| paper | d4-conf-wine-2018-0014 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | no |
The paper’s numbered results establish equilibrium properties, truthfulness criteria, and mechanism guarantees, but none asserts a computational result under the stated anchor test. The proposed shared-discount computation is a new synthesis, and moving to a nonatomic population changes the reporting incentives central to the paper.
fails bit a — no named computational result to mirror
No result qualifies as an anchor. The paper’s results concern strategic incentives and mechanisms in finite prediction forums; its continuous outcomes and signal distributions do not make the population continuous.
Under the required anchor rule, I can’t make an anchor-backed positive case: the paper contains no named computational result asserting a problem is in P, hard, FPT, or otherwise classifying its computational complexity.
The nearest results are incentive and mechanism claims. Proposition 5 gives a condition for truthful and prompt play under logarithmic scoring with jointly normal signals; Proposition 7 says truthfulness is never optimal under quadratic scoring in that setting; Propositions 13 and 14 give conditions under which discounting restores truthful, prompt play; and Theorem 15 extends the result to scheduled forums when the relevant Alice–Bob–Alice subgames are truthful and prompt. These are meaningful results, but they do not meet the specified anchor requirement.
The paper does study continuous outcomes and continuous signal distributions, while its population remains finite: the core game has two players, and Theorem 15 concerns a fixed schedule of experts. So the paper does not already use a continuous population. A possible follow-on would be a large forecasting forum with many analysts grouped by complete signal and schedule types, and mass representing each type’s share of the forum. One could ask whether a common discount schedule can make truthful, prompt reporting an equilibrium while keeping expected market-maker loss bounded. But that would be a new computational synthesis problem, not a mirror anchored in a named computational result of this paper.
The main obstacle is that the paper’s strategic results depend on the effects of ordered individual reports. In a continuum, an individual analyst may have zero influence on the public prediction, changing the incentives the paper studies. Defining a continuum game that preserves those incentives would itself require substantial new modeling. I would treat that as a research lead, not claim this paper supplies a qualifying continuous mirror.
The strongest objection is the one the proponent has already identified: the paper states no computational complexity result. Propositions 5 and 7 classify incentives in a fixed Alice–Bob–Alice game; Propositions 13 and 14 give discount conditions for that game; Theorem 15 extends the argument to a finite prediction schedule. They are equilibrium and mechanism results, not complexity classifications. That leaves no named computational result for ChoCo to mirror.
A better candidate question would be: given a high-multiplicity distribution of forecaster types, compute the least common discount factor that makes truthful, prompt reporting an equilibrium while preserving bounded market-maker loss. The difficulty is making “type” complete. If it includes a forecaster’s realized private signal, continuously distributed signals generally give each exact type zero mass. If types describe only signal-generating processes, they omit information that the paper’s incentive conditions may depend on: Proposition 11, for example, requires a bound across signal values and possible deviations. For repeated copies of the same Alice–Bob interaction, changing their mass does not change each copy’s incentive condition. A common discount can be chosen from the worst relevant condition, while the type masses add no new optimization structure.
Putting those copies into one market introduces a different problem. In a continuum, an individual report has zero mass and cannot move the aggregate prediction. The market-score incentive tied to that movement disappears, along with the strict incentive for promptness studied here. Giving each individual a larger influence restores an atomic market; treating a positive-mass cohort as a single reporting unit makes that cohort the strategic agent. Either choice changes the unit of deviation in the paper. Scaling rewards to compensate for vanishing market movements also requires a new bounded-loss argument; the paper’s market-maker guarantee relies on the usual telescoping market-score increments.
Theorem 15 does not resolve this obstacle. Its reduction combines the intervening experts’ information into a composite Bob while preserving a finite sequence of individual opportunities. It supplies a way to analyze scheduled finite games, not a population limit in which each report has vanishing influence. The analyst-forum example makes many forecasters plausible, but their private signals and positions in the reporting sequence remain central to the incentives.
So the proposed shared-discount computation would be a new research question, not a continuous mirror of a computational result in this paper. The negative case is strong on fit with ChoCo’s stated programme; it cannot establish that mean-field prediction-market incentives would never be worthwhile as a separate project.
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.