| paper | d4-conf-aaai-2026-0911 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | no |
The numbered propositions and corollaries state axiomatic properties and representation results, not computational guarantees or hardness results. The proposed audience robustness problem adds a new aggregate objective, so it does not supply a continuous analogue of a computational result in this paper.
fails bit a — no named computational result to mirror
The paper characterizes targeting methods and analyzes their axiomatic properties; the audience-level robustness extension is not a mirror of a named computational result.
The strongest honest case is that this paper has no named computational result to anchor a continuous mirror on. Propositions 6–12 establish which axioms the proposed targeting functions satisfy; Proposition 4 represents a targeting method with a preorder; and Corollary 2 gives an aggregation representation. None asserts a complexity result such as polynomial-time solvability, hardness, or fixed-parameter tractability. They are results about the axiomatic structure of targeting, not about computing it.
A plausible adjacent scenario would be a large customer population with target types given by evaluation vectors, where each type occupies mass \(\mu_a\) and receives recommendations according to its targeting method. But that would extend the paper’s single-target selection problem into a many-target recommendation problem. It would not mirror a named computational result in this paper, and the paper supplies no computational anchor for it.
So I can’t state a qualifying continuous problem per anchor without misrepresenting the paper. Its results are axiomatic, and continuizing the population would not turn them into the computational questions this programme studies.
The proponent is right about the strongest objection: this paper states no complexity theorem. Its numbered results characterize targeting methods and their axiomatic properties. That makes it a less direct ChoCo mirror than a paper about bribery or control, but it does not establish that no worthwhile computational question arises from the results.
Proposition 6 gives a concrete choice rule, and Corollary 2 explains the broader family it belongs to. In the paper’s expert-finder setting, imagine a large audience whose members each choose a target profile and receive recommendations under one of these rules. Let each type \(t\) fully specify a target’s evaluation vector and any other information the recommendation process uses. For a fixed candidate set \(O\), let \(e_t(p)\) denote the share of type \(t\) that endorses \(p\), splitting ties by a specified convention. A society is a distribution \(\mu\) over these types, and the endorsement mass of \(p\) is \(E_\mu(p)=\sum_t\mu_t e_t(p)\).
That gives a specific continuous robustness question: what is the minimum cost of transferring mass between target types so that a chosen alternative \(p^\star\) becomes a most-endorsed recommendation, meaning \(E_{\mu'}(p^\star)\ge E_{\mu'}(q)\) for every \(q\in O\)? The targeting methods in the paper determine the type-level endorsements; the new question studies how the recommendation outcome responds to changes in their population mix. With a finite type space, it has the familiar mass-transfer form. With compactly represented or combinatorial alternatives, computing the endorsements may itself require studying the targeting rule’s optimization problem.
This does add an aggregate recommendation objective that the paper does not study, so it is not a clean mirror of one of its theorems. But it uses the paper’s actual targeting rules, in an audience setting supported by its expert-finder example, and asks a computational question about a distribution over complete types. The lack of a complexity result is a real weakness in the case against a mirror; it is not enough to rule out this more ambitious one.
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.