| paper | d4-conf-ecai-2012-0002 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | yes |
The proposed distributional check is statable and closely tracks Theorem 1, but the theorem establishes incentive compatibility rather than a computational result. The proponent acknowledges that no named computational result anchors the proposed check, so the paper fails the computational-result gate.
fails bit a — no named computational result to mirror
The proposed check covers Theorem 1’s baseline-inflation incentive property; it leaves the selection, bidding, revenue redistribution, learning, and simulation results unmirrored.
There is no named computational complexity result here in the sense required for an anchor. The closest is Theorem 1, proved in this paper: at a trading slot \(k\), if an agent is selected to reduce with probability \(\rho\leq 0.5\) and the rate it is paid is at most its retail price \(p_i^c\), then it cannot increase its expected profit by burning extra electricity to inflate its baseline. That is an incentive compatibility theorem, not a result that a computational problem is in P, NP-hard, or belongs to another complexity class. The paper gives no qualifying named anchor, so I cannot present a fully compliant anchor-based case.
The most credible positive case is nevertheless a narrow continuous version of the paper’s baseline-inflation question. Imagine a large portfolio of standardized commercial sites—such as branches of one retailer—with many sites but a relatively small set of exact operating types. A type \(t\) records the site’s relevant baseline load, reduction capacity, minimum acceptable rate, retail price, and performance parameters. The society is a distribution \(\mu\) over these types, where \(\mu_t\) is the fraction of sites of type \(t\). This is plausible only where standardization keeps the number of distinct complete types \(\tau\) much smaller than the number of sites.
Call the proposed problem Continuous CDSM Baseline-Incentive Verification. Its input is a finite-support society \(\mu\), a set of trading slots \(K\), a horizon of \(N\) baseline observations, and a candidate policy giving each type \(t\) and slot \(k\) a selection probability \(\rho_{tk}\), an awarded rate \(r_{tk}\), a retail price \(p_t^c\), and a maximum feasible baseline inflation \(\Delta_{tk}\). The question is whether any positive-mass type can profit by burning \(\delta\in(0,\Delta_{tk}]\) extra energy on days it is not selected, to claim that much extra reduction on selected days. For a type and slot, its expected incremental profit is
\[ G_{tk}(\delta) = N\delta\bigl(\rho_{tk}r_{tk}-(1-\rho_{tk})p_t^c\bigr). \]
A solution is either a certificate that \(G_{tk}(\delta)\leq 0\) for every supported type, slot, and feasible \(\delta\), or a violating type, slot, and profitable \(\delta\). Under the paper’s conditions, \(\rho_{tk}\leq 0.5\) and \(r_{tk}\leq p_t^c\), the policy passes. More generally, the test reduces to checking whether \(\rho_{tk}r_{tk}\leq(1-\rho_{tk})p_t^c\) for every supported type and slot. This proposed verification problem is tractable by a scan over the \(\tau |K|\) type-slot pairs; it is a Class A candidate.
The mirror is recognizable to the authors because it retains their central mechanism and their specific strategic deviation. It covers only the baseline-inflation incentive result, not the paper’s full slot selection, bidding, revenue redistribution, or learning scheme. The weakest point is that \(\mu\) barely affects this verification problem: the incentive check is essentially per agent, so the continuum does little computational work. Theorem 1 also does not meet the required complexity-anchor test. Optimizing mass allocations and payments for an entire cooperative could make the population genuinely central, but that would be a new research question, not a result anchored by this paper.
The strongest case against a mirror is that this paper has no computational result to mirror. Theorem 1 is about incentive compatibility: under stated assumptions, an agent cannot profit in expectation by inflating its baseline. Algorithm 1 describes a selection rule, and the experiments report simulated revenues and learning behavior. The paper does not analyze the complexity of a computational problem. That is a substantial mismatch with ChoCo’s aim of charting a computational landscape.
The proposed Continuous CDSM Baseline-Incentive Verification makes the gap visible. Its test reduces to scanning the supported type-slot pairs and checking whether \(\rho_{tk}r_{tk}\leq(1-\rho_{tk})p_t^c\). The fraction of society of type \(t\) does not affect that check. Making the population a distribution changes how many agents each type represents, but adds no computational work or population-level decision to this incentive question. A more faithful version could include the paper’s historical baseline and performance updates, but with complete types those features still enter as type-specific data; they do not turn Theorem 1 into a computational problem about masses.
A stronger proposed mirror would optimize the CDSM’s selection probabilities and payments across a large portfolio of standardized facilities, subject to reliability, incentive, and market constraints. That is a sensible adjacent research question, and repeated sites could make a high-multiplicity regime plausible. But it introduces an optimization objective the paper does not study: the paper specifies a mechanism and evaluates it rather than solving for an optimal mechanism. Under a finite supported type model, the natural version appears to have type-level variables and aggregate linear constraints. That may be useful mechanism design, but it is not a complexity result anchored in this paper; a continuous version would be a new problem.
I would not claim that no plausible continuous population can be imagined. Standardized portfolios could support one, and the paper’s aggregate bid and revenue calculations would make a type-level model recognizable to its authors. The negative case is weaker on that universal claim than on the central point: a population continuum may support new demand-management analysis, but this paper offers no named computational result whose continuous mirror would advance the programme’s complexity agenda.
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.