| paper | d4-conf-atal-2015-0383 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | yes |
The paper groups CP-nets by relative frequency, and its PR aggregation uses those frequencies exactly as the proposed type masses. Clearing denominators realizes any rational mass vector as a finite profile, so the proposed PRI question restates the paper’s existing model and result.
fails bit b-already — the paper is already native to the continuum
The paper already encodes each finite profile as distinct CP-net types with relative frequencies, and its PR aggregation uses those masses directly; clearing denominators realizes any rational mass vector.
fatal: True
The proposed mirror covers PRI winner determination, anchored by Theorem 1; it leaves PRO, the LS variants, and the dominance results aside.
The best mirror is Continuous PRI winner determination. The population is a large membership choosing one joint configuration of \(n\) binary features. A type is a complete CP-net: its dependency edges and every conditional preference statement are fixed. Members of the same type are indistinguishable for this decision. The society is a distribution \(\mu\) over the \(\tau\) types present, with rational masses summing to \(1\); plausibly, \(N\) is very large while \(\tau\) is modest because members reuse a small number of preference templates. Assume the CP-nets share a topological order and their possible parent sets are bounded, as in the paper’s \(O\)-bounded regime.
For each feature \(X_i\) and assignment \(u\) to its possible parents, let \(q_i(u)\) be the mass of types whose CP-net prefers \(x_i\) to \(\bar{x}_i\) at that row. Form the PR aggregate using these masses. The question is: which complete assignment does PRI select? The solution is the lexicographically tie-broken outcome obtained by processing features in topological order and, at each feature, choosing the value preferred by at least half the mass at the applicable row. This is a social choice rule, not a utilitarian objective: the output is the rule’s winning outcome. The masses represent actual population shares, not uncertainty about an individual’s preferences.
The anchor is Theorem 1, proved in this paper: for any profile of CP-nets, PRI—PR aggregation followed by choosing the optimal outcome of the most probable induced CP-net—returns the same result as sequential voting with majority. In particular, it identifies the paper’s collective outcome computation with a sequence of local majority decisions. The paper states polynomial-time computability for PRI under bounded width; the mass version is also polynomial in \(n,\tau,L\) for bounded parent sets, where \(L\) is the mass encoding length. One can sum the relevant type masses for each row and then make the sequential choices. This is a Class A, tractable mirror.
The authors should recognize the connection: the mirror retains their CP-net types, PR aggregation, PRI rule, and sequential-majority result. It covers that winner-computation result, not the paper’s LS aggregation, its other notion of optimality, or its dominance analysis. It also opens natural follow-up questions: how the rule behaves at exact half-mass ties; how its complexity changes when the parent-set bound is part of the input; and whether similar population formulations of PRO or the paper’s dominance queries remain tractable. The paper reports #P-hardness of general PCP-net dominance as a result of Bigot et al. [2], but that is cited prior work, not a numbered theorem here, so it is not an anchor for this case.
The weakest point is substantial: the paper already defines frequencies for repeated CP-nets and its PR rule operates on those proportions. Replacing finite counts by rational masses may therefore look like a high-multiplicity formalization of a rule whose mathematics is already in the paper. The positive case is that the paper studies finite profiles, while the mirror makes support size and mass bit length the explicit input regime, with \(N\) potentially huge. That is a legitimate computational question, but this is a close mirror, not a claim that the paper’s underlying aggregation idea was previously discrete.
For the proposed PRI mirror, the population is already present in the paper’s input. It represents a profile as CP-net types with relative frequencies, \(P=((C_1,\mathrm{freq}_1),\ldots,(C_\tau,\mathrm{freq}_\tau))\). Those frequencies are the type masses: the paper’s PR rule sets each conditional preference probability by summing the frequencies of the CP-nets that make that choice. That is precisely the proposed \(q_i(u)=\sum_{t:\,t\text{ prefers }x_i\text{ at }u}\mu_t\).
Theorem 1 then already gives the proposed outcome: process features in topological order and choose the locally more popular value, using the stated tie-breaking convention. Its claim is for any profile, and its proof uses the frequencies directly. Every finite rational mass vector can also be realized as a finite profile by clearing denominators. So the continuous question here does not reveal a different population model or winner rule; it restates the paper’s result in mass notation.
The strongest version of the mirror would make the high-multiplicity encoding explicit: list the \(\tau\) distinct CP-nets and binary rational masses, then bound the running time in \(n,\tau,L\). Under the paper’s bounded-parent assumption, this amounts to summing rational masses for each applicable row and comparing them with \(1/2\). That is a useful bit-complexity clarification, but the paper already groups CP-nets by frequency and uses those frequencies in the aggregation. The remaining work looks like encoding and arithmetic bookkeeping, not a new algorithmic question. I am treating this as a collision because of what this paper itself models—not because high-multiplicity work exists elsewhere.
The negative case is strong for this anchor and its close variants. The supplied text does not support the literal universal claim that no worthwhile continuous mirror could arise from any newly added problem, such as a distinct intervention task. But such a task would introduce a new computational operation rather than uncover a missing population version of this paper’s PRI result.
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.