Fairly Dividing a Cake after Some Parts Were Burnt in the Oven

The proposed cohort version is a recognisable extension of the paper's three-agent cake-division question, but Theorem 4.13 and Lemma 6.6 are existence results rather than named computational results. Section 5 contains an algorithmic observation, yet it is unnumbered, oracle-based, and lacks a finite-input complexity theorem. Therefore bit (a) fails, so the mirror is red.

The objection that survived

The proposed exact cohort constraints make the masses \(\mu_t\) disappear, while the paper supplies no qualifying numbered computational theorem to mirror.

What the mirror covers

The proposed mirror covers connected envy-free division for three homogeneous valuation types and the prime-agent existence extension, but leaves the paper's topological lemmas, conjecture, and cited protocols without a qualifying computational-population result.

Open questions for a prover

The case FOR (proponent)

The strongest honest case is a qualified one: the paper has a natural population mirror, but it has no qualifying numbered computational anchor. Its numbered results are existence, topological, or structural results, not claims of membership in \(P\), hardness, approximation complexity, or query complexity. I therefore would not present this as a fully compliant ChoCo case.

The best lead anchor is Theorem 4.13, proved in this paper:

“For \(n=3\) selective agents, there always exists a connected envy-free division of a mixed cake.”

Here a mixed cake is the interval \([0,1]\), divided into three connected intervals; each selective agent has a continuous preference-selection function over the partition and must receive one of their preferred pieces. Additive agents with integrable, possibly negative value densities are a special case.

The corresponding population mirror is Continuous Mixed-Cake Envy-Freeness\(_3\). An instance consists of three positive rational masses \(\mu_1,\mu_2,\mu_3\) summing to \(1\), representing three large homogeneous cohorts, together with one complete preference type for each cohort. In the additive version, type \(t\) has a rational piecewise-constant density \(v_t:[0,1]\to\mathbb R\). The decision variables are two cut points \(0\le z_1\le z_2\le1\), the resulting connected pieces \(X_1,X_2,X_3\), and an assignment \(\pi\) of types to pieces. The solution must satisfy

\[ \int_{X_{\pi(t)}}v_t(x)\,dx \ge \int_{X_j}v_t(x)\,dx \qquad \text{for every positive-mass type }t\text{ and every }j. \]

Equivalently, one may minimize the population-weighted envy

\[ \sum_t \mu_t \max_j \left( \int_{X_j}v_t(x)\,dx - \int_{X_{\pi(t)}}v_t(x)\,dx \right)_+, \]

with the exact problem asking whether the optimum is zero and, if so, returning the cuts and assignment.

A plausible regime is a very large estate or contiguous land strip divided among three large beneficiary cohorts. Millions of heirs may fall into only three valuation types because they share tax status, zoning constraints, or location: for example, agriculture-preferring, development-preferring, and mixed-valuing cohorts. Thus \(N\gg\tau\), with \(\tau=3\), and \(\mu_t\) records the population share of each cohort. The paper’s own burnt-cake and taxed-land examples make this scenario recognizable to its authors.

The expected direction is tractable in an approximate query model. Section 5 gives an unnumbered binary-search procedure; for three agents, the paper states an \(O(1/\delta)\)-query construction for a suitable triangulation. For rational piecewise-constant densities, this suggests an approximate algorithm, although the paper does not prove a bit-complexity bound. The exact existence claim follows directly from Theorem 4.13.

The main co-modification is unavoidable: the continuum object is a cohort-level connected bundle, not a literal allocation of separate connected intervals to every atomless individual. This is best understood as a group or per-capita version of cake division. If one insists that every one of \(N\) individuals must receive a distinct connected subinterval of one fixed finite cake, the limit becomes degenerate and Theorem 4.13 does not provide the desired mirror. The positive case survives only if the authors accept homogeneous cohorts as the relevant claimants and permit the corresponding type-level average interpretation.

Further questions would include whether the prime-\(\tau\) extension suggested by Lemma 6.6 yields a genuine algorithmic result, what happens for composite \(\tau>3\), and whether maximizing population-weighted welfare subject to envy-freeness is polynomially solvable.

So the substantive mirror is plausible, but its weakest point is decisive under the stated rubric: Theorem 4.13 is not a computational anchor, and the paper contains no numbered result that is. The case is therefore a credible population-level extension of the paper’s fair-division problem, not a qualifying ChoCo mirror of a computational result.

The case AGAINST (opponent, writing after the proponent)

The negative case is unusually strong because the paper has no qualifying computational anchor. Theorem 4.13 is an existence theorem: it says that three named agents with continuous preferences admit a connected envy-free division. It asserts neither a finite-input algorithm nor a complexity bound. Asking whether this existence statement survives for a distribution of agents is a mathematical extension, but not a ChoCo mirror of a computational result.

The proponent’s best population model does not repair this. If three positive-mass types receive three connected pieces, the exact envy-free constraints are

\[ V_t(X_{\pi(t)})\ge V_t(X_j) \]

for every type \(t\) and piece \(j\). The masses \(\mu_t\) disappear entirely whenever \(\mu_t>0\). Thus the proposed “continuous society” is simply the original three-agent problem with agents renamed as cohorts. The population distribution contributes no high-multiplicity structure. Adding a population-weighted envy objective introduces a new optimization problem, not a mirror of Theorem 4.13.

The faithful alternative is to replicate each type into many individuals and require every individual to receive a connected piece. But then the allocation has \(N\) pieces and \(N-1\) cuts, not three pieces and two cuts. As \(N\) grows, this is a different cake-cutting problem; if the cake is scaled with \(N\), that scaling is an additional supply-and-allocation model rather than a limit of the paper’s theorem. Giving an entire cohort one common piece instead changes individual cake division into group ownership. That may be a sensible problem, but it is not what the paper proves.

Section 5 does not provide a stronger anchor. Its \(O(1/\delta)\) statement is an unnumbered query-complexity observation for exactly three agents with oracle access to preferences. There is no finite encoding, bit-complexity analysis, or population parameter. A three-type distribution leaves this procedure essentially unchanged; the masses again do not enter. Restricting to rational piecewise-constant densities could create a new computational problem, but that would be an added formalization, not a computational result from the paper.

Lemma 6.6 is weaker still. It is a topological existence result for prime numbers of agents, not an algorithm or complexity theorem. Varying the number of agents is not the same as introducing many agents per type, and the lemma supplies no finite representation of the arbitrary selection functions it concerns.

A large estate with many homogeneous heirs is a plausible application story, so one cannot honestly claim that every population interpretation is nonsensical. But every promising interpretation either collapses to the already-studied fixed-agent problem or changes the allocation semantics enough to become a new fair-division programme. The paper therefore offers no worthwhile continuous-population mirror in the programme’s computational sense.

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.