{"id":"4c0dc63a-30c7-4a43-a70f-524b050cc089","arxiv_id":"2611.17017","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A genealogy-valued Cannings process with block-resampling is constructed on the continuum hierarchical group in the stochastic regime, and its ancestral web is shown to be strongly dual to a Levy web.","lead":"This paper constructs the family-tree process, the ancestral web, and its dual Levy web for a population model on the continuum hierarchical group, and only in the regime where the underlying frequency process is random. It gives uniqueness, long-time behavior, and scaling universality classes, tools that can be reused in mathematical population genetics.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek/deepseek-v4-flash","headline":"The central theorems rest on the block-resampling extension of the in-preparation [GKWb]/[GKWa] theory; Proposition 3.12(a) and Lemma 4.4(c), flagged with 'It remains to prove', are not established here, and Appendix A.2 defers Polishness of A and A* to the same sources.","rationale":"This paper attempts something genuinely hard: continuum-space genealogy-valued processes via L-truncation and duality. Credit is due for the clean statements of the main theorems, the explicit regime criterion (2.2), the explicit operators (2.17)-(2.34), and the skeletal but real proofs of Lemmas 4.1-4.7 and of the theorems in Sections 5-6. The reader's weakest assumption is exactly right: the countable-space genealogy and web theory of [GKWb]/[GKWa] with block-resampling is the foundation of the L-truncation construction, and it is not available inside this text. My pass sharpens the reader's concern in three places. First, Section 4.4 flags Lemma 4.4(c) with 'It remains to prove that this holds still with block-resampling added', and the proof of Lemma 4.4 in Section 4.5 does not cover part (c); this is a directly located admission of an incomplete proof on the path to X^dagger and hence to Theorems 2.13-2.14. Second, Proposition 3.12's 'must be extended' wording indicates the extension is not contained in [GKWb]/[GKWa] either; it is work that remains to be done. Third, Theorem 2.7's well-posedness is reduced to a one-sentence preservation argument for (3.21), and the proof of Theorem 2.16 justifies uniform cadlag tightness by 'jump rates are summable by condition (2.2)', which (2.2) does not imply (e.g., c_k = 1, lambda_k = 2^{-k} gives regime A with sum c_k = infinity); this is a local overstatement in a sketch, but it corroborates the pattern of key convergence steps being asserted rather than demonstrated. I found no internal contradiction forcing a stronger verdict and no basis for accepting the claims as fully established; the concern lands exactly as a conditional gap, so the reader's CONDITIONAL verdict stands unchanged.","tokens_in":57699,"tokens_out":22229,"duration_ms":133117,"concrete_test":"Complete the flagged step inside the paper's own framework: for fixed L, prove Proposition 3.12(a) and Lemma 4.4(c) by verifying the generator duality identity (3.21) between A_Cann,b of (2.28)-(2.30) and the Lambda_k-block-coalescence operator on the full monomial domain Pi_web, including the local finiteness of the countable sum in (2.28) on that domain, the factor alpha = integral x Lambda_L(dx) of (2.30), and the projective dyadic-grid limit m -> infinity that defines the entrance law inf-inf cW^L, all without invoking [GKWa]/[GKWb]. If the verification uses only (2.2) and the stated parameter conditions, the central claims stand conditional on the companion papers; if it requires new hypotheses or genuinely unpublished results, then Theorems 2.1, 2.7, 2.13 and 2.14 are not established as stated.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Theorems 2.1 and 2.7 claim that in regime A, characterized by (2.2), the L-truncated genealogy processes converge to a unique strong Markov Feller U^{Omega_N^infinity x K}-valued process solving a well-posed martingale problem, and Theorems 2.13-2.14 claim strong graphical duality of the ancestral web W and the Levy-web cW in X^dagger. Every one of these theorems is built on the L-truncation of finite-L processes on the countable groups Omega_L^N, whose genealogy and web theory is supplied by [GKWb] and [GKWa], both listed as 'In preparation 2026'. The load-bearing step is the extension of that theory to block-resampling. Proposition 3.12 states outright that those references 'must be extended to include the mechanism of block-resampling for the genealogies' and then asserts parts (a)-(c) as consequences; the paper's own contribution (Appendix A.3, Section 4.5) gives