On the Fairness of Additive Welfarist Rules

The paper’s numbered results are fairness characterizations, not named algorithmic or complexity results. The unnumbered EF1 algorithm and cited MNW computational results do not satisfy the anchor rule. The proposed high-multiplicity configuration problem is a sensible follow-up, but it is not a computational mirror of a result proved in this paper.

The objection that survived

The configuration optimization and universal-optimum decision problem are new tasks appended to Theorem 4.7, so the anchor lacks the required computational character.

What the mirror covers

The proposed model covers only the binary-instance EF1 characterization in Theorem 4.7; all other characterizations and cited computational results remain without a computational result from this paper to mirror.

Open questions for a prover

The case FOR (proponent)

Under the programme’s strict anchor rule, this paper has no qualifying computational result. Theorems 3.1, 3.5, 3.6, 4.1, 4.7, and 4.11, and the propositions around them, characterize which functions \(f\) guarantee EF1; they do not assert that an allocation problem is in \(P\), \(NP\)-hard, FPT, or otherwise computationally classified. The polynomial-time EF1 statement in the introduction is unnumbered and attributed to [5,15]. MNW hardness appears only in cited papers [1], [9], and [13]. Thus there is no honest named computational anchor from which to build the requested case.

The closest positive route, explicitly outside that strict gate, is Theorem 4.7, proved in this paper. It gives a natural high-multiplicity extension.

Call the problem \(\mathrm{Binary\text{-}Welfarist\text{-}EF1}_{\infty}\). There are finitely many agent types \(\Theta\), rational masses \(\mu_\theta\) summing to \(1\), and finitely many good types \(q\in Q\), with rational per-capita supplies \(\lambda_q\). Each type has a binary valuation vector \(v_{\theta q}\in\{0,1\}\). Goods remain indivisible: a bundle is an integer vector \(B\in\mathbb Z_{\ge0}^{Q}\), not a fractional bundle. Let \(z_{\theta,B}\) be the mass of type-\(\theta\) agents receiving exactly bundle \(B\). Feasibility is

\[ \sum_B z_{\theta,B}=\mu_\theta \]

for every \(\theta\), and

\[ \sum_{\theta,B} B_qz_{\theta,B}=\lambda_q \]

for every good type \(q\). The welfarist objective is

\[ \max_z \sum_{\theta,B}z_{\theta,B}\, f\!\left(\sum_q v_{\theta q}B_q\right). \]

The computational question is to find an optimal \(z\), and to decide whether every optimum is EF1. The latter means that for every pair of positive-mass configurations \((\theta,B)\) and \((\theta',B')\), there is some \(q\) with \(B'_q>0\) such that

\[ \sum_q v_{\theta q}B_q \;\ge\; \sum_q v_{\theta q}B'_q-v_{\theta q}. \]

This is recognizably the authors’ problem in a plausible regime: a large population of customers or recipients falls into finitely many complete preference/eligibility types, while inventory consists of repeated indivisible goods. Clearing denominators turns rational masses and supplies into cloned finite agents and goods, preserving the objective and EF1 exactly. The model does not permit a good to be split; it only permits different clones of one type to receive different whole bundles.

The expected fairness characterization should transfer: the universal EF1 guarantee remains Condition 4,

\[ \Delta_{f,k}(1)>\Delta_{f,k+1}(1) \]

for every \(k\ge0\). The counterexamples in the proof of Theorem 4.7 already use binary, finite, rational instances, so failure of Condition 4 survives the mirror. The optimization problem itself is a new configuration-LP/pricing question: likely tractable under bounded type structure, but not classified by this paper.

The weak point is decisive: this is a faithful, plausible continuous extension of a named fairness theorem, not a continuous mirror of a named computational result. Presenting it as a Class A, B, or C theorem would overstate what the paper supports.

The case AGAINST (opponent, writing after the proponent)

The negative case is decisive under ChoCo’s anchor rule: this paper contains no qualifying computational result to continuize.

Theorem 4.7, the proponent’s anchor, is not a complexity theorem. It characterizes the functions \(f\) for which every optimal additive-welfarist allocation is EF1 on binary instances. The same is true of Theorems 3.1, 3.5, 3.6, 4.1, and 4.11 and the surrounding propositions. None defines a computational input-output problem or proves membership in \(P\), hardness, approximation, or parameterized tractability. The paper’s only polynomial-time statement—that an EF1 allocation exists and can be computed efficiently—is an unnumbered fact attributed to earlier work, and it concerns finding some EF1 allocation, not computing an additive-welfarist optimum or characterizing the admissible welfare functions. The MNW computational results are likewise only cited, not proved here.

The proposed \(\mathrm{Binary\text{-}Welfarist\text{-}EF1}_{\infty}\) therefore changes the paper’s question. It takes an axiomatic universal statement about \(f\) and appends a new optimization problem: compute a welfarist optimum over a configuration distribution, or decide whether every optimum is EF1. That may be a sensible fair-division problem, but it is not a continuous mirror of a computational result in this paper. The paper itself supplies neither the computational formulation nor an algorithmic claim to preserve.

The strongest repaired version still does not fix this. One could choose a concrete function such as \(f=\log\), fix finite agent and good types, use rational masses and repeated indivisible goods, and ask for exact or approximate optimization. That yields a legitimate high-multiplicity fair-division project. But it is an externally designed follow-up, not an anchor furnished by Theorem 4.7. The theorem’s content concerns only the discrete utility increments \(f(k+1)-f(k)\); replacing repeated agents by masses does not generate a new theorem-level computational object.

Indeed, with rational masses, the proposed model is exactly equivalent to a finite cloned instance after clearing denominators. The objective and EF1 constraints are preserved. This equivalence is not a defect in ChoCo—high multiplicity is explicitly in scope—but it shows that the proposed construction is a compressed fair-division formulation, not evidence that this paper has computational population-continuization content. Any interesting complexity would come from the newly introduced configuration optimization, not from the paper’s fairness characterizations.

So I would reject Theorem 4.7 as an anchor and give this paper no ChoCo mirror on the programme’s stated terms. The universal claim that no conceivable high-multiplicity fair-division scenario could be worthwhile is, however, too strong: the configuration-LP problem is plausibly worth studying in its own right. The honest negative conclusion is narrower but firm—this paper does not motivate it as a continuous computational-social-choice result.

The adversarial triple: the proponent anchors on up to three named results; the opponent sees that case and must defeat every anchor; the judge decides which case convinced it. These are the pipeline’s own outputs, generated by tools/triple_run.py — no human edited them. The paper’s own text is not reproduced here beyond the quoted statement above.