{"id":"380b02a7-a83d-4970-a39d-f2a0cd78058a","arxiv_id":"2411.17587","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A new stochastic extensive form framework replaces nature with a lottery draw and oracles, enabling continuous-time noise such as Brownian motion to be represented while preserving classical extensive form foundations.","lead":"This paper builds a general extensive-form framework in which a single lottery draw selects a decision tree and each agent receives private updates from an oracle, replacing the classical 'nature' agent. It gives a precise decision-theoretic foundation for stochastic games with continuous-time noise such as Brownian motion.","discovery_kind":"unification","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The strict-generalization claim for Brownian noise is asserted but not demonstrated: Theorem 3.22 excludes continuous time, and Section 3.8 refutes a nature representation without exhibiting a well-posed stochastic extensive form for Brownian motion.","rationale":"The reader identifies the companion paper [53] as the weakest load-bearing premise. That is a legitimate concern about self-containedness, but the more directly load-bearing gap is internal to this paper's central claim: the strict generalization for Brownian motion requires a well-posed stochastic extensive form in continuous time, and none is constructed. Theorem 3.22 explicitly limits well-posedness to well-ordered time; Section 3.8 only argues that a particular nature representation fails; the promised third paper is invoked for the approximation theory. Thus the conclusion overstates what is demonstrated. This supports the CONDITIONAL verdict already given, but for a different reason than the reader's companion-paper concern. The concern is not that the framework is inconsistent, only that the headline application—Brownian noise as a strict generalization—is unproven as of this paper. A concrete order-theoretic check can settle whether the standard action path construction even yields a well-posed continuous-time sef; if it fails, the authors need either a new construction or a careful scoping of the claim. Therefore no verdict change is warranted; the condition is exactly that the Brownian strict-generalization claim be either proved or explicitly deferred.","tokens_in":61354,"tokens_out":9984,"duration_ms":108877,"concrete_test":"Construct the action path sef candidate of Section 2 with continuous time T = R+, action space A = R, exogenous scenario space Omega = C(R+, R) under Wiener measure, W = Omega x A^T, Brownian filtration F, and history structure H_t partitioning by past paths. Then check the order-theoretic conditions of Corollary 3.13 on the induced decision forest. For a node x_t (outcomes sharing the same past up to t), the chain {x_u : u > t} in downset(x_t)\\{x_t} has no maximum because R+ is dense, so weak up-discreteness fails. Hence this continuous-time candidate is not well-posed, confirming that Theorem 3.22 does not extend and that the paper contains no exhibited well-posed Brownian sef. If a different construction is intended, it must be supplied explicitly; otherwise the strict-generalization claim should be scoped to the announced approximation program.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that stochastic extensive forms strictly generalize classical extensive form theory by accommodating Brownian motion rests on the existence of a well-posed stochastic extensive form for such noise. That existence is not established in the paper. Theorem 3.22 proves well-posedness only for action path sef data with well-ordered time, and the text immediately cautions that 'nothing is said about continuous time.' Section 3.8 argues that a nature representation of Brownian motion in the action-path class fails: perfect endogenous recall for nature forces singleton history partitions, and the resulting forest is not weakly up-discrete. But this negative argument is about the nature representation, not about the new framework. No stochastic extensive form with Brownian-noise filtration is constructed, and its well-posedness is not verified. Corollary 3.13 makes the burden precise: well-posedness requires the underlying decision forest to be weakly up-discrete, coherent, and regular. For the natural continuous-time action path construction, the density of the time axis appears to destroy weak up-discreteness, so the standard construction does not supply the needed example. The conclusion's assertion that general stochastic processes can be implemented 'without encountering outcome generation problems for a nature agent' is therefore a promissory note: it shows that arbitrary filtrations can be encoded as exogenous information structures, but not that the resulting extensive form has a well-defined outcome map or equilibrium concept in continuous time.