' about voting and voter dilution is not expressed in the formal model; it would strengthen the paper to show how the dilution effect is represented in terms of v and V.","section":"Section 4"}],"recommendation":"major_revision","confidential_remarks":"The manuscript reads as a preliminary conference paper. The total-function issue in Section 2 and the probability error in the voting example are substantive but fixable. Given the paper's brevity and informal style, I would not recommend acceptance without substantial revision. The authors should also engage more with existing formal treatments of capabilities, such as van Hees's 'Analyzing Capabilities,' to clarify the novelty of their model."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know this up front: it is a short, explicitly preliminary conference note, not a finished paper. What it has is a useful one-line model. Capabilities are transformations from worlds to worlds, each agent's capability set is closed under composition, and V(i,w) is the maximum value the agent can reach by applying any capability. That simple vocabulary makes it easy to talk about capability externalities (the gun-threat game) and about public infrastructure (the bus line vs. the pool). The examples are clear and the central point, that one person's capability gain can be another's loss, is worth stating.\n\nThe soft spots are real, and they are of three sizes. Smallest: the voting probability is wrong. (22000! × 22000!)/44000! is the reciprocal of a binomial coefficient, not the probability of a tied fair-coin election. The number is off by a factor of 2^44000, so the 'very very small' claim comes from a miscalculation. Medium: the model assumes interpersonally comparable v(i,w) without saying so. The gain comparisons G(i,w,b) across individuals only make sense if value is comparable, and the paper does not defend that. Largest: capabilities are total functions from W to W. That means the closure under composition includes transformations the agent cannot actually execute in the relevant world. The bus-line example is supposed to need walking to the stop and then the bus, but as a total function the bus is already executable from home. This overstates V and weakens the claim that the model captures real opportunity. The paper would need partial functions or an availability relation to fix this.\n\nWhat the paper does well beyond the formula: the independence idea using product worlds is a nice way to formalize Sen's personal-liberty intuition. The author is candid that the development is preliminary, and the reference list is the standard capability canon, though it does not actually engage van Hees's prior formal analysis, which is a missed connection.\n\nIn proportion: the central sketch is defensible, the details are not. The total-function issue is the one that touches the meaning of the model, the probability error is embarrassing but isolated, and the comparability assumption is an omission rather than a fatal flaw.\n\nI would not cite it yet, but I would send it to a referee who knows the capability literature. A good referee could push the author to fix the math, add partial functions or an availability semantics, and state the comparability assumption. If those are repaired, this becomes a genuinely useful pointer. For a reading group, it is a maybe: it will provoke good discussion, but you will spend part of the session correcting the voting calculation.","headline":"A likable but underbaked sketch: the V=max formula gives a serviceable vocabulary for capability externalities, but the total-function closure lets infeasible actions count and the voting probability is wrong.","tokens_in":7227,"tokens_out":3033,"would_cite":false,"duration_ms":32239,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["91A10","91B14"],"pacs":[],"model":"deepseek-v4-flash","headline":"The capability approach is formalized as the maximum value an agent can reach by composing her capabilities; increasing those capabilities can hurt others.","keywords":["capability approach","formal model","game theory","well-being","social choice","public projects","interpersonal comparability","composition closure"],"falsifier":"Find two public projects $p$ and $q$ such that multiplying every $v(i,w)$ by a positive constant reverses the ordering of the induced gains $G(i,w,p)$ and $G(i,w,q)$ for some agent; the model's claim that gains are a meaningful basis for project choice would then fail, because the social ranking would depend on the arbitrary scale of $v$.","tokens_in":6305,"feed_emoji":"📐","tokens_out":8714,"duration_ms":89637,"temperature":0.7,"pith_summary":"The paper proposes a mathematical formalization of the capability approach, replacing the question of what a person owns with the question of what she can do with it. In the model, each agent $i$ has a set of capabilities represented as functions from worlds to worlds, and her capability value is $V(i,w)=\\max(v(i,f(w)): f\\in C_i)$, where $C_i$ is the closure of her basic capabilities under composition. The paper's central claim is that enlarging $C_i$ can only weakly increase $V(i,w)$, and that society should compare gains $G(i,w,b)=V(i,w,C_i+b)-V(i,w,C_i)$ across individuals when choosing public projects such as a swimming pool or a bus line. It also argues, through a bargaining game in which a buyer acquires a gun and threatens the seller, that one person's capability increase can reduce another person's value, so capabilities are not an unmixed blessing.","feed_headline":"Well-being equals your best reachable world, says new model","feed_subtitle":"A formal capability approach shows why a better bus line helps some people and a gun hurts others.","key_machinery":"The carrying object is the capability net value $V(i,w)=\\max(v(i,f(w)): f\\in C_i)$, where $C_i$ is the least set containing the agent's basic capabilities