the lifted operators (2.28)-(2.34) and a generator-relation check, but not the well-posedness of the finite-L martingale problem, the Feller property, or the entrance laws. Lemma 4.4(c), needed to obtain the entrance law inf-inf cW^L defining X^dagger and hence cW (the basis of Theorems 2.13-2.14), is explicitly flagged in Section 4.4 with 'It remains to prove that this holds still with block-resampling added', and the proof of Lemma 4.4 in Section 4.5 addresses only the duality and rebirth-identification parts. Proposition 3.6 and Appendix A.2 likewise defer the Polishness of A and A* to [GKWb, GKWa], and Section 5.4 reduces the well-posedness of the limiting martingale problem to a one-sentence preservation argument for (3.21). The authors acknowledge these gaps, so this is a conditional gap rather than an observed contradiction, but as it stands the chain from Proposition 3.12 through Lemma 4.4(c) to Theorems 2.1, 2.7, 2.13 and 2.14 is not completed within the text.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek/deepseek-v4-flash","summary":"The paper constructs and analyses genealogical processes and an ancestral web for a continuum-space hierarchical Cannings model with block-resampling on the uncountable geographic space Ω_N^∞. The main results include: (i) construction of a U^{Ω_N^∞ × K}-valued genealogy process U as an L-truncation limit, with a well-posed martingale problem and a dichotomy between stochastic (regime A) and deterministic (regime B) behaviour; (ii) construction of an ancestral web W and its dual Lévy-web Ŵ, with strong graphical duality in an ancestral space X^†; (iii) identification of universality classes via continuum-space rescaling limits. The proofs rely on duality, L-truncation, tightness, and martingale problem techniques, building on the companion paper [GdHa] and the in-preparation works [GKWb, GKWa].","tokens_in":58155,"tokens_out":1775,"duration_ms":17576,"significance":"If the results hold, this paper would provide a rigorous framework for evolving genealogies and ancestral webs on a continuum hierarchical geographic space, extending the theory from countable groups to an uncountable Polish space. The construction of the Lévy-web and the strong graphical duality are novel and could serve as a template for similar models on other ultrametric spaces. The identification of universality classes via rescaling limits is also valuable. However, the significance is tempered by the heavy reliance on unpublished companion papers for load-bearing steps.","major_comments":[{"comment":"Proposition 3.12 is the cornerstone for the existence of the genealogical coalescents on countable geographic spaces with block-resampling, yet the text states that the references [GKWb, GKWa] 'must be extended to include the mechanism of block-resampling for the genealogies'. The paper does not carry out this extension; it only sketches the lifted operators in Appendix A.3 and checks a generator relation in §4.5. Without a complete proof of Proposition 3.12, the existence of the finite-L processes U^L, W^L, and their duals is not established, and the entire truncation limit argument collapses. This is a load-bearing gap.","section":"§3.3.2, Proposition 3.12","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"Lemma 4.4(c) asserts that the entrance law ∞∞Ŵ^L exists and defines a unique X^{†,L} for all dyadic times simultaneously, but the text explicitly says 'It remains to prove that this holds still with block-resampling added'. The proof in §4.5 addresses only the duality and rebirth-identification parts, not the existence of the entrance law for the block-resampling case. This lemma is needed for the construction of Ŵ and X^†, so the gap is critical.","section":"§4.4, Lemma 4.4(c)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"Proposition 3.6 claims that the state spaces A and A* are Polish, but the justification is deferred to the in-preparation references [GKWb, GKWa] and Appendix A.2 merely states that 'the necessary information is collected' there. Since the topology and completeness of these spaces are essential for the martingale problem and tightness arguments, the lack of a self-contained proof is a significant omission.","section":"§3.1.1, Proposition 3.6 and Appendix A.2","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The construction of U as the L→∞ limit of U^L relies on the convergence of the dual processes Y^L and the duality function H^L. The convergence of Y^L is asserted in Proposition 3.11(b), which itself depends on Proposition 3.12 and Lemma 4.4. As noted above, those are not fully proved. Hence the central