{"id":"99d6ac54-dcaf-4c66-9cd8-efed1eb9a296","arxiv_id":"2411.13684","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The paper builds a unified theory of flow-based allocation rules for cooperative games with generalized coalition configurations and characterizes two-step (Owen-type) flow methods by two axioms on the flow.","lead":"This paper introduces a framework for cooperative games where agents face both a coalition configuration (a family of possibly overlapping groups) and restricted cooperation inside each group. It defines and axiomatically characterizes flow-based allocation methods for this setting, including a two-step Owen-style construction, and shows when these methods reduce to methods on ordinary coalition games.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 5's sufficiency direction is false: the two flow axioms constrain only the aggregate edge flow, not the per-agent split, so a flow method can satisfy the axioms without being a two-step method.","rationale":"The reader's weakest-assumption analysis correctly identifies a gap in the proof of Lemma 1: the claimed sink whose deletion preserves connectedness need not exist. That gap is real and should be repaired. However, Lemma 1 itself is true (the oriented incidence matrix of a weakly connected digraph has rank |V|-1), so the gap does not threaten the main theorems once the proof is fixed. The load-bearing problem is different and more severe: Theorem 5, the paper's strongest claim, is false as stated. Theorem 4 is a correct characterization of two-step flows, but Theorem 5 upgrades this to a characterization of two-step flow methods without controlling the individual coefficients λ_i. On edges where several agents enter a coalition together, the aggregate flow does not determine the payoffs; the two axioms are silent about the split, while the two-step procedure forces the split to be independent of the upper-game support S. The concrete 2×2 example exhibits a flow method whose aggregate flow is two-step and satisfies both axioms, but whose per-agent coefficients cannot be written in the form required by Lemma 4. Therefore the central characterization of two-step flow methods fails under the paper's own definitions. The paper contains valuable correct results about flows (Theorems 3 and 4), and a repaired version could restrict Theorem 5 to flow-determined methods satisfying an equal-split condition such as (5), or state the characterization at the level of flows. But as written, the main value-level theorem is false, so the verdict should move to reject.","tokens_in":32363,"tokens_out":18213,"duration_ms":1054238,"concrete_test":"Implement the counterexample above. Set F1={∅,{1},{2,3},{1,2,3}}, F2={∅,{4}}, M={1,2}; choose Λ_M(∅->{1})=Λ_M({2}->{1,2})=1/2 and Λ_1(∅->{2,3})=1 (so each displayed edge has total flow 1/2). Define λ_2=0.45, λ_3=0.05 on ((∅,∅)->({2,3},∅)) and λ_2=0.05, λ_3=0.45 on ((∅,{4})->({2,3},{4})), with all singleton-cover coefficients set equal to the aggregate flow. Verify: (i) the induced Λ is a unitary flow satisfying Null flow and Flow proportionality; (ii) the payoff vector defined by (3) is efficient and marginalist; (iii) the system λ_i(KS,Kq->KS,K'q)=Λ_M(RS\\q,RS)λ^q_i(Kq,K'q) has no solution, because it would force λ_2/Λ_M to be identical on the two displayed edges, but 0.45/0.5 ≠ 0.05/0.5. This directly falsifies the 'if' direction of Theorem 5.","verdict_should_be":"REJECT","load_bearing_attack":"The paper's central value-level claim, Theorem 5, does not follow from Theorem 4 and is false under the paper's definition of a flow method (eqs. (3)-(4)). Theorem 4 characterizes flows: a unitary flow satisfying Null flow and Flow proportionality factorizes as Λ_M(RS\\q,RS)·Λ_q(Kq,K'q). But a flow method is a marginalist value whose coefficients λ_i sum to the aggregate flow. When a cover adds several agents simultaneously, i.e. |Q(K,K')| > 1, the aggregate flow does not determine how it is split among those agents. A two-step flow method (12)-(13) has, by Lemma 4, coefficients of the form λ_i(KS,Kq -> KS,K'q) = Λ_M(RS\\q,RS)·λ^q_i(Kq,K'q). Hence the split among agents entering together on a fixed lower edge must be the same for every upper-game support S. The two axioms are conditions on the total flow Λ only and impose no restriction on that split. Concretely, take M={1,2}, P1={1,2,3}, F1={∅,{1},{2,3},{1,2,3}}, P2={4}, F2={∅,{4}}. Let a=b=1/2 and c=1 define a two-step aggregate flow, so Λ((∅,∅)->({2,3},∅)) = 1/2 and Λ((∅,{4})->({2,3},{4})) = 1/2, with the rest determined by unitary flows on the two-cube and on F2; this flow satisfies both axioms. Now choose a flow method with aggregate flow equal to Λ but with agent 2 receiving 90% of the first edge's flow and 10% of the second, agent 3 receiving the complementary shares. The induced flow is unchanged, so the axioms hold. But Lemma 4 requires agent 2's coefficient on both edges to be 0.5·r for a single r; the values 0.45 and 0.05 cannot both arise. Thus Φ0 is a flow method satisfying the axioms yet is not a two-step flow method. The proof's sentence 'from any two-step flow one can construct a two-step flow method' only constructs one such method, not the given one.