and closed under composition $f\\circ g$. Composition closure is essential because real actions are sequences, such as walking to the bus stop, taking the bus, and buying ice cream. The gain $G(i,w,b)$ from a new social procedure $b$ is the difference the added capability makes to $V$, and the paper uses this quantity to frame which sub-population benefits from a public project. A secondary device is the independence model $W=\\prod_i W_i$, in which each agent's actions alter only her own coordinate, offered as a formal reading of the idea that sleeping position or bathroom color is a private matter.","core_discovery":"The central discovery is a minimal formal model in which an agent's well-being is not her current assets but the best value she can achieve by applying compositions of her capabilities. Formally, with a finite set of agents, a set of worlds $W$, and per-agent functions from $W$ to $W$, the agent's capability net value is $V(i,w)=\\max(v(i,f(w)): f\\in C_i)$, where $C_i$ is the closure of her basic capabilities under composition. Adding a new procedure $b$ to the capability set produces a gain $G(i,w,b)=V(i,w,C_i+b)-V(i,w,C_i)$, which the paper claims can differ sharply across individuals. The paper also demonstrates that capability increases can harm others, using a bargaining game where one player's acquisition of a gun shifts the Nash equilibrium from mutually beneficial trade to extortion, so an individual gain can be a social loss.","pith_inferences":["If the model is taken literally, the monotonicity of $V$ in $C_i$ gives a formal reason to keep capability lists open-ended: adding any capability weakly improves the agent's own position, though it may not improve others'.","The model could be tested empirically by inferring an agent's composition-closed capability set from observed choices and checking whether estimated gains $G$ match welfare rankings from survey data.","The need for interpersonal comparability is likely the binding constraint: the formal machinery works for a single agent without cross-person comparison, but social choice requires a common scale the model does not construct.","A probabilistic extension, replacing deterministic capabilities with stochastic maps, would connect the model to dynamic programming and expected-utility calculations, allowing risk to enter the capability value."],"forward_implications":["Enlarging an agent's capability set always weakly increases her capability value $V(i,w)$, so any capability gain is at least weakly good for the person who obtains it.","Comparing $G(i,w,p)$ and $G(i,w,q)$ across individuals gives a quantitative framework for choosing between public projects such as a swimming pool and a bus line.","The gun-threat game implies that a legal system can restore cooperative outcomes by making threats unprofitable, so institutions help determine which capabilities are worth having.","Public infrastructure such as a subway system should be counted as part of residents' effective capability wealth, not just as income.","Social decisions in this model depend on a common scale for $v(i,w)$ that allows gains to different people to be compared."],"supporting_citations":[{"why":"Supplies the original capability perspective and the liberal-paradox example that frames the paper's treatment of rights and independence.","marker":"Sen (1970)"},{"why":"Gives the philosophical account of capabilities and functionings that the paper formalizes.","marker":"Nussbaum (2011)"},{"why":"Provides the survey definition of the capability approach quoted in the paper's epigraph.","marker":"Robeyns (2005)"},{"why":"Presents an earlier formal analysis of capabilities that the paper extends.","marker":"van Hees (2020)"},{"why":"Introduces the social-software perspective used for the game-theoretic examples.","marker":"Parikh (2002)"},{"why":"The encyclopedia entry whose normative claims the model is designed to capture.","marker":"Robeyns and Byskov (SEP)"}],"fun_headline_variants":["Well-being is your best reachable world, not your assets","Adding capabilities can shrink others' well-being","Capability approach: your best reachable world, not assets","Why a better bus line helps some but a gun hurts everyone"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the value $v(i,w)$ is a single real number with the same meaning for every person, because the model compares gains across individuals to choose public projects; if values are only ordinal or personal, the social decisions it describes have no well-defined basis.","fun_headline_variants_meta":{"raw":{"variants":["Well-being is your best reachable world, not your assets","Adding capabilities can shrink others' well-being","Capability approach: your best reachable world, not assets","Why a better bus line helps some but a gun hurts everyone"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000814,"raw_usage":{"total_tokens":3448,"prompt_tokens":702,"completion_tokens":2746,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":318,"completion_tokens_details":{"reasoning_tokens":2679}},"tokens_in":318,"tokens_out":2746,"duration_ms":24812,"temperature":1.0,"reasoning_tokens":2679,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T20:41:03.943710+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Find two public projects $p$ and $q$ such that multiplying every $v(i,w)$ by a positive constant reverses the ordering of the induced gains $G(i,w,p)$ and $G(i,w,q)$ for some agent; the model's claim that gains are a meaningful basis for project choice would then fail, because the social ranking would depend on the arbitrary scale of $v$.","supporting_citations":[{"cited_title":"Rational Fools: A Critique of the Behavioral Foundations of Economic Theory","cited_arxiv_id":null,"evidence_quote":"Presents an earlier formal analysis of capabilities that the paper extends."}],"review_version":1}