Incentives for Early Arrival in Cooperative Games

The paper has no numbered theorem, lemma, corollary, or proposition asserting a complexity bound, hardness result, approximation guarantee, or parameterized algorithm. The proposed anchors establish mechanism properties and structural characterizations, so the literal computational bit fails even though an anonymous high-multiplicity extension is conceivable. The opponent also correctly identifies that the proposed synthesis needs new finite-representation and limiting semantics for criticality and Shapley fairness.

The objection that survived

The proposed synthesis is either tautological when its layer decomposition is supplied or lacks a finite input/output model, while infinitesimal criticality and Shapley fairness require new limiting semantics.

What the mirror covers

The proposal concerns the paper's eRFC mechanism properties and the symmetric sufficient case, but supplies no computational mirror for the RFC characterization, GM decomposition, examples, or any other named result.

Open questions for a prover

The case FOR (proponent)

The strongest honest positive case is weak but real: the paper admits a natural high-multiplicity mirror for online value sharing. The important caveat is that the paper contains no named complexity theorem of the kind your programme normally requires. It proves mechanism-design characterizations and constructions, but nowhere states that a problem is in \( \mathrm{P} \), NP-hard, FPT, or similar. Thus, under the literal anchor rule, this paper has no valid complexity anchor. What follows is the strongest salvage, using its named algorithmic theorems as structural anchors rather than complexity results.

The lead anchor is Theorem 5.5, proved in this paper: the extended RFC mechanism \(eRFC\) is Shapley-fair and online individually rational. Theorem 5.6, also proved here, adds that \(eRFC\) incentivizes early arrival whenever every \(0\)-\(1\) component in the greedy monotone decomposition has that property. Corollary 5.7 gives a concrete sufficient case: symmetric monotone games.

A credible mirror is a startup, project, or investment platform with a very large population of agents falling into a small number of complete types. A type might specify an investor’s standardized ticket size and expertise, or a student’s skill bundle. Agents of the same type have the same contribution function and face the same mechanism. There are \(n\gg \tau\) agents, with type proportions \( \mu\in[0,1]^\tau \), where \( \tau \) is fixed or modest. This is precisely a high-multiplicity regime, not a claim that every cooperative game has such a population.

Let \(T\) be the finite type set and let \( \mu_t \) be the available mass of type \(t\). A coalition is represented by a mass vector \(x\in[0,\mu]\), and its value is \(V(x)\), where \(V(0)=0\) and \(V\) is monotone. An arrival history is a coordinatewise nondecreasing path \(z(q)\) from \(0\) to \(\mu\): \(z_t(q)\) is the mass of type \(t\) that has arrived by time \(q\). The action variable is therefore the timing of mass arrivals. The mechanism assigns an irrevocable reward density to each arrived unit of mass after every prefix.

The local-information requirement is inherited directly from the paper: at state \(z(q)\), the mechanism may use the restriction of \(V\) to mass vectors \(y\le z(q)\), but not information about future arrivals. Its constraints are the continuous versions of the paper’s three properties. Total assigned reward at state \(x\) must equal \(V(x)\); each arrived unit’s share may only increase as further mass arrives; delaying a tagged unit while keeping the other arrivals fixed may not increase that unit’s eventual share. Shapley fairness is defined by the high-multiplicity limit: discretize \( \mu \) into \(n\mu_t\) identical agents, use a uniformly random order, normalize each agent’s payoff by \(n\), and take the limit. For differentiable \(V\), the resulting per-unit benchmark is the Aumann–Shapley expression \( \int_0^1 \partial_t V(\lambda\mu)\,d\lambda \).

This gives the following precise lead problem.

Continuous eRFC Synthesis. The input is a finite type set \(T\), rational mass vector \( \mu \), and a monotone value function supplied as a finite layer decomposition \(V(x)=\sum_{k=1}^K \lambda_k g_k(x)\), where each \(g_k\) is a succinctly represented monotone \(0\)-\(1\) function on the mass box \( [0,\mu] \). The task is to output a local-information online mechanism assigning reward densities to every arrival path, or report that none exists, such that the mechanism is Shapley-fair and online individually rational. The output is a policy, not merely an allocation for one fixed arrival order.

This is recognizably the paper’s problem. The \(g_k\) are the continuum counterparts of its \(0\)-\(1\) games, the coefficients \( \lambda_k \) are its decomposition weights, and summing the component mechanisms is exactly the paper’s \(eRFC\) construction. I would expect this restricted version to be Class A: run the RFC rule on each compactly represented layer and aggregate the resulting reward densities. The computational question is whether the layer policies and their integration can be generated in time polynomial in the type description, \(K\), and the encoding length of the rational data. If the layers are given only by an arbitrary coalition oracle, the exponential representation problem returns and no tractability claim follows.

The second, narrower problem is anchored in Theorem 5.6 and Corollary 5.7.

Continuous Early-Arrival Feasibility. Given the same mass-layer representation, decide whether the induced \(eRFC\) policy is incentive-compatible for early arrival for every measurable arrival path, and, if so, output the policy together with a certificate of the property. A natural certificate consists of a proof that every layer \(g_k\) has no continuum analogue of a type becoming the unique critical contributor only after other mass has arrived. Symmetric layers are automatically accepted by the analogue of Corollary 5.7.

I would also expect this restricted problem to be tractable when each layer is a threshold or polyhedral monotone region: the relevant critical boundaries can be enumerated or separated over the fixed-dimensional type-mass space. For arbitrary succinct monotone layers, however, continuum-specific hardness is quite plausible, because the mechanism must quantify over all arrival paths and all possible tagged-unit delays. This is a useful boundary question rather than a defect in the mirror.

