| paper | d4-conf-aaai-2014-0067 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | no |
Theorem 1 is a computational result, but its question concerns one balanced tournament and one distinguished player's chance to win. The proposed LP instead asks whether a continuum of bracket outcomes can all have winners of a target type; the paper's type-level result supports studying types, but does not bridge that change in objective.
fails bit b-changes — the continuum limit changes the question
The flows describe an aggregate of brackets with positive mass of winners, while Theorem 1 asks whether a single draw lets one distinguished player win; the proponent acknowledges changing player to type but does not address this mismatch.
fatal: True
The proposed flow model is aimed only at deterministic TFP in Theorem 1; it leaves PTFP, counting draws, and the paper's algorithmic results untreated.
My lead mirror is a continuous type-level version of deterministic tournament fixing. The anchor is Theorem 1, proved in this paper: TFP is NP-complete. In TFP, the instance is a set of \(n=2^h\) players, a deterministic pairwise comparison matrix specifying the winner of every match, and a distinguished player \(i^\star\). The question is whether some balanced knockout draw lets \(i^\star\) win. The paper proves hardness by reduction from a restricted 3SAT problem.
A plausible regime is a very large knockout tournament whose entrants fall into a small number \(\tau\) of matchup classes. Members of a class have the same deterministic result against every other class. For example, a tournament might have many entrants but only a modest number of reproducible strength tiers. The number of entrants is then far larger than \(\tau\), and the organizer wants a champion from a specified tier. In a same-class match, the individual winner is immaterial: the surviving entrant still has that class. This is not a probabilistic model; the class matchups remain deterministic.
The paper itself gives good evidence that this is a question its authors would recognize. It defines \(\#\mathrm{TFP\text{-}types}\), where players are grouped by their results against other types and the objective is to count draws won by a given type; it also proves in Theorem 2 that this counting problem is polynomial-time solvable when the number of types is constant. I use that as support for the type-level regime, not as a second anchor: my mirror asks a yes/no feasibility question, rather than counting discrete draws.
Here is the continuous problem I would hand to a prover. Continuous Type Tournament Fixing has as input a number of rounds \(h\), a finite set of types \(T\), rational masses \(\mu_t\geq 0\) summing to \(1\), a deterministic winner map \(w(t,s)\) for each pair of types, and a target type \(t^\star\). A tournament has \(2^h\) entrants at the discrete scale, but type mass is divisible: there is no requirement that \(2^h\mu_t\) be an integer. A solution is a sequence of nonnegative match-mass flows \(x^r_{\{t,s\}}\), one for each unordered type pair and round \(r\), together with the surviving type masses \(q^r_t\).
Set \(q^0_t=\mu_t\). In each round, all live mass must be paired exactly once: \(2x^r_{\{t,t\}}+\sum_{s\ne t}x^r_{\{t,s\}}=q^{r-1}_t\). The winners determine the next masses: \(q^r_t=x^r_{\{t,t\}}+\sum_{\{a,b\}:a\ne b,\ w(a,b)=t}x^r_{\{a,b\}}\). The question is whether such flows exist with \(q^h_{t^\star}=2^{-h}\), meaning all surviving mass has the target type. Each match consumes two units of mass and produces one unit of winner mass, so the total mass halves each round. The flows are the decision variable; they describe which type masses meet at each stage. Since the population is atomless, it can be split to realize fractional pairings.
I expect this mirror to be tractable (Class A). Its feasibility constraints form a rational linear program with \(O(h\tau^2)\) match variables and polynomially many balance constraints, so it is solvable in time polynomial in \(h,\tau\), and the input bit length. That gives a concrete way the indivisible-player hardness in Theorem 1 could disappear when the question is about a divisible population of interchangeable types. The mirror covers Theorem 1’s deterministic “can the desired winner be arranged?” question at the type level. It does not cover PTFP’s probabilistic win-probability optimization, \(\#\mathrm{TFP}\)’s counting question, or the paper’s exact algorithms.
The weakest point is that Theorem 1 asks whether a named individual can win, while this mirror asks whether a type can win and lets its mass split across matches. That is a real change. I think the case survives because the paper explicitly treats type winners as meaningful and motivates them by players of similar intrinsic ability; the mirror preserves the organizer’s central decision—how to arrange a balanced knockout so the desired champion class emerges. The next questions are whether continuous feasible flows can be rounded to finite type-count tournaments, and how the rounding error behaves as type multiplicities grow.
The strongest objection is that the proposed flow model does not have one knockout tournament at its root. A balanced tournament with \(2^h\) entrants has one winner. In the LP, \(2^{-h}\) is the mass remaining after \(h\) rounds, so a positive mass of champions remains. Splitting that mass through the rounds describes an aggregate of many fractional brackets, not one draw assigning all entrants to a single bracket. The LP is coherent, but it answers a different question: whether a population of brackets can be routed so that their winners have the target type.
A more faithful limit of one large tournament exposes the problem. If \(N=2^h\), the winner has mass \(1/N\), which vanishes as the tournament grows. A type distribution cannot identify that winner; tracking a named target as a zero-mass tag restores the identity, but its success depends on its particular path through the tree. The aggregate flows do not capture that path. Replacing the target individual with a target type is more natural, and the paper itself recognizes type winners. But a single finite bracket still has exact type counts and indivisible entrants. If the atomless version instead asks whether *some* target-type champion can be produced, any type with positive mass supplies arbitrarily many entrants, so the proportions cease to constrain that existential question. To make proportions matter, one must ask about an ensemble of brackets or the fraction of target-type winners—an interesting neighboring problem, but a different objective.
Theorem 2 is evidence against a blanket claim that high multiplicity is unnatural here: the authors explicitly study type-level winners and motivate types by similar ability. It is not evidence that the flow LP already answers a continuous version; its result concerns finite tournaments and finite draws. So the best negative case is a mismatch between a single tournament’s one-winner outcome and the proposed model’s divisible mass of winners. That is a substantial objection to this particular mirror, but it does not defeat every type-based continuous formulation. Given the paper’s own type-level result, I cannot honestly make 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.