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces cooperative games with generalized coalition configuration: a coalition configuration P covering the agent set, with each element equipped with a normal set system F_q, and a coalition profile function on the Cartesian product of the feasible coalitions. It defines marginalist values and flow methods on this domain, shows (Theorem 1 and Theorem 2) that linearity plus the null-agent axiom characterizes marginalist values and that adding efficiency characterizes flow methods. It then constructs Owen-type two-step flow methods via an upper game on the elements of the configuration and lower games on each set system. Theorem 3 shows the induced aggregate flow decomposes as a product of an upper flow and lower flows. Section 7 introduces two axioms, Null flow for non-relevant directed edges and Flow proportionality, proves in Theorem 4 that a flow on the product digraph is a two-step flow iff it satisfies these axioms, and then claims in Theorem 5 that this also characterizes two-step flow methods at the value level. Section 6 also gives a configuration-type value on the unrestricted domain, and Section 8 studies when a flow method on coalition profiles induces a flow method on reachable coalitions. The paper is clearly written and the flow-level decomposition result is interesting, but the central value-level characterization in Theorem 5 is false as stated.","tokens_in":32816,"tokens_out":12832,"duration_ms":137925,"significance":"If corrected, the framework would be a useful and fairly general extension of Weber's flow methods and Owen's two-step procedure to restricted cooperation and coalition configurations. The paper contains detailed axiomatic derivations and no parameter fitting or circular reliance on the target results; Theorem 4's characterization of decomposable flows is a substantive and plausibly correct contribution. However, the advertised value-level characterization of two-step flow methods is not established: Theorem 5 is false under the paper's own definition of a flow method. Since this is the headline result announced in the abstract and introduction, the current manuscript cannot be accepted. A revised version that reframes the main characterization at the level of flows, or that adds an additional axiom governing the split of edge flows among agents, could be worth reconsidering.","major_comments":[{"comment":"The sufficiency direction of Theorem 5 is false. A flow method is not determined by its induced aggregate flow: definition (4) only fixes the sum of the coefficients λ_i over i ∈ Q(K,K'), not the individual coefficients. The two axioms Null flow for non-relevant directed edges and Flow proportionality are conditions on the aggregate flow Λ only, so they are inherited by any flow method with the same aggregate flow, regardless of how the flow is split among the agents entering together on an edge. In contrast, Lemma 4 shows that a two-step flow method has coefficients of the form λ_i^0(K_{S,K_q}, K_{S,K'_q}) = Λ^M(R_{S\\q}, R_S) λ_i^q(K_q,K'_q). Consequently, for a fixed lower edge (K_q,K'_q) and two different upper supports S,S', the ratio λ_i^0(e_S)/λ_i^0(e_{S'}) must be the same for every i ∈ Q; the aggregate axioms impose no such constraint. Concretely, let M={1,2}, P1={1,2,3}, F1={∅,{1},{2,3},{1,2,3}}, P2={4}, F2={∅,{4}}. Let Λ^M(∅,{1})=Λ^M({2},{1,2})=1/2 and let the lower flow on F1 be Λ^1(∅,{2,3})=Λ^1({2,3},{1,2,3})=1, all other lower flows being zero. This defines a two-step aggregate flow Λ satisfying both axioms. Now define a flow method Φ with induced flow Λ but with coefficients on e^1=(∅,∅)→({2,3},∅) given by λ_2=0.45, λ_3=0.05, and on e^2=(∅,{4})→({2,3},{4}) given by λ_2=0.05, λ_3=0.45, completing the remaining coefficients consistently with Λ. Then Λ satisfies the two axioms, but Φ is not a two-step flow method: Lemma 4 would force λ_2(e^1)/λ_2(e^2)=λ_3(e^1)/λ_3(e^2)=1, whereas the chosen ratios are 9 and 1/9. Thus Theorem 5 must be corrected; the axioms characterize two-step flows (Theorem 4), not two-step flow methods. A value-level statement would require an additional axiom on the per-agent split, such as the equal-splitting condition (5).","section":"Section 7, Theorem 5"},{"comment":"The proof of Lemma 1 contains a genuine gap. In the induction step it asserts that there exists a vertex j0 whose deletion leaves the acyclic digraph connected and such that j0 is the tail of no directed edge. The sink-deletion claim is false: in the connected acyclic digraph with vertices a, b, s and edges a→s, b→s, the only sink is s, and deleting s leaves two components. The statement of the lemma is true, and the gap appears repairable (for example, by a different inductive or network-flow argument), but as written the proof of Theorem 1 is incomplete. Since Theorem 1 is the characterization of marginalist values that underlies the