| paper | d4-conf-atal-2021-0050 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | medium |
| authors would recognise it | yes |
other→P discrete result → its continuous mirror
Answer: The exact policy-selection problem is solvable in deterministic polynomial time in \(m,\tau,L\), including the exact rational score and the prescribed tie-break. · discrete side: other · adjudication: proved
full proof & adversarial review →
Lemma 2
statement extracted from the paper’s text layer
Given rational masses \(\mu_\theta\) of owner types and \(\nu_\phi\) of viewer types, complete type data, and a rational relationship kernel \(w_{\theta\phi}\), choose a policy \(a\) maximizing \(\sum_\theta \mu_\theta u_{\theta,a}v_{\theta,a}\). The finite candidate set contains policies preferred by positive-mass owner types, a specified compromise policy, and the distinguished uploader’s preferred policy if the uploader is represented separately; each \(u_{\theta,a}\) is computed from the type-level allowed, extra, excluded, and excluded-extra viewer masses, and \(v_{\theta,a}\) is the paper’s value-promotion score. Return a maximizing policy and its score.
A finite-support population of co-owners and viewers, grouped by complete preference, value, appreciation, and relationship types. Policies are the decision variables, and the objective is the mass-weighted version of Equation (3), with type-level utilities computed from the relationship kernel. With a finite candidate set, each policy can be evaluated by aggregating over the types.
The mirror covers the policy-selection guarantee in Lemmas 1 and 2. It leaves Lemmas 3 and 4, explanation generation, simulations, and the user study outside the mirror.
There is no named complexity result in this paper to anchor a continuization mirror to. The closest results are Lemma 1 in §4.3, proved here, which says ELVIRA’s recommended audience is always optimal—“the one which is the most coherent with everyone’s utility and value preferences”—and Lemma 2, also proved here, which says that, assuming cooperation, ELVIRA finds and recommends an optimal audience if one exists. Those are correctness claims under the paper’s assumptions, not claims that a problem is in P, NP-hard, or otherwise classified. The simulations and user study are empirical results, not numbered computational results. So neither lemma qualifies as an anchor under the requested standard.
The strongest plausible mirror I can offer is therefore unanchored: a continuous version of the policy-selection problem underlying Lemma 1. Consider a large group co-owning an item. An owner type \(\theta\) specifies a preferred sharing policy \(p_\theta\), an appreciation value, a value ordering, and the type-level relationship profile used to assess who would see the item. Let \(\mu_\theta\) be that type’s fraction of the owners. Let \(\Phi\) be a finite set of viewer types with masses \(\nu_\phi\); a rational relationship kernel \(w_{\theta\phi}\) records how type \(\theta\) weights viewers of type \(\phi\). These are complete types: people grouped together have the same preferences, values, appreciation, and relationship profile for this decision.
The candidate policy set \(\mathcal A\) contains the policies preferred by positive-mass owner types, plus a specified compromise policy, as in the paper. For each \(a\in\mathcal A\), type-level audience sets determine the allowed, extra, excluded, and excluded-extra viewer masses. Extending Definition 5, type \(\theta\)’s utility is
\[ u_{\theta,a} = W_\theta(A_{\theta,a}) -\alpha W_\theta(C_{\theta,a}) +\operatorname{app}_\theta\bigl( W_\theta(B_{\theta,a})+\beta W_\theta(D_{\theta,a}) \bigr), \]
where \(W_\theta(S)=\sum_{\phi\in S}\nu_\phi w_{\theta\phi}\). Let \(v_{\theta,a}\) be the paper’s value-promotion score for that type and policy. The problem is: given these rational masses, type data, relationship kernel, and candidate policies, find an \(a^*\in\mathcal A\) maximizing
\[ S(a)=\sum_\theta \mu_\theta u_{\theta,a}v_{\theta,a}. \]
The solution is a maximizing policy and its score. This mirrors the paper’s utility–value aggregation directly: the objective is its Equation (3) with the finite sum over owners replaced by a mass-weighted sum.
A plausible regime is a large group photo at a graduation or company event: hundreds or thousands of co-owners, but a relatively small number of combinations of sharing preference, value ordering, appreciation, and relationship profile. The social-network structure must also have repeated relationship profiles, such as block-like groups; without that, types could proliferate until the compression disappears. This is a property of that scenario, not of multiuser privacy generally. The authors could plausibly recognize the model as the high-multiplicity form of their policy-selection objective, though the single-uploader setup makes the limit delicate: one uploader’s mass vanishes as the group grows, even though the paper gives every user a score in the aggregate.
For the stated finite candidate set, this problem should be tractable by evaluating every policy and aggregating type scores, in time polynomial in the number of owner types, viewer types, and input bit length. That is a straightforward consequence of the proposed representation, not a complexity result established by the paper. The mirror covers policy selection only; it does not continuize the paper’s explanation generation, formal role and audience properties, simulations, or user study.
The weakest point is also decisive for the requested case: this mirror has no qualifying named anchor in the paper. It turns the paper’s optimization into a population-mass problem, but it does not continue a discrete hardness result whose hardness might dissolve, nor extend a named algorithmic complexity result. Further questions include whether the relationship profiles stay few in realistic networks, how to relate finite-group recommendations to the mass optimum, and how to handle the vanishing influence of a unique uploader. Without a named computational anchor, these remain promising questions around the paper, rather than a result-based case for its continuous mirror.
The strongest negative point is the one the proponent concedes: the paper has no named computational result to continue. Lemmas 1 and 2 are correctness claims under the agents’ cooperation assumptions; Lemmas 3 and 4 are invariance properties. The simulations and user study are empirical. None classifies a computational problem or states an algorithmic complexity bound. Replacing the sum in Equation (3) with an integral, \(\int u_{\theta,a}v_{\theta,a}\,d\mu(\theta)\), gives a natural new objective, but it does not mirror a computational result in this paper.
The best version of that proposal would keep one distinguished uploader and model many co-owners as a few complete types, with repeated relationship profiles and a mass-weighted score. That could make sense for a large, structured group sharing an item. But the paper’s setting relies on a social graph: each user’s audience and utility depend on their relationships to others. In a general graph, those relationship profiles can be highly individual, so grouping owners into a few types requires substantial regularity. If such block structure is imposed, the resulting model is coherent, but it is still a new population-weighted extension. Making it a more substantial computational problem—for example, by introducing an implicit policy space or additional constraints—would require adding a problem the paper does not formulate.
So the proposed mirror may be worth exploring as a standalone model, but it is not anchored to a result from this paper, and the paper supplies no computational claim for a continuous version to confirm, sharpen, or overturn. That makes the negative case strong against this candidate as a ChoCo mirror. It does not prove that no future model of privacy decisions could be worthwhile; the claim that every possible scenario fails would go beyond what the paper supports.
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.