existence theorem is not yet supported.","section":"§2.1.1, Theorem 2.1(a)(i)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"Theorem 2.16 claims convergence of Ŵ^L to Ŵ^∞ in A*,V. The proof in §6.6 invokes Lemma 4.6(a) for tightness, but Lemma 4.6(a) is itself stated as a claim whose proof relies on the unproven Lemma 4.4(c) and the tightness criteria that are not fully verified. This creates a circular dependency.","section":"§2.2.4, Theorem 2.16","authors_acknowledge":null,"authors_acknowledge_p":null}],"minor_comments":[{"comment":"The notation (Λ_{-∞}, d) is introduced but Λ_{-∞} is not defined in the text; it appears to refer to the limiting Cannings measure at level -∞. Please define explicitly.","section":"§1.2, Eq. (1.5)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The inequality in (2.4) uses '≥1− bδ∀ bδ >0' which is ambiguous; it should be '≥1−δ for all δ>0' or similar.","section":"§2.1.1, Eq. (2.4)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The inner product ⟨ν_u, (∇φ,2)⟩ is not defined; presumably it is the pairing of the sampling measure with the gradient of φ in the direction of aging. Please clarify.","section":"§2.1.4, Eq. (2.25)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The skeleton D is defined as Ω_N^{∞,disc} × {k2^{-n}: k,n∈ℕ}, but Ω_N^{∞,disc} is defined in (3.26) as a union over L. The product structure is not fully rigorous because the time grid is dyadic rationals while the space skeleton is a countable subgroup; the combination is countable and dense, but the notation could be made precise.","section":"§3.3.2, Eq. (3.27)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"These lemmas are stated without proof and refer to [GdHa] for the argument. Since [GdHa] is a companion paper that is presumably also under review, the referee cannot verify the claims. It would be helpful to include a brief sketch or a more detailed reference to specific sections of [GdHa].","section":"§4.1, Lemma 4.1 and Lemma 4.2","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The proof says 'This is done in Section 4, see 4.5' but Section 4.5 proves lemmas for the Lévy-web, not directly the tightness of U^L. The logical chain should be spelled out more clearly.","section":"§5.1, Proof of Theorem 2.1","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The construction of X^† via the metric d in (4.6)–(4.7) uses the exponential transform 1−exp(−δ). The choice of transform is not motivated; it would be helpful to note that this is a standard bounded metric compatible with the ultrametric.","section":"§6.5, Proof of Theorem 2.14","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The lifting of the block-resampling operator from M_loc to U is described, but the verification that the lifted operator is well defined on the domain Π (i.e., that it maps polynomials to bounded measurable functions) is not given. This is needed for the martingale problem.","section":"Appendix A.3","authors_acknowledge":null,"authors_acknowledge_p":null}],"recommendation":"major_revision"},"author_rebuttal":null,"desk_editor":{"model":"deepseek/deepseek-v4-flash","letter":"You should know two things. First, the paper does something new: a genealogy-valued Cannings process on the uncountable hierarchical group Omega_N^infty with block-resampling, an ancestral web W, a dual Levy web cW, and strong graphical duality in the ancestral space X^dagger. None of that exists in the prior literature -- [GKWb] treats countable spaces without block-resampling, [GSW16] does Brownian genealogies on Z. Second, the main theorems are conditional: they sit on top of the block-resampling extension of the countable-space theory in [GKWb]/[GKWa], both listed as in preparation.\n\nCredit where earned. The theorem statements are precise and the strategy -- L-truncation, duality, tightness, martingale problem -- is coherent and matches the measure-valued picture from the companion [GdHa]. The regime dichotomy in (2.2) is the right structural condition. The paper is honest about what it does not contain: Proposition 3.12 says plainly that [GKWb, GKWa] \"must be extended to include the mechanism of block-resampling,\" Lemma 4.4(c) carries the sentence \"It remains to prove that this holds still with block-resampling added,\" and Appendix A.2 defers the Polishness of A and A* to the same sources. Nothing here is hidden.