| paper | d4-conf-atal-2016-0264 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | unclear |
The paper contains algorithm descriptions and simulation comparisons, but no numbered theorem, lemma, corollary, or proposition with an algorithmic guarantee or hardness claim. The proposed optimization over a distribution of algorithm types is a plausible follow-up to the experiments, yet it is not a computational result of this paper, so bit (a) fails.
fails bit a — no named computational result to mirror
None of the paper's simulation findings in Figures 1–11 or Table 3 is covered by an anchored mirror; the proposed portfolio optimization is an unproved follow-up.
There are no qualifying anchors in this paper: it contains no numbered theorem, lemma, corollary or proposition asserting a computational result such as membership in \(P\) or NP-hardness. Figures 7–11 and Table 3 report simulation results, not named computational results. So, under your anchor rule, there are zero anchor-to-mirror questions.
The strongest positive case the experiments suggest is a continuous version of the paper’s algorithm-mix study. In a city-scale route-choice system, many commuters could share one of a finite number of app configurations and current states, while responding to the same public traffic history. A type would include its algorithm family, parameters and internal state; its mass would be the fraction of commuters with that profile. Millions of commuters could then be represented by a modest number \(\tau\) of types. Their choices remain decentralized, though selecting the mix of configurations is a design decision.
One precise candidate is Continuous Minority-Game Portfolio Design. Given a finite set \(T\) of complete types, a time horizon \(H\), the game’s update rules, rational constraints on the allowed family proportions, and a required mean reward \(\rho\), choose an initial mass vector \(\mu\) to minimize the variance of cumulative per-agent rewards, subject to \(\mathbb{E}[R_H]\ge \rho\). Each round, types choose actions according to their algorithm and state; action masses determine the minority, which updates the shared history and the types’ states. A solution is an optimal \(\mu\), its induced trajectory, and its reward variance. The model must specify how a \(50\)-\(50\) split is resolved.
This is motivated by the paper’s reported results, but is not anchored in a named result. Figure 11 reports that greater measured diversity is associated with more even rewards; Table 3 reports, for example, that the Evolutionary–Roth-Erev mix at \(d=5.13\) has lowest, median and highest payoffs of \(774\), \(957\) and \(1046\), compared with \(368\), \(483\) and \(1001\) for pure Best Play. Those observations make portfolio design a plausible computational question for the authors to recognize, but they do not establish its complexity.
Provisionally, I would expect this optimization question to be continuum-specifically hard when \(H\) is part of the input: changing \(\mu\) can change which action is in the minority, and that change feeds back into later behavior. That is a hypothesis, not a result from the paper. The discrete-to-continuous bridge would also need attention near ties, where finite odd populations and exact continuum masses can produce different outcomes.
The weakest point is that portfolio optimization adds a design decision to a paper whose stated contribution is experimental evidence for decentralized fairness. The article varies mixtures, so that extension is plausible, but it does not analyze an optimization problem or prove a computational claim. The honest positive case is therefore a candidate mirror suggested by the experiments—not an anchored case under the requested standard.
No qualifying anchor was raised: the paper has no named theorem, lemma, or corollary establishing a computational result. Its figures and tables report simulation outcomes. That is the strongest case against a *result-by-result* mirror: there is no computational result here for one to mirror.
The proposed portfolio problem is a plausible extension, but it changes the paper’s question. The experiments vary the mix of algorithms to study whether diversity produces fairer rewards without centralized oversight. Choosing \(\mu\) to minimize reward variance makes the mix a planner’s decision. A closer version would fix \(\mu\) and ask how to compute the resulting reward distribution over horizon \(H\). That is more faithful, but it turns an experimental claim into a new computational question rather than translating a computational result in the paper.
There is a further modeling strain in the evolutionary strategy. Its update selects the ten best and ten worst agents. In a continuum, selecting a fixed ten agents has vanishing mass; selecting a fixed fraction instead changes the update rule. And a type detailed enough to track an agent’s strategy pool, learned state, and reward history may split into many types as agents’ histories diverge. One can imagine commuters sharing an app configuration and public traffic history, but whether enough also share the relevant internal state is a substantive assumption, not something the paper establishes.
That case against is limited. The paper explicitly varies algorithm proportions, so a fixed-mixture continuous model of commuters and its long-run reward distribution is recognizable and potentially useful. The objections weaken the connection and point to modeling choices; they do not rule out a sensible high-multiplicity regime. I can defend the narrower claim that this paper offers no anchored computational mirror. I cannot honestly defend the universal claim that no worthwhile continuous mirror exists in any scenario.
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.