The mirror covers the paper’s central positive contribution—online value sharing with Shapley fairness, nondecreasing shares, and early-arrival incentives—not its unrelated literature review or every finite-player example. The high-multiplicity story is particularly plausible for standardized investment cohorts or large student cohorts: the paper itself motivates exactly these applications, and replacing named agents by proportions of repeated contribution types preserves the timing decision and the three desiderata.

The weakest point is serious. The paper’s RFC mechanism is built around individual pivotal and critical players. In an atomless population, one individual has zero mass and cannot change \(V\) by herself; “the first critical player” may therefore disappear or become a boundary-density notion. Exact Shapley fairness also needs normalization and a limiting definition. The proposed repair is to define the continuum mechanism through its finite \(n\)-agent approximants and require convergence of per-unit rewards. If the authors insist on literal individual criticality rather than this high-multiplicity limit, the mirror fails.

So my verdict is: a credible but non-anchored positive case exists. The paper is a good source for a continuous online value-sharing model, especially through Theorems 5.5 and 5.6, but it does not supply the named computational-complexity result that would make this a full ChoCo case.

The case AGAINST (opponent, writing after the proponent)

The positive case fails at the programme’s most important screening criterion: this paper contains no named computational result to continuize. Theorems 5.5 and 5.6, Corollary 5.7, and the RFC completeness theorem are mechanism properties and structural characterizations. They state neither a complexity bound nor an algorithmic decision, search, approximation, or parameterized problem. Algorithm 1 is merely an exponential decomposition over all coalitions; the paper makes no claim about its running time. Under ChoCo’s stated rules, that is decisive.

The proposed “Continuous eRFC Synthesis” does not repair this. If the layer decomposition \(V(x)=\sum_k\lambda_k g_k(x)\) is supplied, the policy is already given symbolically by \(\sum_k\lambda_k\mathrm{RFC}(g_k)\). There is no synthesis problem left beyond applying the definition. If the layers are instead supplied by an oracle or an unspecified “succinct representation,” the problem has no complexity model: a monotone \(0\)-\(1\) function on a continuum is not a finite input until the representation and query operations are fixed. If the mechanism must be output explicitly for every arrival path, the output itself is an infinite object. Thus the proposed anchor is either tautological or underspecified.

There is also a substantive mismatch between the paper’s object and a population mass model. The paper allows an arbitrary set function \(v:2^N\to\mathbb{R}_{+}\), and its critical-player notion depends on which named players are removed. A finite type quotient is legitimate only after imposing exchangeability: \(v(S)\) must depend solely on the vector of type counts. That is a sensible high-multiplicity restriction, but it is a new subclass supplied by the proposed mirror, not a regime identified or analyzed by the paper.

More seriously, the paper’s incentive condition concerns a tagged player’s position in a discrete arrival permutation. In a mass state \(x\), delaying one infinitesimal unit changes no coordinate of \(x\). Any mechanism depending only on the mass path therefore sees exactly the same path before and after the deviation; early-arrival incentive compatibility becomes equality, hence vacuous. To preserve a nontrivial deviation, the model must retain the tagged unit’s microscopic rank or arrival time in addition to the mass state. That reintroduces the individual-level object the continuization was supposed to replace.

The same problem undermines Theorem 5.5. RFC rewards the first critical individual. In an atomless population, a critical individual has zero mass, and a monotone \(0\)-\(1\) layer typically changes value at a threshold without any positive mass unit being responsible for the jump. With an unscaled value, a tagged player’s reward and timing incentive vanish as the population grows. With extensive scaling, the finite-population pivotal reward becomes a singular boundary allocation. Neither limit is the paper’s eRFC mechanism, and the paper proves no convergence or representation theorem for either one.

Replacing Shapley fairness by an Aumann–Shapley integral does not solve this; it changes the benchmark. A finite uniformly random permutation has a well-defined Shapley value, whereas a uniformly random order of an atomless population has no canonical first arrival. A high-multiplicity sequence can define a limit under smoothness and a chosen normalization, but different scalings and regularizations need not yield the same online allocation. That may motivate new cooperative-game theory, but it is not a continuous mirror of Theorem 5.5.

Theorem 5.6 and Corollary 5.7 fare no better. Theorem 5.6 is only a conditional implication about a decomposition already chosen; Corollary 5.7 gives symmetry as a sufficient case. The proposed “Continuous Early-Arrival Feasibility” is therefore a newly invented verification problem, not a computational question raised by the paper. For arbitrary succinct layers, its complexity is determined almost entirely by the chosen representation. For threshold or polyhedral layers, it becomes a geometric check designed by the continuizer. Either version may be interesting independently, but neither is an anchored ChoCo result.

A large cohort of standardized investors or students could indeed make a legitimate high-multiplicity cooperative game. That is not the objection: high multiplicity is a valid modelling regime, and existing high-multiplicity work would support it. The objection is that, after making the game anonymous enough for mass states, the paper’s central objects—individual criticality, discrete first arrival, and permutation-based Shapley fairness—either disappear, become vacuous, or require microscopic information to be restored.

So the strongest negative verdict is that this is not a worthwhile ChoCo mirror of the paper. It may inspire a separate continuum theory of anonymous online cost sharing, with carefully chosen scaling and an Aumann–Shapley benchmark. But that would be a new axiomatic mechanism-design programme, not a computational continuation of any named result in this paper. The universal claim that no interesting mass model could ever be formulated is not mathematically airtight; the decisive point is narrower and sufficient: this paper supplies no computational anchor, and the proposed repairs do not create a faithful one.

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.