| paper | Combinatorial Civic Crowdfunding with Budgeted Agents: Welfare Optimality at |
| authors | — |
| venue | AAAI 2023 |
| filed under | coalition · wvg |
| judged by | gpt-5.6-luna / xhigh (triple__luna__xhigh__c2r1) |
| judge confidence | medium |
| authors would recognise it | unclear |
The paper clearly contains computational anchors in Theorem 6 and Corollary 2. However, a nonatomic population makes a unilateral agent non-pivotal, whereas the proposed positive-mass cohort requires coordination and a coalition-deviation interpretation. That interpretation is plausible but not clearly recognizable as the paper's original game.
fails bit none — no continuous question survives
High multiplicity supplies repeated individual agents, not pooled coordination, so the cohort-account formulation may be a new coalition game rather than a population relaxation of the paper's unilateral-agent problem.
fatal: False
It is undecided whether ChoCo may treat a positive-mass homogeneous cohort as one coordinated strategic unit: accepting that yields a plausible Class B mirror, while rejecting it makes the deviations degenerate. A formal programme convention on cohort-level coalition deviations would settle the grade.
The proposed mirror covers the optimal-deviation hardness of Theorem 6 and Corollary 2, but not the equilibrium impossibility and guarantee results, real-valued nonexistence result, or heuristic simulations.
There is a credible positive case, but it is a narrow one: this paper supports a continuous mirror whose main result is Class B hardness transfer, not a claim that continuization makes the problem tractable. My strongest anchor is Theorem 6.
The natural regime is a large municipal crowdfunding platform with many residents but relatively few complete resident types. A type is a valuation vector and budget,
\[
t=(\theta_{t1},\ldots,\theta_{tp},\gamma_t).
\]
It may represent, for example, a neighbourhood or income-and-interest cohort whose members have the same benefits from the projects and the same contribution capacity. A society is a rational distribution \(\mu\) over these types, with perhaps millions of residents but only dozens of distinct types. Projects remain discrete; only the population is continuized. Thus
\[
V_j=\sum_t\mu_t\theta_{tj},\qquad
G=\sum_t\mu_t\gamma_t
\]
are per-capita aggregate valuation and budget. Targets and refund bonuses are normalized per capita as well. Multiplying everything by a common population denominator recovers the corresponding finite instance.
The lead problem is Continuum Cohort Optimal Deviation. Its instance consists of:
If the distinguished cohort contributes \(x_j\) per member to project \(j\), then
\[
C_j(x)=C_{j,-\star}+\lambda x_j,
\]
where \(C_{j,-\star}\) is the fixed contribution of the rest of society. The problem is to find
\[
x\in(\delta\mathbb Z_{\ge 0})^p,\qquad
\sum_jx_j\le\gamma_{t^\star},
\]
maximizing the cohort’s per-member utility
\[
\sum_j\left[
\mathbf 1[C_j(x)\ge T_j](\theta_{t^\star j}-x_j)
+
\mathbf 1[C_j(x)<T_j]R_j(B_j,x_j,C_j(x))
\right].
\]
The output is an optimal contribution vector, or equivalently the answer to the threshold decision question “is the optimum at least \(K\)?”
This is a genuine population mirror of the paper’s optimal-strategy question: the budget constraint, project thresholds, refunds, funded and unfunded utilities, and strategic deviation are all retained. The only change is that repeated agents are represented by mass and a deviation is made by a homogeneous mass of them. It is not participatory budgeting or fractional project selection.
The anchor is Theorem 6, quoted in substance as: “Given an instance of MCC with discrete contributions and for any set of \((R_j)\) satisfying Condition 1, computing optimal strategy for agent \(i'\), given the contributions of \(N\setminus\{i'\}\), is NP-Hard.” This theorem is asserted and proved in the paper; its proof refers to Damle, Padala, and Gujar (2022) for the detailed reduction, so it is not merely cited as prior work.
The continuous cohort problem should also be NP-hard, and the hardness should be classified as Class B. The reduction is from KNAPSACK and its combinatorics live in the number of projects and their contribution/value parameters, not in the number of citizens. A finite hard instance can be represented by rational type masses and normalized monetary quantities. More importantly, the population representation does not remove the project-allocation choice that encodes the knapsack instance. Continuization therefore preserves the difficulty rather than dissolving it.
A second, worthwhile anchor is Corollary 2, which states that, when all agents other than \(i'\) follow a strategy funding the welfare-optimal subset \(P^\star\), computing \(i'\)’s optimal deviation is NP-hard. I would mirror it with Welfare-Preserving Continuum Defection: the input additionally supplies a unique welfare-optimal \(P^\star\), a budget-deficit society, and fixed other-cohort contributions that fund \(P^\star\). The task is to compute the distinguished cohort’s optimal deviation, or decide whether it has a deviation yielding utility at least \(K\), together with the resulting funded set.