\n\nNow the soft spots, in proportion. These are load-bearing. Proposition 3.12 is the base of the chain to Theorems 2.1, 2.7, 2.13 and 2.14, and the paper's own contribution (Appendix A.3, Section 4.5) is the lifted operators and a generator-relation check -- not the finite-L well-posedness, Feller property, or entrance laws. Lemma 4.4(c) is needed to get the entrance law that defines X^dagger and hence cW, and its proof in Section 4.5 covers duality and rebirth identification but not the flagged part. The uniqueness step in Section 5.4, preservation of the duality criterion (3.21) in the L to infinity limit, is a one-sentence claim. A few steps are waved off with \"we omit the standard details.\" None of this is an observed contradiction and the authors flag the gaps themselves, so conditional is the right verdict, not reject. But as written the pillars are not checkable inside the text.\n\nWho is this for: mathematical population genetics and interacting particle systems people working on genealogies, ancestral webs, and duality. The architecture is worth seeing. It deserves a serious referee -- this is a within-field structural advance and the proof design is credible -- but peer review should be made conditional on the authors supplying [GKWb]/[GKWa], or at least the block-resampling extension sections, so the load-bearing steps can actually be checked. I would not cite it for anything load-bearing until those companions exist. Reading group: maybe, with the caveat spelled out.","headline":"A real construction of continuum-space genealogies, an ancestral web, and a L\\'evy web with strong duality, but the load-bearing block-resampling extension is deferred to in-preparation companions; referee it, with the companions demanded.","tokens_in":58839,"tokens_out":4513,"duration_ms":40280},"pith_extraction":{"msc":["60K35"],"pacs":[],"model":"deepseek/deepseek-v4-flash","headline":"Population genealogies on the continuum hierarchical group are unique stochastic processes exactly under one divergence condition — otherwise they collapse to dust.","keywords":["evolving genealogies","continuum hierarchical group","Cannings process with block-resampling","ancestral web","Lévy web","strong graphical duality","marked ultrametric measure spaces","universality classes"],"falsifier":"The regime-A existence claim is falsifiable at the level of pair statistics: fix a concrete parameter set satisfying the divergence condition (for instance $N=2$, $c_k=2^k$, $\\lambda_k=1$ for all $k$) and a time $t>0$, and compute in the $L$-truncated models $\\Omega_2^L$ the probability that two Haar-random individuals share a most recent common ancestor within time $t$; if this probability is zero, or fails to converge as $L\\to\\infty$, then Theorem 2.1(a)(iii) is false and the limit in (2.3) cannot be the claimed $U$-valued process. Conversely, the uniqueness half of Theorem 2.7 could be settled by exhibiting two distinct solutions of the $(\\Pi, L^{\\mathrm{gen}}_{\\Omega_N^\\infty}, \\delta_{U_0})$-martingale problem for one initial state, which the duality argument claims is impossible.","tokens_in":59101,"feed_emoji":"🌳","tokens_out":10630,"duration_ms":84708,"temperature":0.7,"pith_summary":"The paper studies the genealogies — the family trees connecting present-day individuals to their common ancestors — of a spatially extended population model on the continuum hierarchical group $\\Omega_N^\\infty$, obtained as the limit of finer and finer discrete geographies. Its central claim is a dichotomy: under the divergence condition $\\sum_{k\\ge 0} (1/c_k)\\sum_{\\ell\\le k}\\lambda_\\ell = \\infty$ the genealogy of the population is a genuine stochastic process, the unique solution of a well-posed martingale problem, whereas under the complementary condition the genealogy fragments into dust in which no two individuals share an ancestor. The reason this matters is that the same condition separates stochastic from deterministic behaviour of the type-frequency process, and the paper shows the genealogical and measure-valued pictures are governed by the same dichotomy. The paper then constructs a richer object, the ancestral web, a population of forward ancestral paths, together with its backward dual, the L\\'evy web, embedded in one ancestral space with strong graphical duality, and closes by identifying universality classes for these objects under space-time rescaling.","feed_headline":"One divergence condition decides when genealogies survive","feed_subtitle":"For the continuum hierarchical Cannings population, the same sum separates a unique stochastic genealogy from dust.","key_machinery":"The argument runs on four coupled objects. (i) The state space $U^V$ of marked ultrametric measure spaces — equivalence classes of $(U,d,\\mu,\\kappa)$ where $d$ is an ultrametric (the genealogical distance, twice the time back to the most