| paper | d4-conf-aaai-2020-0827 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | no |
Theorem 4.1 gives an explicit witness to failure of Bayesian resilience for \(q<1\), but it asserts neither an algorithm nor a complexity bound or hardness result. The proposed pool packages the paper's prior as masses while each auction remains a finite \(n\)-bidder game, so it does not yield a continuous-population mirror of a computational result.
fails bit a — no named computational result to mirror
Theorem 4.1 proves an equilibrium boundary, not a computational result, so it fails the required anchor test.
fatal: True
The proposed mirror addresses only the arbitrary-distribution resilience boundary in Theorem 4.1; it does not cover the other threshold, distribution-specific, or split-attack results.
The paper has no named computational-complexity result of the kind your anchor rule asks for: none of its theorems classifies a problem as polynomial-time, NP-hard, FPT, or similar. Its numbered results concern Bayesian equilibrium and granularity thresholds. The closest anchor for a positive case is Theorem 4.1, proved in the paper.
Theorem 4.1 says that for every \(n>2\) and every \(q<1\), there is a valuation distribution \(F\) and a false-name attack that benefits some bidder in the \(n\)-bidder, two-item VCG auction. The proof gives a concrete family: a bidder of type \((1,1)\) splits into bids \((1,1)\) and \((1,\tfrac12)\); the other bidders’ values are drawn from a distribution with masses \(1-2\varepsilon\) at \(0.6\), and \(\varepsilon\) each at \(0.5\) and \(1\), for a suitably small \(\varepsilon(q,n)>0\). Thus no \(q<1\) guarantees Bayesian resilience for every valuation distribution when \(n>2\).
A plausible continuous setting is a large online-advertising pool feeding many two-slot auctions. A bidder type is \((g,\theta)\): demand \(g\in\{1,2\}\) and per-item value \(\theta\). The pool’s mass \(\mu_{g,\theta}\) records the fraction of advertisers of each type; auctions sample \(n\) participants from it. In the theorem’s witness, only three value levels and two demands are needed, so the pool has at most six types even as the number of advertisers grows. The market-wide pool is the continuous object; each local auction still has two indivisible items and the paper’s false-name bidding rules.
The corresponding problem is Population-Pool VCG Resilience. An instance consists of \(n\), a finite type set \(\{1,2\}\times V\), and rational masses \(\mu\) summing to one. One bidder knows their type; the other \(n-1\) bidders are sampled from \(\mu\). The question is whether truthful bidding is a Bayesian Nash equilibrium against all false-name bid vectors. A “no” solution includes a bidder type and attack whose expected utility is strictly higher than truthful bidding. Formally, compare the truthful utility with the expected utility under \(\mu^{n-1}\); the objective is the bidder’s expected quasi-linear utility. Theorem 4.1 supplies negative instances for every \(n>2\) and \(q<1\), with the type masses induced by its \(q,F\).
For this finite-support version, I would expect resilience certification to be Class A, though that is a conjecture about the mirror, not a result of the paper. The two-item structure reduces the relevant opponent information to the highest two one-item bids and the highest two-item bid, while Lemma 1 reduces useful deviations to a pair of one-item bids or truth-telling. That suggests exact computation from the support and order statistics rather than enumerating all opponent profiles. Further questions include the complexity in \(n\), support size \(\tau\), and bit length \(L\); the exact resilience threshold for a given mass distribution; and when the split attack is worst.
This mirror covers the paper’s Bayesian resilience question for two-item VCG, not its broader discussion of alternative mechanisms or welfare guarantees. Its weakest point is that the paper already uses \(q\) and \(F\) as a distribution over bidders’ private types. The positive case depends on those masses being an actual large advertiser pool randomly matched into auctions, rather than merely uncertainty about a fixed finite group. And although Theorem 4.1 gives a strong equilibrium boundary to study, it is not itself a complexity classification.
The strongest objection is that Theorem 4.1 is not a computational result: it establishes a Bayesian-equilibrium boundary, not the complexity of deciding or computing anything. The proposed “Population-Pool VCG Resilience” problem adds a complexity question to the paper; it does not mirror a computational result the paper proves.
The advertiser-pool scenario is plausible, and it is the best version of this anchor. But keeping each auction at \(n\) bidders leaves the strategic society finite. The pool’s masses only supply the prior from which those bidders are sampled. Increasing the pool while holding \(\mu\) fixed changes nothing about the auction or its equilibrium calculation; the expected-utility comparison is the paper’s existing Bayesian analysis, with the same \(q\) and \(F\) expressed as population shares. A complexity study of that calculation might be worthwhile, but it would be a new algorithmic project, not a benefit produced by continuizing this game.
If instead the auction itself has a continuum of participants, two indivisible items leave each participant with zero probability of allocation in the population limit. For a randomly selected participant, the chance of being among the two winners vanishes as the number of bidders grows; with atomless values, the winning surplus also tends to collapse toward the top of the support. The individual false-name incentive that Theorem 4.1 studies then loses its role. Scaling the item supply with the population could preserve a nontrivial market, but it changes the two-item auction that anchors the theorem.
So the negative case is strongest as a scope objection, and weak as a claim that no useful related model exists. The paper’s distributional setup makes a pool-based interpretation natural; the difficulty is that it either leaves the paper’s finite Bayesian game intact or changes the auction enough that Theorem 4.1 no longer anchors the continuous question.
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.