This second problem is also Class B. It preserves the paper’s central economic tension more faithfully than the general version: the population is currently implementing the welfare-optimal project set, but a positive-mass type may rationally deviate, and finding the best such deviation remains hard. Corollary 2 is derived in this paper from Theorem 6, rather than being imported from elsewhere.
The paper also gives evidence that the mirror is not artificial on the tractable side. Theorem 3 translates almost verbatim: with Budget Surplus, \(B_j\le V_j-T_j\), and the type-level version of Subset Feasibility,
\[
\gamma_t\ge\sum_j\bar x_{tj}\quad\text{for every type }t,
\]
one can construct a type-symmetric equilibrium funding every project. This is a simple finite-type, closed-form subcase. I do not count Theorem 3 as a complexity anchor because it is a welfare guarantee rather than a named P/FPT/NP-hardness result, but it shows that the continuous formulation retains the paper’s positive equilibrium structure as well as its negative result.
The main weakness is strategic semantics. In an ordinary nonatomic population, one individual cannot change \(C_j\), so the paper’s pivotal-agent reasoning may disappear. My mirror therefore treats a positive-mass homogeneous cohort as the strategic unit. That is not literally the paper’s unilateral-agent game; it is a population-level version of it. The scenario is nevertheless plausible for neighbourhood associations, workplace groups, or collective civic accounts, and the paper’s anonymous refund mechanism makes such type-level aggregation natural. A referee who insists on strictly individual deviations should instead use a tagged-agent hybrid, but then the continuous limit becomes less pure.
A second limitation is that I retain the contribution grid \(\delta\), because Theorem 6 itself is a discrete-contribution result. For genuinely real-valued contributions, Theorem 5 warns that an optimum may fail to exist; the appropriate continuous question would ask for a supremum or an \(\varepsilon\)-optimal deviation. That is a further research question, not something to conceal.
So the honest verdict is: this paper admits a defensible continuous population mirror, centred on optimal strategic deviation in a high-multiplicity civic-crowdfunding society. Its expected outcome is Class B hardness transfer, with tractable equilibrium subcases and new questions about type-level equilibrium existence, minimum destabilizing mass, approximation, and the limit as \(\delta\to0\).
The negative case turns on a mismatch between the paper’s strategic unit and the continuous population model. The paper’s computational results concern one named agent’s unilateral deviation. That agent is pivotal: changing its contribution can move a project across its funding threshold. In a genuine nonatomic population, that pivotality disappears.
Let \(x(t)\) be a contribution policy and \(C_j=\int x_j(t)\,d\mu(t)\). A single agent has measure zero, so changing \(x(t)\) at that agent does not change \(C_j\), the funded set, or any threshold event. If a project is funded, the agent strictly prefers contributing zero. If it is unfunded, the agent chooses contributions only to maximize refunds; it cannot buy funding by itself. Thus the threshold discontinuity that drives Theorems 4–6 vanishes. Under the paper’s PPR refund, the remaining best response is simply a continuous allocation problem over already-funded or already-unfunded projects, not the paper’s optimal-deviation problem.
The proponent’s repair is a positive-mass homogeneous cohort that deviates jointly. But that is not what high multiplicity supplies. High multiplicity says that many agents have the same valuation and budget; it does not make them coordinate their actions or pool their budgets. If they remain separate agents, their individual deviations are null. If they share a “neighbourhood account,” then the model has replaced many paper-agents by a new coalition player with a pooled budget and a collective objective. That may be an interesting coalition-crowdfunding model, but it is not a continuous mirror of the paper’s PSNE or of Theorem 6.
A tagged-agent version does not repair this. One can place a single strategic agent in front of a large continuous background, but then the population is only an exogenous source of the \(p\) aggregate contributions \(C_{j,-i'}\). Theorem 6 already conditions on precisely those aggregates. The only identity whose utility is optimized remains the tagged individual’s identity, so the continuum has not replaced the society relevant to the computational question.
Corollary 2 inherits exactly the same problem. Fixing other agents’ contributions so that they fund \(P^\star\) leaves a single-agent deviation problem. In an ordinary continuum, the deviator cannot destroy \(P^\star\); in a bloc model, it can, but only under the newly introduced coalition solution concept. Calling the bloc a “type” does not bridge that semantic gap.
Theorem 3 supplies no independent rescue. Its type-level Subset Feasibility inequality is easy to state, but its equilibrium conclusion does not survive the ordinary nonatomic interpretation: once projects are funded, every infinitesimal agent wants to reduce its positive contribution because it cannot affect funding. A coordinated-bloc interpretation can restore the conclusion only by changing the strategic model in the same way.
The proponent is right that the paper contains genuine computational results, and a coordinated civic association could motivate a new continuous coalition-deviation problem. That is the weak point in this negative case: such a model is not absurd, and it might deserve study. But every faithful population limit of the paper’s actual unilateral-agent game either loses pivotality and degenerates, or retains a tagged individual and treats the continuum as background. The only version preserving the hardness is a new positive-mass coalition game. On the paper as written, therefore, there is no worthwhile continuous population mirror of its named results.
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.