recent common ancestor) and $\\kappa$ marks each individual by a type in $K$ and a location in $\\Omega_N^\\infty$ — supplies a Polish topology in which convergence is tested through finitely sampled distances and marks. (ii) The $L$-truncation approximation replaces the continuum geography by the countable groups $\\Omega_N^L$ with Fleming-Viot resampling at the bottom level, so each truncated genealogy is a known Markov jump process on a countable Abelian group, and the limit $L\\to\\infty$ is taken through duality with a spatial coalescent rather than directly. (iii) The regime condition $\\sum_{k\\ge0}(1/c_k)\\sum_{\\ell\\le k}\\lambda_\\ell=\\infty$ is the criterion that the effective difference migration is recurrent enough for the dual coalescent to meet itself, and it is exactly the condition separating existence of a stochastic genealogy limit from dust. (iv) The L\\'evy web $\\widehat{W}$ is the jump-process analogue of the Brownian web: a population of backward paths of descent obtained as the entrance law of a block-coalescing system of L\\'evy processes started from every space-time point, and it is the engine that builds the ancestral web $W$ and the embedded ancestral space $X^\\dagger$ in which forward and backward geodesics realize $W$ and $\\widehat{W}$ simultaneously.","core_discovery":"The central claim is that the marked genealogy process $U=(U_t)_{t\\ge0}$ with values in $U^{\\Omega_N^\\infty\\times K}$ — equivalence classes of ultrametric measure spaces whose points carry type and location marks — exists as the $L\\to\\infty$ limit of $L$-truncated models on the countable discrete groups $\\Omega_N^L$, and this limit is a strong Markov Feller process with c\\`adl\\`ag paths characterized by a well-posed martingale problem, precisely in the stochastic regime A where $\\sum_{k\\ge0}(1/c_k)\\sum_{\\ell\\le k}\\lambda_\\ell=\\infty$. In regime B no such limit exists: the probability that two sampled individuals share a most recent common ancestor within time $t$ tends to zero as $L\\to\\infty$, so the genealogy is dust and the type-frequency dynamics is deterministic. At positive times in regime A the population decomposes into random depth-$h$ subfamilies covering disjoint unions of balls in the geography, and the founders form a countable set of Haar measure zero that carries all future stochastic volatility. On top of the genealogy process the paper constructs the ancestral web $W$, a forward-evolving population of marked lineages, and its dual, the L\\'evy web $\\widehat{W}$, a backward population of paths of descent generated by a spatial coalescent; both are embedded as oppositely oriented geodesics in a single marked grounded tree-like fibred metric measure space $X^\\dagger$, in which they are in strong graphical duality. A continuum-space rescaling limit then yields universality classes parametrized by stable processes, with convergence of $U$, $W$ and $X^\\dagger$ in the stochastic regime and dust formation otherwise.","pith_inferences":["If the dichotomy is as sharp as claimed, the divergence sum is a directly checkable diagnostic for whether any candidate hierarchical population model supports genealogical analysis at the continuum level, and slowly growing block-resampling rates $\\lambda_k$ relative to migration rates $c_k$ fall on the dust side.","The L\\'evy-web construction suggests the Brownian-web machinery on $\\mathbb{R}^1$ is the continuous-path special case of a jump-process web on the hierarchical group, so the strong graphical duality established here may transfer by suitable limits to spatial Lambda-Fleming-Viot models where only the Brownian web was previously available.","Because the paper explicitly leaves the $d\\to\\infty$ instantaneously-coalescing limit outside its main theorems, a natural next step is to extend the strong graphical duality to that regime, where the L\\'evy web should reduce to a web of instantaneously coalescing paths.","The claim that a Haar-null founder set carries all future volatility is testable in the truncated models: conditioning on the founder locations at a positive time should render the future of the genealogy process deterministic, a form of noise arising from a null set."],"forward_implications":["In regime A the genealogy process is determined by its initial state through a well-posed martingale problem, so all finite-sample distance-and-mark statistics of the population are in principle computable from the four generators (migration, aging, resampling, and type resampling).","In regime B no $U$-valued genealogy process exists on the continuum geography; ancestry there is