rest of the paper, this needs to be fixed.","section":"Section 5, Lemma 1"}],"minor_comments":[{"comment":"In Case (b), the expression Λ^ΦM(µ(R_{µ(K)\\q}, R_{µ(K)}) contains an extraneous 'µ(' and should read Λ^ΦM(R_{µ(K)\\q}, R_{µ(K)}); in Case (c), the summation ∑_{(K'_q,K_q)∈E_i^F} should be over E_{F_q}, not E_i^F.","section":"Section 6, proof of Theorem 3"},{"comment":"In the computation of γ(K), the intermediate displayed line omits the minus sign in front of the outgoing-flow term; the preceding and following lines have the correct sign.","section":"Section 5, proof of Theorem 2"},{"comment":"The phrase 'each pair {S,S'} ⊆ M such that q ∈ S∩S′' is imprecise: S and S' are subsets of M, so the axiom should say 'each pair of subsets S,S′ ⊆ M with q ∈ S∩S′'.","section":"Section 7, Flow proportionality"},{"comment":"References [20] and [29] list the same paper (van den Brink, Khmelnitskaya, and van der Laan, 'An Owen-type value for games with two-level communication structure') and should be merged into a single entry.","section":"References"}],"recommendation":"reject","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The main result to know about this paper is that Theorem 5, the advertised characterization of two-step flow methods, is false as stated. The two axioms (null flow on non-relevant edges, flow proportionality) constrain the aggregate edge flow Λ, but a flow method is a marginalist value whose coefficients λ_i sum to Λ on each edge; the axioms put no restriction on how that aggregate flow is split among the agents entering on that edge. The sufficiency proof only shows that from a two-step flow you can construct a two-step flow method—one such method—not that an arbitrary method inducing that flow must be two-step. The stress-test counterexample works: take M={1,2}, P1={1,2,3}, F1={∅,{1},{2,3},{1,2,3}}, P2={4}, F2={∅,{4}}, and a two-step aggregate flow with half on each lower edge; then split the first edge 90/10 between agents 2 and 3 and the reverse on the second. The aggregate flow is unchanged, so both axioms hold, but the per-agent coefficients are not of the product form required by Lemma 4. The reader's report is too kind here: it flags the Lemma 1 gap but misses this more serious failure.\n\nWhat the paper does well: the framework of generalized coalition configurations is natural and cleanly executed. Theorems 1 and 2 extend Aguilera et al. to product digraphs; Lemma 1's proof gap (the claimed sink whose deletion preserves connectedness does not always exist, e.g., two sources feeding one sink) is real but the lemma is true and easily repaired, so Theorem 1 survives. Theorem 3's flow decomposition is sound, modulo minor typos in case (b). Theorem 4, characterizing two-step flows at the aggregate level, appears correct—the two axioms do single out flows that factor as Λ_M·Λ_q. Theorem 6 gives a nice new configuration-type value for the unrestricted case, and Theorem 7 is a useful reduction result with the caveat that Γ*_{F0} is not always the covering digraph. There is no circularity or fitting; the axiomatic derivations are honest.\n\nThe soft spot is exactly the leap from Theorem 4 to Theorem 5. The distinction between flow and flow method is where the characterization fails, and Theorem 5 is the paper's central advertised claim. Repairing it requires an additional axiom that constrains the per-agent split (for instance, requiring the split on a given lower edge to be independent of the upper support), or re-scoping the result to two-step flows rather than flow methods.\n\nWho this is for: researchers working on Owen-type values, restricted cooperation, and flow methods in cooperative game theory. The framework and the aggregate-level results are worth having, but the paper needs substantial revision before it should appear. I would send it to a serious referee, but with explicit instructions to scrutinize Theorem 5 and the method-level interpretation of the flow axioms.","headline":"Theorem 5's sufficiency direction is false: the two flow axioms characterize two-step flows at the aggregate level, but not flow methods, because the per-agent split of each edge's flow is unconstrained by the axioms.","tokens_in":33391,"tokens_out":3354,"would_cite":false,"duration_ms":35852,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["91A12"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper characterizes flow methods for cooperative games with restricted cooperation inside coalition configurations, and shows that two-step allocation procedures are exactly the flow methods whose induced flows vanish on non-relevant…","keywords":["cooperative games","coalition configuration","restricted cooperation","flow methods","marginal values","two-step allocation","product digraph","set systems"],"falsifier":"Inspect the deletion step in the proof of Lemma 1 on a connected acyclic digraph with two sources and one sink: the lemma's stated induction needs a sink whose removal leaves the graph connected, and whether such a sink always exists is precisely what