| paper | d4-journals-jair-2018-0035 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | medium |
| authors would recognise it | yes |
other→P discrete result → its continuous mirror
Answer: Polynomial-time solvable. · discrete side: other · adjudication: proved
full proof & adversarial review →
Proposition 4.10
Given a finite market-state model, rational masses μ_t summing to \(1\), and each supported account type \(t\)’s asset blocks and proposed structure-preserving edit encoded by rational multiplicative factors, compute \(q_{t,\min}^{\mathrm{post}}=\min_v q_t^{\mathrm{post}}(v)\), where \(q_t(v)=\exp(a_t(v)/b)\). Decide for each type whether \(q_{t,\min}^{\mathrm{post}}\ge 1\), equivalently whether its post-edit minimum cash \(b\ln q_{t,\min}^{\mathrm{post}}\) is nonnegative, and report the eligible mass \(\sum_t \mu_t\mathbf{1}[q_{t,\min}^{\mathrm{post}}\ge 1]\). Treat this as a snapshot audit against the current market distribution: do not pool accounts or assume their proposed edits are committed together.
A distribution over complete account types, where each type specifies its current asset profile and proposed edit. Compute solvency independently from that type’s asset blocks, then weight the eligible types by their masses; the outcome variables remain discrete.
The audit batches independent account checks, with μ used only to total eligible mass, so it adds little beyond type-level deduplication; the proponent specifies the compression regime but does not establish that this population summary matters operationally.
fatal: False
Covers Proposition 4.10 and Algorithm 4.2’s minimum-cash computation for Task 3 edit eligibility; leaves probability queries, market-distribution updates, expected assets, resolutions, and variable additions aside.
The strongest positive case I can defend is a narrow mirror of the paper’s collateral check for traders’ assets. The society is the forecasting crowd; the market’s possible outcomes remain discrete. That distinction matters because the paper’s joint probability distribution over outcomes is not itself a continuous population.
Consider a large forecasting platform with many users or automated forecasters, including repeated deployments that share a strategy, starting balance, account history, and pending order. A type must include all of those details: in particular, two traders are the same type only if their current asset profiles and proposed edits are identical for this task. If there are \(N\) active orders but only \(\tau\) distinct account-and-order types, the society is the distribution \(\mu\) over those types, with \(\mu_t\) the fraction of orders of type \(t\). This is plausible for replicated forecasting services or managed accounts, though it is a narrower regime than a crowd of users with unique trading histories.
I would call the mirror Mass-Weighted Solvency Audit\(_\infty\). Its input is a market’s discrete outcome variables and junction-tree structure, plus the supported trader types and their masses. Each type contains a trader’s asset blocks, their scopes and tables, and a proposed structure-preserving edit. To make exact computation well-defined, encode the current LMSR assets in multiplicative form: for type \(t\), let \(q_t(v)=\exp(a_t(v)/b)\), with rational local factors and rational edit ratios. An edit from \(p\) to \(x\) multiplies the affected local factor by \(x/p\), as follows from the paper’s asset update rule. The audit asks whether each post-edit account remains solvent in every market state, and what fraction of the order population is eligible.
Writing \(c_t^{\mathrm{post}}=\min_v a_t^{\mathrm{post}}(v)\), the decision for each type is \(y_t=1\) exactly when \(c_t^{\mathrm{post}}\ge 0\). The requested population output is the accepted mass \(\sum_t\mu_t y_t\), together with the per-type minimum cash values (and, if desired, states attaining them). This is a snapshot preflight of individual orders: it does not pool traders’ assets or claim that different orders can all be committed simultaneously without updating market prices between them.
My anchor is Proposition 4.10, proved in this paper. It says that if a user’s assets \(a_v^u\) are additively decomposable over blocks \(\mathcal B=\{B_1,\ldots,B_k\}\), and \(V_B\) consists of the variables appearing in more than one block, then the user’s minimum assets are exactly
\[ c^u=\min_v a_v^u =\min_{v_B}\sum_{i=1}^{k} \min_{\{v_{B_i}:V_B=v_B\}}\delta_{B_i}(v_{B_i}). \]
The paper derives this formula and computes cash from it with Algorithm 4.2. It is a named computational result used by the paper, but it does not claim an overall polynomial-time bound. The global-separator formula can be expensive when many variables overlap; the paper’s dynamic asset cluster method instead uses min-propagation on a junction tree.
For the mirror, I expect Class A when the asset-block interaction width is bounded: run the cash computation once per supported type, then sum the eligible masses. With bounded variable domains and width, the work scales with \(\tau\) and the represented block tables, rather than with the number \(N\) of individual orders. I would expect hardness to survive for unrestricted width, since the hard part is minimizing over assignments to the market’s discrete state variables, not over the number of traders. That is a prediction about the mirror, not a hardness theorem proved here. It raises the natural follow-up: what is the exact complexity of the unrestricted-width audit, even for one trader type?
This is a problem the authors could plausibly recognize: checking nonnegative assets and deciding whether to allow an edit are explicit market-maker tasks in the paper. The mirror covers that collateral-check component, not the full market maker, its probability-query task, or its expected-asset calculations.
The weakest point is the regime assumption. A trader type includes the current account profile and pending edit, so distinct trade histories can make \(\tau\) nearly \(N\). The mirror is compelling for repeated or managed account profiles; it offers little population compression when every user is unique. There is also no numbered theorem in the paper that classifies this computation as polynomial-time or hard. Proposition 4.10 is the best defensible numbered anchor because the paper actually uses its minimum-assets characterization in an algorithm, but the complexity landscape would need to be established by new work.
Proposition 4.10 is the proponent’s best anchor: it gives an exact way to compute a trader’s minimum assets from asset blocks. But the proposed mirror puts the population outside that computation. For each type \(t\), compute \(c_t=\min_v a_t^{\mathrm{post}}(v)\), then report \(\sum_t \mu_t\mathbf{1}[c_t\ge 0]\). The market-state minimization happens separately for each type; the population mass appears only in the final weighted sum.
That makes the Mass-Weighted Solvency Audit a batch of independent account checks. If many traders have identical asset profiles and proposed edits, compute once and reuse the answer. If their profiles differ, the support grows and the work scales with the number of distinct profiles. High multiplicity is a plausible regime for replicated bots or managed accounts, but here it only removes duplicate computations. It does not create a population-level feasibility problem or a new algorithmic structure for continuous optimization.
A stronger mirror might let a continuum of traders submit orders together, with their mass changing the market distribution and each account’s collateral checked along the way. But then the orders interact through changing prices, and the result can depend on how they are processed. Pooling accounts would also change the paper’s rule: it requires each user to remain solvent, not merely that aggregate assets cover aggregate losses. Preserving individual solvency leaves a market-dynamics problem with account-specific state, rather than a continuous version of Proposition 4.10’s snapshot computation.
So I would reject this anchor as not worthwhile for ChoCo: it does not expose a computational question created by continuizing the society. The negative case is not airtight. A carefully specified market with many repeated account profiles could make batch solvency checks useful, and Proposition 4.10 is genuinely algorithmic support for that task. But the mirror as stated adds a population-weighted report to an existing per-account computation; its continuous dimension does no substantive work.
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.