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a general theory of stochastic extensive forms built on the companion paper's stochastic decision forests. A stochastic extensive form replaces the traditional nature agent with a single lottery draw selecting a decision tree, while personal agents receive dynamic exogenous information through sigma-algebra-valued information structures and make partition-refining adapted choices. The paper defines strategies as Savage acts, characterizes well-posedness scenario-wise (Theorem 3.11), gives an order-theoretic characterization (Corollary 3.13), constructs action-path stochastic extensive forms (Section 2, Theorem 2.6), proves well-posedness for well-ordered time (Theorem 3.22), and argues that Brownian noise cannot be represented through a nature agent but can be accommodated by the new framework (Section 3.8). It then defines expected-utility preferences, dynamic consistency, dynamic rationality, and a generalized perfect Bayesian equilibrium.","tokens_in":61564,"tokens_out":7767,"duration_ms":78351,"significance":"If the framework is sound, this is a substantial conceptual unification: it bridges refined-partitions extensive form theory with filtrations from probability theory, and it provides scenario-wise tests for well-posedness that import the classical Alós-Ferrer–Ritzberger machinery. The action-path construction (Theorem 2.6) and the measurability characterization of adapted choices (Theorem 2.8) are useful and nontrivial tools. The paper also gives a clean reduction, Theorem 3.11, of stochastic well-posedness to scenario-wise classical well-posedness, and Corollary 3.13 expresses well-posedness in verifiable order-theoretic terms. The main caveat is that the paper's headline claim about continuous-time processes such as Brownian motion is not backed by a positive well-posedness result in this manuscript; it is a goal deferred to the third paper.","major_comments":[{"comment":"The central claim that stochastic extensive forms strictly generalize classical extensive form theory by accommodating Brownian motion is not established in this paper. Theorem 3.22 proves well-posedness only for action-path sef data with well-ordered time and explicitly says \"nothing is said about continuous time.\" Section 3.8 only shows that a particular nature-agent representation of Brownian motion is not well-posed because perfect endogenous recall forces singleton history partitions and the resulting forest is not weakly up-discrete. No well-posed stochastic extensive form with Brownian-noise filtration is constructed, and no proof is given that any continuous-time action-path sef is weakly up-discrete, coherent, and regular, as required by Corollary 3.13. The concluding sentence \"It is possible to implement general stochastic processes as background noise ... without encountering outcome generation problems\" is therefore a promissory note. The paper should either provide such a construction and its well-posedness proof, or explicitly rephrase the claim as a program for the third paper.","section":"Section 3.8 and Conclusion"},{"comment":"Two load-bearing verification steps are explicitly omitted. Footnote 13 in Section 3.3 states \"We omit the formal argument behind this statement\" when claiming that after any endogenous information set a well-posed stochastic extensive form is induced. Remark 3.19 states \"We do not bother the reader with the verification of the claim that this yields again a well-posed stochastic extensive form and an eu preference structure\" for the multiple-selves construction. These are not cosmetic omissions: the dynamic rationality and equilibrium definitions condition on induced decision situations, and Remark 3.19 is used to justify the multiple-selves analysis of the absent-minded driver. Both statements should be proved in the appendix or explicitly marked as assumptions rather than consequences.","section":"Footnote 13 and Remark 3.19"},{"comment":"The negative argument about the nature representation of Brownian motion is itself asserted rather than proved. The paper says that with singleton history partitions \"the corresponding decision forest is not weakly up-discrete\" and cites well-known counterexamples from [62, 64, 5], but it does not formalize which continuous-time action-path pseudo-sef is being considered or show in detail why weak up-discreteness fails for Brownian paths. Since this is the only concrete evidence offered for the strict-generalization claim, the argument should be made precise, with the relevant nodes, histories, and maximal chains described explicitly or with a formal derivation from Theorem 2.9 and Corollary 3.13.","section":"Section 3.8, Brownian motion argument"}],"minor_comments":[{"comment":"Axiom 6 uses the condition P(c') = P(c), while Lemma 1.12's construction uses the weaker condition P(hat c) ⊆ P(c). The relationship between these two conditions should be clarified, since the lemma claims to produce a stochastic extensive form satisfying Axiom 6.","section":"Definition 1.7 and Lemma 1.12"},{"comment":"The statement \"By Theorem 2.9, perfect (endogenous) recall requires H_t to contain only singletons\" is not immediate from the theorem as stated, because Theorem 2.9 gives a condition on pairs of partition cells rather than a direct singleton characterization. A short derivation or a pointer to the relevant part of [53] would help.","section":"Section 3.8, Theorem 2.9 usage"},{"comment":"The symbol W is used both for the underlying set of outcomes in Section 1 and for the sigma-algebra on W in Definition 3.15. This overloading is confusing and should be resolved, for example by denoting the sigma-algebra by S or A.","section":"Definition 3.15"},{"comment":"The paper relies heavily on [53], which is cited as \"Mimeo,\" for Definitions 1.3-1.6 and many lemmas. Since the current paper is not self-contained without [53], the author should indicate how the companion paper can be accessed or include a more detailed summary of the imported definitions and lemmas.","section":"References and self-containedness"}],"recommendation":"major_revision","confidential_remarks":"The paper is well structured and the main technical machinery appears promising, but the central Brownian-motion claim is not supported by a positive well-posedness result in the current text. The two explicitly omitted verifications (footnote 13 and Remark 3.19) are also load-bearing for the equilibrium concept as stated. These issues are fixable by adding proofs or by substantially qualifying the claims, so I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Thanks for the report. I read the paper and I largely agree with your conditional verdict, though I'd put a bit more weight on the unproven continuous-time claim.\n\nWhat's genuinely new: the stochastic extensive form (sef) built on stochastic decision forests is a real extension of the Alos-Ferrer-Ritzberger framework, and the action path construction is a useful bridge to stochastic process models. The scenario-wise well-posedness theorem (3.11) is clean and gives a practical reduction to classical results. The treatment of the absent-minded driver via the Heraclitus property is sensible and gives a decision-theoretic resolution consistent with Gilboa. The verification of the axioms in the simple examples is also careful.\n\nThe weak spots: the paper's central selling point, that it strictly generalizes classical extensive form theory by accommodating Brownian noise, is not actually demonstrated. Theorem 3.22 only gives well-posedness for well-ordered time, and the text explicitly says continuous time is not covered. Section 3.8 argues that a nature representation with perfect endogenous recall fails, but that's a negative result. No well-posed stochastic extensive form with a Brownian-motion filtration is constructed or verified. So the conclusion's sentence about implementing general stochastic processes 'without outcome generation problems' is a promissory note. This is a load-bearing gap, not a technicality, because the strict-generalization claim rests on it.\n\nSecond, the paper imports core machinery from the companion paper [53] (definitions of stochastic decision forests, order consistency, adapted choices) and all main theorems rely on lemmas from it. That paper is unpublished, so a referee cannot fully check the foundation. The author should either make [53] available or include the necessary parts.\n\nThird, there are two explicitly omitted verification steps: footnote 13 and Remark 3.19. Both are auxiliary, but they should be supplied.\n\nOn balance, the framework is coherent, the proofs I checked are careful, and the mathematical structure looks internally consistent. The main issue is that the headline application is unfinished. This deserves a serious referee—it should go to peer review—but the referee should require either a concrete well-posed continuous-time example with Brownian noise or a much more modest claim. I'd temper the conclusion until then.