| paper | d4-conf-wine-2017-0027 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | medium |
| authors would recognise it | yes |
Theorem 3.3
Given rational \(B,p,\varepsilon>0\) and a finite-support rational distribution \(\mu\) over types \(t=(\alpha_t,\delta_t)\), with rational \(0<\alpha_t,\delta_t<1\) and common share success probability \(p\), compute a finite-support rational payout sequence \(x_i\ge0\), \(\sum_i x_i\le B\), whose welfare \(p\sum_t\mu_t\sum_{i\ge0}\delta_t^i x_i^{\alpha_t}\) is within \(\varepsilon\) of the supremum over feasible payout sequences.
A large pool represented by finitely many contributor types \(t=(\alpha_t,\delta_t)\) with mass \(\mu_t\), all submitting statistically identical shares; choose one common fixed-rule payout \(x_i\) per share age to maximize mass-weighted steady-state discounted expected utility under \(\sum_i x_i\le B\).
Cross-type welfare depends on cardinal utility normalization, which the paper does not justify across miners; specifying \(u_t(x)=x^{\alpha_t}\) and weights \(\mu_t\) makes the extension precise but leaves that welfare convention open.
fatal: False
Covers Theorem 3.3’s power-utility pool design extended from one \((\alpha,\delta)\) profile to finite-support \(\mu\); it does not cover generalizations of Theorems 3.1–3.2, the PPLNS comparisons, or incentive compatibility.
The strongest positive case is Risk-Mix Pool Design\(_\infty\), anchored in Theorem 3.3, proved in this paper. There is no named result here classifying a problem as polynomial-time, NP-hard, or FPT. Theorem 3.3 is instead an exact optimization result: for utility \(u(x)=x^\alpha\), \(0<\alpha<1\), and discount factor \(\delta<1\), a fixed-rule geometric pool is steady-state optimal, with allocation \(X_i=B(1-\delta^{1/(1-\alpha)})\delta^{i/(1-\alpha)}\) to the share \(i\) places before a block. That gives a concrete computational anchor—the pool-design problem and its optimal allocation—though it is not a complexity theorem. [The paper](https://arxiv.org/abs/1703.03846)
A plausible regime is a large mining pool or crowdsourced-computation service with many small contributors but relatively few utility profiles. A type \(t\) records a contributor’s risk parameter \(\alpha_t\) and discount factor \(\delta_t\); assume all types submit equivalent shares with the same success probability \(p\). The society is a finite-support distribution \(\mu\), where \(\mu_t\) is the fraction of contributed shares from type \(t\), and the number of contributors is much larger than the number of types. This fits the paper’s focus on large pools of small miners, while making population composition part of the question.
Given rational \((\mu_t,\alpha_t,\delta_t)\), block reward \(B\), success probability \(p\), and additive error \(\varepsilon>0\), choose a fixed-rule allocation \(x=(x_i)_{i\ge0}\). Here \(x_i\) is the reward paid to a share \(i\) submissions before a successful block, with \(x_i\ge0\) and \(\sum_i x_i\le B\). Type \(t\)’s steady-state discounted expected utility is \(p\sum_{i\ge0}\delta_t^i x_i^{\alpha_t}\). Define population welfare as \(W_\mu(x)=p\sum_t\mu_t\sum_{i\ge0}\delta_t^i x_i^{\alpha_t}\). The task is to return a feasible allocation with welfare at least \(\sup_y W_\mu(y)-\varepsilon\). This keeps the paper’s allocation rule and payoff objective, while making the society itself a distribution over complete contributor types.
I expect the finite-horizon version to be Class A: its objective is concave and its budget constraint is linear, so it is a convex-optimization problem. The infinite tail can be bounded because every \(\delta_t<1\). A uniform polynomial-time claim in the binary input length needs more care, however: when a discount factor is very close to \(1\), the horizon needed for a given error can be large. Natural next questions are whether the mixed-type optimum has a useful closed form, whether it can be approximated efficiently under binary encoding, and how participation constraints change the solution.
This case covers only the optimal fixed-rule pool-design result, not the paper’s other claims. The authors would likely recognize the problem as a population-level extension of their social-welfare objective, but not as a result already established by Theorem 3.3. The weakest point is that their theorem treats one miner utility profile, whereas this mirror makes welfare a mass-weighted average across profiles; that extension is plausible, but it must be presented as the new question rather than attributed to the paper. Also, the paper’s stochastic share process is not itself population continuization: the proposed mirror is the distribution \(\mu\) over contributor types.
The only strong anchor is Theorem 3.3: for one miner utility \(u(x)=x^\alpha\) and discount factor \(\delta\), it gives a closed-form optimal geometric payout rule. Theorem 3.2 exposes the underlying task as maximizing a separable concave objective over a payout sequence subject to one budget constraint. Neither theorem classifies the complexity of a computational problem, and the paper supplies no discrete complexity result to compare against. [The paper’s results](https://arxiv.org/html/1703.03846v1)
For a homogeneous population, adding mass changes nothing: the theorem’s per-share utility and optimal schedule do not depend on how many miners there are. The paper assumes large pools of small miners so a share’s utility can be treated as separable; that makes a large population plausible, but it does not itself create a population-complexity question.
The proposed risk-mix version is more substantial. Yet its welfare objective needs an extra normative choice. The paper notes that von Neumann–Morgenstern utility functions are defined only up to positive affine transformations. Rescaling one type’s utility preserves that type’s preferences while changing its influence on the proposed aggregate objective, and potentially changing the optimal pool. Fixing a shared normalization—such as the specified power functions and share-based weights—makes the optimization well-posed, but adds a cross-type welfare criterion the theorem does not establish. The resulting problem may be worth studying; its computational content comes from that new heterogeneous-agent model, rather than from continuizing the paper’s existing result.
I would argue against prioritizing this as a ChoCo mirror of Theorem 3.3: the paper offers an economic pool-design result, while its natural homogeneous version has no population-sized computational question. I cannot honestly defend the stronger claim that no worthwhile mirror exists in any scenario. With a chosen utility normalization, the finite-support mixed-type problem is credible and could support approximation and bit-complexity questions. That is the negative case’s main weakness.
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.