must be checked. A counterexample to that step would leave Theorem 1 underived as written, even if the lemma itself can be proved by another argument.","tokens_in":32167,"feed_emoji":"🔀","tokens_out":7802,"duration_ms":82981,"temperature":0.7,"pith_summary":"The paper introduces cooperative games in which a coalition configuration covers the agent set and, inside each element of the configuration, only certain coalitions are feasible; the worth is a function on profiles of feasible coalitions. It proves that any value satisfying linearity and the null-agent axiom must be marginalist: each agent's payoff is a sum of coefficients times marginal contributions along covering relations. Adding efficiency forces those coefficients to form a unitary flow on the product digraph built from the set systems, so payoff rules of this kind are exactly flow methods. The paper then characterizes the flow methods that come from a two-step allocation procedure: they are exactly the flows that vanish on non-relevant edges and obey a flow-proportionality condition on relevant edges. This extends the flow-method theory of transferable-utility games to restricted cooperation inside coalition configurations.","feed_headline":"Restricted coalition payoffs become flows on one product digraph","feed_subtitle":"Efficiency, linearity, and null-agent axioms pick out unitary flows; two flow axioms identify the two-step procedures.","key_machinery":"The central object is the product digraph $\\Gamma_{\\mathcal{F}} = (\\mathcal{F}, E_{\\mathcal{F}})$ formed from the cartesian product of the covering-relation digraphs of the set systems $(P_q, \\mathcal{F}_q)$. A flow method is a value whose coefficients aggregate to a unitary flow on $\\Gamma_{\\mathcal{F}}$; relevant directed edges are those whose endpoints are relevant coalition profiles, built from a support $S$ of the configuration and a coalition $K_q$ in the $q$-th set system. The two-step construction factors each relevant edge's flow as $\\Lambda_{\\mathcal{M}}(R_{S\\setminus\\{q\\}}, R_S) \\cdot \\Lambda_q(K_q, K'_q)$, a product of a flow on the $m$-dimensional directed hypercube and a flow on $\\Gamma_{\\mathcal{F}_q}$. Flow proportionality is the relation $\\Lambda(K_{S,K_q}, K_{S,K'_q})\\Lambda(K_{S',L_q}, K_{S',L'_q}) = \\Lambda(K_{S',K_q}, K_{S',K'_q})\\Lambda(K_{S,L_q}, K_{S,L'_q})$ that, together with the null-flow condition on non-relevant edges, characterizes two-step flows.","core_discovery":"On the domain of games with generalized coalition configuration, the combination of Linearity, Null agent, and Efficiency characterizes exactly the flow methods: values whose coefficients define a unitary flow on the product digraph $\\Gamma_{\\mathcal{F}}$. The paper's central structural result is that the two-step procedure—first allocate worth among the elements of the configuration through an upper game, then allocate each element's share to its agents through lower games—produces precisely those flow methods whose induced flow vanishes on non-relevant directed edges and satisfies flow proportionality on relevant edges (Theorem 5). For the unrestricted case, three axioms (null flow, intracoalitional anonymity, coalitional anonymity) single out a closed-form configuration value that extends the classical configuration value to coalition profiles. Finally, under a condition on the product digraph that holds for partitions and for regular set systems, a profile-level flow method induces a coalition-level flow method on reachable coalitions, so the theory projects back to games whose unit of cooperation is an ordinary coalition.","pith_inferences":["If the characterization holds, the same flow calculus should extend to any coalition configuration whose set systems are normal; the only obstacle is the projection condition in Theorem 7, which is sufficient but not shown necessary.","The worth of a coalition profile can be read as modeling externalities between elements of the configuration; flow proportionality then gives a testable separability condition on how cross-element externalities are priced.","One natural extension is to replace uniform splitting of the flow among entering agents by other sharing rules and ask whether the same two flow axioms still characterize the resulting two-step methods.","Because Theorem 5 characterizes two-step methods solely by flow axioms, an algorithm could check whether a given allocation rule is two-step by computing its induced flow and verifying the two axioms on the product digraph."],"forward_implications":["Efficiency, linearity, and the null-agent axiom are exactly equivalent to unitary flows on $\\Gamma_{\\mathcal{F}}$, so every such value can be studied through its flow rather than through its coefficients.","Two-step flow methods are precisely those with zero flow on non-relevant edges and flow proportionality on relevant edges; checking these two conditions on a flow decides whether it arises from the two-step procedure.","For unrestricted feasible sets within each element, the three axioms determine