\n\nThe right readers are people working on extensive form foundations and stochastic games. I'd engage with it, but with the above caveats.","headline":"Serious, coherent framework, but the Brownian-motion strict-generalization claim is not yet backed by a constructed example.","tokens_in":62143,"tokens_out":2358,"would_cite":true,"duration_ms":23348,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["91A15","91A18","91B06"],"pacs":[],"model":"deepseek-v4-flash","headline":"Stochastic extensive forms strictly generalize classical extensive form theory by representing Brownian motion as noise without a nature agent.","keywords":["extensive form games","stochastic games","dynamic games","decision making","sequential decision theory","stochastic processes","Bayesian games","Brownian motion"],"falsifier":"Exhibit a classical extensive form with a nature agent that is well-posed and whose induced outcomes are Brownian motion paths; such an example would directly refute the paper's central strict-generalization claim that Brownian noise cannot be nature-represented.","tokens_in":61093,"feed_emoji":"🎲","tokens_out":9650,"duration_ms":72303,"temperature":0.7,"pith_summary":"This paper introduces stochastic extensive forms, a generalization of classical extensive form game theory in which exogenous randomness is modeled by a single lottery draw over decision trees rather than by a nature agent making moves. The framework claims to represent general stochastic processes, including Brownian motion, that cannot be represented by any nature agent while preserving well-posed outcome generation and a dynamic equilibrium concept. If correct, it gives a rigorous decision-theoretic foundation to stochastic differential games and stochastic control problems that currently lack one. The paper also constructs a broad class of action-path stochastic extensive forms and shows when they are well-posed.","feed_headline":"Brownian motion enters game theory without a nature agent","feed_subtitle":"Single lottery draws over decision forests let continuous-time stochastic processes become rigorous decision problems.","key_machinery":"The central object is the stochastic decision forest (sdf): a forest of decision trees indexed by an exogenous scenario space, with a surjective map sending nodes to scenarios and a set of random moves—sections of moves defined on events—together with exogenous information structures (sigma-algebras on random-move domains) and adapted choices. This machinery, imported from the companion paper and combined here with the axioms of a stochastic extensive form, carries the argument: it replaces the nature agent, allows filtrations to model noise, and enables the scenario-wise reduction of well-posedness to classical extensive forms.","core_discovery":"The central claim is that stochastic extensive forms—decision forests over an exogenous scenario space equipped with filtration-like information structures and partition-refining choices—provide a strict generalization of classical extensive form theory. Section 3.8 argues that while many stochastic forms can be represented by a nature agent, a class of continuous-time noise, paradigmatically Brownian motion, cannot: any nature representation forces perfect endogenous recall on the nature agent, which makes the induced classical extensive form ill-posed. Stochastic extensive forms avoid this by making exogenous information flow through sigma-algebras rather than through a virtual decision maker. The paper proves scenario-wise well-posedness (Theorem 3.11) and defines a dynamic rationality and equilibrium concept that generalizes perfect Bayesian equilibrium, thereby claiming that every well-posed stochastic differential game can in principle be given an extensive form foundation.","pith_inferences":["The scenario-wise reduction of well-posedness suggests that approximation theory for stochastic differential games can be built by approximating continuous-time decision forests by well-ordered time-indexed forests, at the level of trees and choices rather than payoffs.","The framework's independence from nature representations implies that concepts like closed-loop versus open-loop strategies in stochastic differential games could be defined purely in terms of the exogenous information structures attached to random moves.","The trade-off the paper identifies between richness of endogenous information and existence of expected-utility preference structures points to a testable design principle: coarse endogenous information partitions make non-trivial expected-utility preference structures easier to construct, supporting an approximation program in which fine information is approached by coarser well-posed forms."],"forward_implications":["Every stochastic extensive form with action-path data and well-ordered time is well-posed (Theorem 3.22), so large classes of discrete-time and long-cheap-talk-like stochastic games receive a rigorous extensive form foundation.","Well-posedness of a stochastic extensive form is equivalent to the underlying decision forest being weakly up-discrete, coherent, and regular (Corollary 3.13), a purely order-theoretic test.","Dynamic rationality and equilibrium, generalizing perfect Bayesian equilibrium, are defined for all well-posed stochastic extensive forms, covering Bayesian, correlated, and subgame-perfect equilibria as special cases (Remark 3.18).","Brownian motion cannot be represented by a nature agent in a well-posed classical extensive form (Section 3.8), so the stochastic extensive form framework is a strict