The Fairness Fair: Bringing Human Perception into

The paper contains no numbered theorem, lemma, corollary, or proposition asserting a computational result. Population-EF1 Allocation is a sensible new high-multiplicity problem, but it is not a mirror of a named result in this paper. Therefore bit (a) fails regardless of the model’s potential value.

The objection that survived

The proposed model fixes EF1 and omits the paper’s central questions about human judgement, procedure, information, and explanation.

What the mirror covers

Covers only a newly formulated population version of the paper’s EF1 fair-division background; it leaves the perception, procedural justice, dynamic, privacy, and verification agenda uncovered.

Open questions for a prover

The case FOR (proponent)

The strongest honest positive case is conditional: the paper’s subject admits a sensible high-multiplicity continuous mirror, but the paper itself supplies no eligible computational anchor.

There is no numbered Theorem, Lemma, Corollary, or Proposition in the paper asserting a complexity or algorithmic result. Its only numbered item is “Example 1.” The closest statement is the unnumbered claim that EF1 allocations exist and can be computed in polynomial time, explicitly attributed to Lipton et al. (2004) and Caragiannis et al. (2019). The EFX and MMS remarks are likewise summaries of external literature or open problems, not results proved here. Therefore there are zero anchors under the programme’s rules.

If an agenda-derived mirror is still useful, the cleanest one is Population-EF1 Allocation. Let \(T\) be a finite set of complete valuation types, with rational masses \(\mu_t\) summing to one. A type includes the entire additive valuation vector \(v_t\), together with any information or perception parameters the model uses. Let \(G\) be a finite set of indivisible good categories, with per-capita supply \(r_g\). An allocation is a mass distribution \(x_{t,B}\), where \(x_{t,B}\) is the fraction of type \(t\) receiving bundle \(B\subseteq G\). It must satisfy

\[ \sum_B x_{t,B}=\mu_t,\qquad \sum_{t,B:g\in B}x_{t,B}\le r_g. \]

Every pair of bundles receiving positive mass must satisfy EF1 from the perspective of every observing type: for all positive \(x_{t,B}\) and \(x_{u,B'}\),

\[ v_t(B)\ge v_t(B') \quad\text{or}\quad \exists g\in B'\; v_t(B)\ge v_t(B'\setminus\{g\}). \]

The question is whether such an \(x\) exists; an optimization version can maximize total welfare \(\sum_{t,B}x_{t,B}v_t(B)\) subject to these constraints.

This is a genuine population continuization. Each individual still receives an indivisible bundle; only the population is represented by mass. A plausible regime is a university, employer, or public-service system with millions of participants but a few dozen recurring cohorts—degree programmes, household profiles, accessibility needs, or job families—with many repeated copies of each resource category. Thus \(N\) is enormous while \(|T|=\tau\) is modest. The type is complete, so grouping agents does not discard any feature used by the problem.

I would expect the pure high-multiplicity feasibility problem to be tractable in structured regimes, through configuration-style optimization and pricing over bundles; additive valuations make the basic pricing step easy. In unrestricted form, however, combinatorics in the number of goods may remain, so hardness could transfer from discrete allocation variants. I would not expect continuum-specific hardness without a separate reduction.

The weakest point is faithfulness to the paper’s central contribution. EF1 and additive valuation capture its fair-division substrate, but not yet human perception, procedural justice, framing, or explainability. To capture those, one would extend each type with a rational perception function \(h_t(B,p)\), where \(p\) is a chosen procedure or explanation, and optimize aggregate perceived fairness \(\sum_t\mu_t\sum_Bx_{t,B}h_t(B,p)\). That is a promising continuous research question, but it is newly formulated rather than a mirror of a named result in this paper.

So the positive conclusion is: the paper’s agenda has a credible population mirror, especially for repeated fair-allocation markets, but it cannot support the requested anchor-based case because it contains no named computational result of its own.

The case AGAINST (opponent, writing after the proponent)

The strongest negative point is dispositive under the programme’s rules: this paper has no eligible computational anchor. It contains no Theorem, Lemma, Corollary, or Proposition asserting a complexity or algorithmic result. The statement that EF1 allocations exist and are computable in polynomial time is reported from earlier papers; the EFX and MMS statements likewise summarize external literature and open problems. The paper’s own contribution is an agenda about human perception, cognition, procedure, explanation, privacy, and verification—not a computational theorem about any of them.

That makes “Population-EF1 Allocation” a new problem inspired by the paper, not a continuous mirror of one of its results. It is a perfectly intelligible high-multiplicity fair-division model: repeated valuation types, mass assigned to indivisible bundles, and per-capita supplies. I would not object that agents with the same type are being grouped; under the programme’s definition, that is exactly what high multiplicity means. Nor would I object that the resulting question might be mathematically interesting in its own right.

But it mirrors only the paper’s standard fair-division background, not its stated subject. EF1 is used in the paper as an example of an existing axiom whose alignment with human judgement is uncertain. The proposed feasibility problem takes EF1 as fixed and asks whether a mass allocation satisfies it. It says nothing about whether EF1 is perceived as fairer than proportionality, EFX, MMS, inequality aversion, or a procedurally generated alternative. The paper’s central question—what fairness concepts people actually endorse—is absent.

The proposed repair, adding a perception function \(h_t(B,p)\), does not solve that mismatch. If perception is a fixed property of each type and depends on the final bundle and a specified procedure, then the model is simply a high-multiplicity allocation or mechanism-design problem with an additional utility function supplied as input. The human component has been converted into another valuation table. The computation may be legitimate, but it does not investigate perception; it optimizes whatever perception model the modeller assumed.

A stronger repair could let types encode valuation, cognitive traits, role, information, privacy sensitivity, and procedural preferences, while the allocation includes explanations, deliberation, disclosure, and agency. That is still not a rescue of this paper’s computational content. The paper’s phenomena are often relational and endogenous: a participant’s judgement depends on what others know, whether the observer has “skin in the game,” how the outcome was framed, what procedure occurred, and how explanations alter beliefs. A marginal distribution over static agent types does not contain those histories or conditional information structures. If all of them are packed into a “complete type,” the result is either an arbitrary, highly elaborate population mechanism model or a standard high-multiplicity problem with black-box utilities. Neither is a computational mirror supplied by the paper.

The same problem appears for the paper’s other themes. Dynamic arrivals and redistribution require state histories; privacy-preserving verification requires information structures; procedural justice requires comparing mechanisms, not merely allocations; explainability requires modelling explanations and their interpretation. One can formulate continuous versions of each, but there is no unique computational question selected by the paper, and no named result whose complexity landscape ChoCo would be charting. The continuum would be supplying a new application domain rather than revealing a computational phenomenon latent in this article.

There is also no reason to privilege the proponent’s particular population-EF1 definition. Requiring every pair of bundles in the support to satisfy EF1 for every observing type is stronger and more unusual than ordinary EF1; ex-ante EF1, almost-everywhere EF1, distributional EF1, and stochastic notions are all plausible alternatives. That flexibility is not itself a formal objection—the programme rightly permits natural redefinitions—but here it highlights that the proposed target is being designed from scratch. The paper provides no computational result or modelling commitment that selects among these variants.

So the honest negative conclusion is not that continuous fair division is impossible, nor that Population-EF1 could never be worthwhile. It is that this paper does not warrant such a mirror. Its only computational material belongs to cited fair-division literature, and its original contribution is empirical and conceptual rather than algorithmic. The positive case therefore establishes, at most, a sensible new ChoCo problem loosely motivated by the paper—not a worthwhile continuous mirror of the paper itself.

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.