a unique value with explicit closed-form coefficients; this is the profile-level analogue of the classical configuration value.","When the coalition configuration is a partition, or each set system is regular, every profile-level flow method induces a coalition-level flow method on reachable coalitions, so configuration-style values extend to restricted cooperation.","Maximal-path constructions, where a grand profile is formed step by step, are special two-step flow methods, so Shapley-style equal treatment of maximal paths sits inside the characterized family."],"supporting_citations":[{"why":"Supplies the marginalist and flow characterization for TU-games with restricted cooperation that Theorems 1 and 2 generalize to coalition profiles.","marker":"[10]"},{"why":"Defines the configuration value whose two-step construction is extended to coalition profiles and restricted cooperation.","marker":"[2]"},{"why":"Supplies the classic two-step allocation procedure for coalition structures that the paper adapts to its generalized setting.","marker":"[27]"},{"why":"Introduces probabilistic (flow) values, the notion of flow method used throughout the paper.","marker":"[31]"},{"why":"Provides the maximal-path construction used to build Shapley-style two-step flows.","marker":"[28]"},{"why":"Defines Shapley values on convex geometries, an example of regular set systems to which the two-step construction applies.","marker":"[16]"},{"why":"Defines Shapley values on augmenting systems, another source of lower-game flows in the construction.","marker":"[19]"},{"why":"Develops flow methods on regular games, the regular-set-system case used in Section 8.","marker":"[24]"}],"fun_headline_variants":["Coalition payoffs become flows on configuration digraphs","Unitary flows characterize new cooperative game values","Two-step flow method for generalized coalition games","Axioms single out flow methods for coalition profiles","Coalition profiles yield flow-based solution values"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The marginalist characterization rests on the lemma that every zero-sum function on the vertices of a connected acyclic directed graph can be expressed as the divergence of an edge-weight function; if that representation is unavailable, Theorem 1 does not go through.","fun_headline_variants_meta":{"raw":{"variants":["Coalition payoffs become flows on configuration digraphs","Unitary flows characterize new cooperative game values","Two-step flow method for generalized coalition games","Axioms single out flow methods for coalition profiles","Coalition profiles yield flow-based solution values"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000625,"raw_usage":{"total_tokens":2874,"prompt_tokens":904,"completion_tokens":1970,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":520,"completion_tokens_details":{"reasoning_tokens":1897}},"tokens_in":520,"tokens_out":1970,"duration_ms":13636,"temperature":1.0,"reasoning_tokens":1897,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T16:01:20.206432+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Inspect the deletion step in the proof of Lemma 1 on a connected acyclic digraph with two sources and one sink: the lemma's stated induction needs a sink whose removal leaves the graph connected, and whether such a sink always exists is precisely what must be checked. A counterexample to that step would leave Theorem 1 underived as written, even if the lemma itself can be proved by another argument.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the marginalist and flow characterization for TU-games with restricted cooperation that Theorems 1 and 2 generalize to coalition profiles."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines the configuration value whose two-step construction is extended to coalition profiles and restricted cooperation."},{"cited_title":"In: Hernn , R., Moschlin, O","cited_arxiv_id":null,"evidence_quote":"Supplies the classic two-step allocation procedure for coalition structures that the paper adapts to its generalized setting."},{"cited_title":"In: Roth, A .E","cited_arxiv_id":null,"evidence_quote":"Introduces probabilistic (flow) values, the notion of flow method used throughout the paper."},{"cited_title":"In: Kuhn, H.W ., Tucker, A.W","cited_arxiv_id":null,"evidence_quote":"Provides the maximal-path construction used to build Shapley-style two-step flows."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines Shapley values on convex geometries, an example of regular set systems to which the two-step construction applies."},{"cited_title":"Axiomatizations of the Shapley value for games on augmenting systems European Journal of Operational Research , 196: 1008–1014 (2009)","cited_arxiv_id":null,"evidence_quote":"Defines Shapley values on augmenting systems, another source of lower-game flows in the construction."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Develops flow methods on regular games, the regular-set-system case used in Section 8."}],"review_version":1}