| paper | d4-conf-atal-2013-0010 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | no |
The paper contains no named computational result to anchor a continuous mirror, so the required result test fails. The proposed mass allocation also adds a collective objective to a framework where each agent selects a plan for its own goal.
fails bit a — no named computational result to mirror
No named computational result is covered; the declarative preference framework, aggregation functions, and implementation remain outside any surviving mirror.
There are no qualifying anchors in this paper. It contains no numbered Theorem, Lemma, Corollary, or Proposition asserting a computational complexity result. Equations (1)–(3) define aggregation functions; the implementation and claims about expressivity are not named complexity results. So I cannot honestly provide an anchor-backed case of the kind requested.
The strongest plausible mirror is Population BDI Plan Selection\(_\infty\). Take a finite set of agent-context types \(T\), with rational masses \(\mu_t\) summing to \(1\). For each type \(t\), the input gives its current beliefs and the applicable grounded plan instances \(\mathcal I_t\). Each instance has a plan template, bindings, and the paper’s preference description \(\langle d,V,\lambda\rangle\). Its value \(q_{t,i}\) is computed from the preference formulas true under that type’s beliefs and the instance’s bindings, using the specified aggregation function and the default \(d\) when the paper’s rule calls for it.
The question is to find an allocation \(x_{t,i}\) of population mass to applicable plan instances that maximizes total assigned preference:
\[ \max \sum_{t\in T}\sum_{i\in\mathcal I_t} q_{t,i}x_{t,i} \quad\text{subject to}\quad x_{t,i}\ge 0,\qquad \sum_{i\in\mathcal I_t}x_{t,i}=\mu_t. \]
A solution is the allocation \(x\), or equivalently the resulting mass assigned to each plan instance. Each individual still executes one plan; fractional \(x_{t,i}\) records what share of a type’s mass selects that plan. When top-valued instances tie, the continuum can split mass among them.
A plausible regime is the large agent-based simulation that motivates the paper: for example, many simulated travelers using the same BDI plan library, grouped into a modest number of types by origin and destination zone, current beliefs, available routes, and other details that affect applicability or plan value. Mass is the fraction of simulated agents in each type. This is faithful to the paper’s central operation—dynamically valuing applicable plan instances using situation and bindings—and does not turn fractional outcomes into the goal. It covers the snapshot plan-selection step, not the agents’ whole execution over time.
For the explicitly grounded version, where instance values and the necessary formula-reasoning results are available, I expect Class A: the optimization is separable by type, and each type’s mass can be assigned to a highest-valued instance. The paper does not establish that classification. An end-to-end version would also need to account for the cost of evaluating its formulas and the entailment checks used by aggregation.
The further computational questions are whether those checks are tractable for the paper’s full preference language, and how the problem changes when types or grounded plan instances are represented compactly. The bridge to a discrete population is \(\mu_t=n_t/n\), but this mirror mainly aggregates many local choices; it does not reveal a new population-level interaction.
That is the weak point: the authors may reasonably see this as bookkeeping around repeated individual plan selection, rather than a new computational problem. More decisively, there is no named result here for the mirror to carry. This is a concrete possible continuous analogue of the paper’s main operation, but it does not satisfy the required result-anchor test.
The proponent is right that there are no qualifying anchors: this paper states no numbered result about the complexity of a computational problem. Its equations define preference aggregation functions, and it reports an implementation, but neither supplies a complexity claim for population plan selection. That leaves no result to continuize. The proposed question has to stand as a new research direction, rather than a mirror of one of the paper’s results.
The proposed mass-allocation LP also adds a social objective that the paper does not give plan selection. In the paper, an agent compares applicable plan instances for its own goal, using its beliefs and the instance’s bindings. For a type \(t\), if those determine values \(q_{t,i}\), a faithful population summary would apply the agent’s selection rule and report the resulting shares. With a fixed tie-breaker \(\sigma\), for example, those shares are \(\nu_i=\sum_{t:\sigma(t)=i}\mu_t\). Allowing arbitrary fractional splits among tied instances instead introduces a tie policy absent from the paper. Maximizing \(\sum_{t,i}q_{t,i}x_{t,i}\) goes further: it treats plan values as comparable across agents and chooses a collective allocation. The paper’s local plan preferences do not specify that social comparison or allocation problem.
A large agent-based simulation is a plausible setting for repeated BDI states, and a distribution of selected plans could be useful there. But as a mirror of this paper, the computation is just a population summary of local choices. If evaluating a preference formula or entailment is difficult, that difficulty already occurs for one agent; the mass representation does not resolve it. A richer question involving shared capacity, agents influencing one another, or a controller reallocating plans might have substantial computational content, but it would add interactions or a collective decision that the paper’s local framework does not study.
So the strongest negative case is about fit: there is no computational result to anchor, and the proposed optimization changes individual plan selection into interpersonal welfare maximization. I would not claim that no continuous population model of BDI simulations could ever be worthwhile. The paper’s simulation motivation makes that possibility credible. The case is weaker against such a broader follow-on project than against treating this paper as a continuous-mirror candidate.
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.