| paper | d4-journals-ai-2024-0048 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | medium |
| authors would recognise it | yes |
P→P discrete result → its continuous mirror
Answer: Polynomial-time solvable in \(m,\tau,L\). · discrete side: polynomial · adjudication: proved
full proof & adversarial review →
Theorem 10
statement extracted from the paper’s text layer
Given a finite outcome set \(O\) and a rational distribution \(\mu\) of support size \(\tau\) over matched pairs of strict rankings, compute a succinct equilibrium-strategy prescription for each supported pair and the exact mass \(\nu_o=\sum_{\theta:E(\theta)=o}\mu_\theta\) ending at each \(o\in O\), where each pair describes an independent full-information VAOV case.
The population consists of independent VAOV cases with matched-ranking types \(\theta=(\succ_1,\succ_2)\) and masses \(\mu_\theta\). Apply the paper's equilibrium strategy to each supported type and aggregate the resulting mass into \(\nu\); there is no cross-case decision or scalar objective.
Each case is solved independently, so the census adds no shared strategic constraints; its population-level meaning depends on treating negotiation cases as the units of society.
fatal: False
Covers Theorem 10's full-information equilibrium-strategy computation and the resulting outcome masses; leaves the Rational Compromise comparison, monotonicity, and the no-information and asymmetric-information results aside.
The strongest honest case is a narrow one: a population of many standardized bilateral negotiations, not a continuum inside a single negotiation. Imagine a mediation service handling millions of cases over the same finite menu of remedies. Each case has two parties with strict rankings of those remedies, both parties know both rankings, and the first mover is fixed. Repeated ranking pairs make this a plausible high-multiplicity regime: the number of cases \(N\) can be far larger than the number \(\tau\) of distinct supported pairs of rankings.
The individual types remain exactly the paper’s types: strict preference orders over outcomes. Since each game is bilateral, the population is described by a joint distribution \(\mu\) over matched pairs \(\theta=(\succ_1,\succ_2)\), with rational masses summing to \(1\). Here \(\mu_\theta\) is the fraction of cases with that pair of preference orders. A case remains an ordinary VAOV negotiation with one discrete outcome; the mass is over repeated cases, not over fractional outcomes or over the two players in a single game.
My lead anchor is Theorem 10, proved in this paper. It says Strategies 1 and 2 specify a subgame-perfect equilibrium for the full-information VAOV game. In a round with remaining outcomes \(O_t\), let \(L_t^i\) be the offerer’s lowest-ranked \(\lfloor |O_t|/2\rfloor\) outcomes when \(|O_t|\) is odd, or the lowest-ranked \(|O_t|/2-1\) when it is even; let \(L_t^j\) be the responder’s lowest-ranked \(\lfloor |O_t|/2\rfloor\). The offerer offers an outcome in \(L_t^i\cap L_t^j\) if that intersection is nonempty, and otherwise offers the responder’s least-preferred remaining outcome. The responder accepts an offer exactly when it is at least as good for them as the outcome they would get by rejecting and continuing with those strategies. The paper’s proof says a subgame-perfect strategy from any given state can be computed in time linear in \(m=|O|\); the equilibrium-strategy result is proved here.
Continuous VAOV Equilibrium-Outcome Census is the corresponding population problem. Its input is a finite set \(O\), a rational distribution \(\mu\) with support size \(\tau\) over matched pairs of strict rankings, and the full-information VAOV protocol. It asks for a succinct specification of the equilibrium strategies for every supported pair and the exact mass \(\nu_o\) of cases ending at each outcome \(o\), where
\[
\nu_o=\sum_{\theta:\,E(\theta)=o}\mu_\theta
\]
and \(E(\theta)\) is the outcome of the specified equilibrium for pair \(\theta\). A solution is the strategy prescription together with the vector \((\nu_o)_{o\in O}\); there is no scalar objective to optimize, since the defining requirement is subgame perfection. Applying the paper’s per-state algorithm to each supported pair gives polynomial time in \(m,\tau,L\), where \(L\) is the input mass bit length, followed by exact rational aggregation.
I expect this mirror to be tractable. It preserves the paper’s underlying strategic question for every case and uses the support representation to process repeated types without expanding all \(N\) cases. It does not claim a new tractability gain inside an individual negotiation, or claim that the paper proves a hardness boundary. Rather, it makes the aggregate equilibrium distribution a computational question at population scale. The paper’s stated interest in strategies usable in multi-agent systems gives some reason the authors might recognize the deployment question, though the continuous census itself is an extension.
This covers Theorem 10 and its full-information equilibrium computation. The paper’s other results, including the no-information analysis and the connection to the Rational Compromise rule, are outside this mirror. Further questions include how sensitive \(\nu\) is to changes in \(\mu\), and how much mass must shift between preference-pair types to make a chosen outcome reach a target share.
The weak point is the unit being continuized. The paper studies one dyadic negotiation at a time; this mirror studies a large population of independent dyads. So it is a credible continuous population model for deploying and analyzing the protocol, but not a continuum-agent version of the strategic interaction within one VAOV game. If the authors or referee require the population to participate in one shared game, this case may not qualify.
Theorem 10 is the right anchor, and I would not dismiss it by claiming the paper lacks a computational result. It proves the equilibrium-strategy result, and the paper gives a linear-time procedure for computing a strategy from a game state.
The difficulty is that the proposed census aggregates *games*, while Theorem 10 concerns the strategic interaction inside one game. For each supported pair of rankings \(\theta\), the equilibrium outcome is a deterministic value \(E(\theta)\). The census computes the push-forward of \(\mu\) through \(E\): apply the paper’s procedure to each supported pair, then sum the masses. With a succinct strategy generator, that takes \(O(\tau m)\) local computation plus rational aggregation. Repeated cases save repeated calls, but no mass interacts, no shared equilibrium constraints arise, and the continuous model introduces no new optimization question. That may be a useful deployment statistic; it is a thin extension of the theorem’s existing algorithm.
A stronger version would describe a society of individual negotiators by their ranking types and let a platform choose whom to pair, perhaps to maximize the mass of negotiations ending at a target outcome. A mass-matching LP could make that a genuine population problem. But its substantive decision is the platform’s pairing policy and its objective is an added social goal. Theorem 10 supplies outcomes for candidate pairs; the paper studies neither pairing nor population-wide welfare. That could be a worthwhile follow-up, but its computational core comes from the added assignment problem, not a continuous mirror of the paper’s equilibrium result.
The negative case is weakest at the boundary the proponent chose. A high-volume mediation service with repeated standardized bilateral cases is plausible, and if ChoCo accepts a population of matched negotiation cases as its society, the census is well-posed. I cannot show that such an aggregate would never be useful. My case is that it does not earn much as a continuization of this paper: it batches an already efficient finite-game computation, while the more substantial population models require adding a problem the paper does not study.
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.