| paper | d4-conf-atal-2011-0177 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | no |
Corollary 1 gives an asymptotic CSE-to-equilibrium convergence claim, so the computational-character gate is met. But in a continuum, one principal and their extra identity have negligible mass, changing the finite-player Bayesian equilibrium; assigning shill capability to positive mass instead studies prevalence, not one shill changing a rival’s beliefs.
fails bit b-changes — the continuum limit changes the question
A single shill cannot change aggregate bids or rivals’ beliefs in the nonatomic limit; giving that role to positive mass changes unilateral manipulation into a population-prevalence question.
fatal: True
The proposed extension would cover the CSE-based bidding-strategy analysis in Sections 3–4; it leaves the numerical revenue, surplus, and profit comparisons in Section 5 without numbered computational guarantees.
The best positive case is narrow: the paper’s named anchor is about approximate equilibrium computation, not a hardness result. Its reported effects on revenue and bidder profit are numerical findings, not numbered theorems.
The anchor I would use is Corollary 1, cited from Armantier, Florens, and Richard [1], rather than proved in this paper. It says that if the strategy space \(S\) is compact, expected utility \(\tilde U\) is continuous, and a constrained strategic equilibrium (CSE) exists for every parameter \(k>0\), then a Nash equilibrium exists; moreover, every sequence of those CSEs has a subsequence converging to a Nash equilibrium. The paper uses the CSE method to approximate bidding equilibria in its auction subgames. There is no comparable numbered hardness result to anchor.
A plausible mirror is the continuous-population false-name bidding equilibrium problem. Consider a large platform auctioning batches of two complementary resources, \(X\) and \(Y\). A type \(t\) specifies a bidder’s values for each package, information and beliefs about the apparent composition of rivals, and whether the bidder can submit bids through two linked identities. Let \(\mu_t\) be the fraction of principals of each type, with rational masses summing to one. The platform’s supplies scale with the population, so per-capita supplies \(q_X,q_Y\) remain fixed as the number of principals grows. This supply scaling, and the resulting fractional allocation averages by type, are co-modifications needed to make the large-market question well-defined.
Each type’s strategy prescribes its bid or pair of bids after its information is observed. The auctioneer allocates the available mass of goods to maximize reported revenue, subject to supply constraints, and winning identities pay their bids. A principal using two identities receives the combined value of what those identities win and pays both bids. The solution sought is a type-contingent equilibrium strategy profile and beliefs: at every information set, no type can gain more than \(\varepsilon\) in expected utility by deviating, and beliefs are Bayes-consistent wherever possible. One can ask for such profiles in compact, finite-dimensional bid-strategy families indexed by \(k\), as in the paper’s CSE approach, and whether a convergent subsequence approaches an equilibrium of the unrestricted bidding game.
This scenario has a credible high-multiplicity regime: many firms bidding in recurring, large-batch procurement or capacity auctions, with relatively few valuation and information templates. Mass is the share of principals in each complete type, not the number of aliases. The problem retains the paper’s central ingredients—first-price package bids, incomplete information, and the option to split bids across false names. The authors could reasonably recognize it as a large-market version of their equilibrium-bidding question.
I would tentatively expect the restricted \(\varepsilon\)-equilibrium problem to be tractable when the support and bid-strategy representation are modest: type-level masses remove dependence on the raw number of participants, and CSE is the paper’s chosen route to numerical equilibrium approximation. That is a research hypothesis, not a consequence of Corollary 1. The corollary guarantees existence and subsequential convergence under its assumptions; it supplies no polynomial runtime or convergence rate. Establishing those bounds would be a further question. Other natural questions are whether the shill advantage survives as the market grows, how it depends on the mass of shill-capable types, and whether the continuous equilibrium approximates equilibria of finite markets.
The mirror covers the paper’s CSE-based analysis of bidding strategies in Sections 3–4. It does not elevate the unnumbered revenue and profit comparisons in Section 5 into formal anchors, or claim to mirror the entire paper.
The soft spot is substantial: the paper studies one shill facing one or two other bidders, so false naming changes the perceived number of rivals by a whole bidder. In a continuum, one principal’s extra identity is negligible; preserving a material effect may require a positive mass of shill-capable principals. Scaling supplies and passing to aggregate allocation also changes the strategic environment. The mirror is therefore strongest as a question about whether false-name effects persist in a large market, not as a claim that the continuum reproduces the paper’s specific three-bidder game.
The proposed anchor does not give a complexity result from this paper. Corollary 1 is imported from Armantier et al.; it gives an existence and subsequential-convergence claim for constrained equilibria under stated assumptions. It supplies no runtime or approximation guarantee, and the paper’s own contribution is numerical analysis of a game with two or three actual bidders. A continuous computational question could still be built from that analysis, but it has to preserve the strategic effect the paper studies.
That is where the strongest negative case lies. Bidder 2’s extra identity matters because it changes what a particular rival thinks about the number and kind of opponents. In a continuum of principals bidding in one scaled market, one principal and one extra identity have zero mass. They cannot materially change the observed distribution of bids or a rival’s beliefs about it. Scaling supplies and using type-level allocation averages makes a coherent large-market model, but it turns the paper’s individual shill into a negligible price taker. The effect central to the paper’s results—one shill improving their position by inducing a rival to bid as if facing multiple real bidders—has disappeared.
The strongest repair is to put positive masses of naive and shill-capable principals into the market, as the proponent suggests. That supports a coherent question about whether the *population prevalence* of shill-capable types changes equilibrium bids, revenue, or allocations. But it asks how a mass of agents affects the market’s equilibrium. The paper asks how one bidder’s false name changes another bidder’s beliefs and that bidder’s own payoff. Treating a whole mass as one coordinated shill would restore the effect only by giving a group the strategic agency of one player. It would be a different game.
The other natural construction keeps each auction finite and takes a continuum of repeated auctions or auction settings. Then each copy retains the paper’s identity manipulation, but the continuum describes a distribution over games; it does not continuize the population inside an auction. Those options leave a real, though not conclusive, negative case: the paper’s distinctive individual manipulation either vanishes in a nonatomic market or remains inside finite games when the continuum is moved outside the auction. A large-market study of shill prevalence could still be worthwhile, but its connection to this paper’s central result is weaker than the proponent’s case suggests.
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.