generalization of classical theory.","Measurability assumptions on choices in the literature appear as a consequence of adaptedness rather than as an ad hoc technical condition."],"supporting_citations":[{"why":"Supplies the stochastic decision forest machinery (definitions of sdf, order consistency, maximality, adapted choices) on which every construction in this paper builds.","marker":"[53]"},{"why":"Classifies well-posedness of classical extensive forms in order-theoretic terms; the paper extends this classification scenario-wise in Theorem 3.11.","marker":"[4]"},{"why":"Provides the refined-partitions extensive form theory that stochastic forms generalize and the order-theoretic properties used in the well-posedness characterization.","marker":"[5]"},{"why":"Gives continuous-time extensive form counterexamples showing why nature representations of Brownian-like paths fail to be well-posed.","marker":"[62]"},{"why":"Analyzes maximal strategy sets for continuous-time games, used in the argument that classical theory must restrict outcomes to piecewise-constant paths.","marker":"[64]"},{"why":"Supplies the acts-as-maps formulation that strategies in stochastic extensive forms are defined to be.","marker":"[57]"},{"why":"Provides the interpretation of randomisation via exogenous signals and correlation, on which the paper's treatment of randomisation and equilibrium builds.","marker":"[11]"},{"why":"The Bayesian games and Harsanyi doctrine that the paper's nature-representation discussion and equilibrium generalization respond to.","marker":"[38,39,40]"}],"fun_headline_variants":["No nature agent needed for Brownian decision games","Stochastic extensive forms generalize game theory to continuous time","Brownian motion becomes rigorous without a nature player","Single lottery draw replaces nature in stochastic extensive forms","Sigma-algebras let continuous-time noise enter extensive form games"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the stochastic decision forest machinery imported from the companion paper is sound and general enough to represent continuous-time paths; if that foundation has a flaw, the stochastic extensive forms and the strict-generalization claim built on it collapse.","fun_headline_variants_meta":{"raw":{"variants":["No nature agent needed for Brownian decision games","Stochastic extensive forms generalize game theory to continuous time","Brownian motion becomes rigorous without a nature player","Single lottery draw replaces nature in stochastic extensive forms","Sigma-algebras let continuous-time noise enter extensive form games"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000183,"raw_usage":{"total_tokens":1263,"prompt_tokens":842,"completion_tokens":421,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":458,"completion_tokens_details":{"reasoning_tokens":347}},"tokens_in":458,"tokens_out":421,"duration_ms":5305,"temperature":1.0,"reasoning_tokens":347,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T11:56:40.946701+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Exhibit a classical extensive form with a nature agent that is well-posed and whose induced outcomes are Brownian motion paths; such an example would directly refute the paper's central strict-generalization claim that Brownian noise cannot be nature-represented.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the acts-as-maps formulation that strategies in stochastic extensive forms are defined to be."},{"cited_title":"Emanuel Rapsch","cited_arxiv_id":null,"evidence_quote":"Supplies the stochastic decision forest machinery (definitions of sdf, order consistency, maximality, adapted choices) on which every construction in this paper builds."},{"cited_title":"Trees and exte nsive form","cited_arxiv_id":null,"evidence_quote":"Classifies well-posedness of classical extensive forms in order-theoretic terms; the paper extends this classification scenario-wise in Theorem 3.11."},{"cited_title":"The Theory of Extensive Form Games","cited_arxiv_id":null,"evidence_quote":"Provides the refined-partitions extensive form theory that stochastic forms generalize and the order-theoretic properties used in the well-posedness characterization."},{"cited_title":"Simon and Maxwell B","cited_arxiv_id":null,"evidence_quote":"Gives continuous-time extensive form counterexamples showing why nature representations of Brownian-like paths fail to be well-posed."},{"cited_title":"Stinchcombe","cited_arxiv_id":null,"evidence_quote":"Analyzes maximal strategy sets for continuous-time games, used in the argument that classical theory must restrict outcomes to piecewise-constant paths."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the interpretation of randomisation via exogenous signals and correlation, on which the paper's treatment of randomisation and equilibrium builds."}],"review_version":1}