carried by independent L\\'evy-moving single-ancestor lines, so genealogical analysis at the continuum level must switch to the historical-path picture.","The ancestral web embeds both the genealogy process and the measure-valued type-frequency process as functionals of one random graph, making the monotype-ball decomposition of the geography and the family decomposition of the genealogy two views of the same object.","Strong graphical duality implies the forward web and the backward L\\'evy web have the same law after orientation reversal, a pathwise rather than merely distributional duality that lets entrance-law computations run from either direction.","Under the rescaling limit, all models in the stochastic regime fall into universality classes indexed by $(\\hat c,\\hat\\lambda)$ with $\\hat c+\\hat\\lambda>0$, the stable genealogical Cannings processes, so microscopic rate sequences wash out and only the stable parameters survive."],"supporting_citations":[{"why":"Supplies the countable-space evolution of genealogies and its martingale problem that the paper must extend to block-resampling; Proposition 3.12 states that this extension is required.","marker":"[GKWb]"},{"why":"Introduces the ancestral web and the state spaces $A$ and $A^*$ on countable geographies, to which Appendix A.2 defers Polishness of the continuum state spaces.","marker":"[GKWa]"},{"why":"The companion paper that constructs the measure-valued process on $\\Omega_N^\\infty$, proves the stochastic/deterministic dichotomy, and supplies convergence of the dual L\\'evy processes and occupation times used in the $L\\to\\infty$ limits.","marker":"[GdHa]"},{"why":"Provides the construction principle that a limit dual process together with a limit duality function yields the Markov semigroup and a well-posed martingale problem, used throughout Sections 3.2 and 5.","marker":"[DGP24]"},{"why":"Defines the Polish space $U^V$ of marked metric measure spaces, the state space of the genealogy process.","marker":"[DGP11]"},{"why":"Introduced the hierarchical Cannings model with block-resampling on $\\Omega_N$ whose continuum-space version this paper extends to the level of genealogies.","marker":"[GdHKK14]"},{"why":"Supplies the continuum-space genealogy construction on $\\mathbb{R}^1$ via the Brownian web that the L\\'evy web generalizes, together with the $\\sigma$-finite localization of the marked state space.","marker":"[GSW16]"},{"why":"Provides the Liggett-Spitzer space used to control entrance laws of the spatial coalescent in Lemmas 4.1 and 4.2.","marker":"[LS81]"}],"fun_headline_variants":["A single sum decides if genealogies survive","Divergence condition gates stochastic genealogies","One criterion splits genealogies from dust","Stochastic genealogies hinge on one sum"],"cache_read_input_tokens":0,"weakest_assumption_plain":"The argument assumes that the in-preparation countable-space theory of genealogies and ancestral webs ([GKWb] and [GKWa]), extended to include block-resampling, is actually available: Proposition 3.12 states that those references 'must be extended to include the mechanism of block-resampling for the genealogies,' and Appendix A.2 defers the Polishness of the state spaces $A$ and $A^*$ to the same references, so if that extension fails, the truncation limit has no base case.","fun_headline_variants_meta":{"raw":{"variants":["A single sum decides if genealogies survive","Divergence condition gates stochastic genealogies","One criterion splits genealogies from dust","Stochastic genealogies hinge on one sum"]},"model":"deepseek/deepseek-v4-flash","effort":"low","cost_usd":0.000251,"raw_usage":{"cost":0.00025074,"is_byok":false,"cost_details":{"upstream_inference_cost":0.00025074,"upstream_inference_prompt_cost":0.00016534,"upstream_inference_completions_cost":8.54e-05},"total_tokens":1486,"prompt_tokens":1181,"completion_tokens":305,"prompt_tokens_details":{"audio_tokens":0,"video_tokens":0,"cached_tokens":0,"cache_write_tokens":0},"completion_tokens_details":{"audio_tokens":0,"image_tokens":0,"reasoning_tokens":250}},"tokens_in":1181,"tokens_out":305,"cost_source":"openrouter_usage","duration_ms":3459,"image_pages":[],"temperature":1.0,"billed_attempts":1,"reasoning_tokens":250,"transport_provider":"openrouter","cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-11-13T07:22:11.965024+00:00","model_set":{"reader":"deepseek/deepseek-v4-flash"},"pipeline_version":"pith-pipeline@v1.4.1-daily-deepseek-or-medium","input_kind":null,"input_chars":null,"input_total_chars":null,"passes_missing":null,"published_standing":null,"published_confidence":null,"objection_audit":null,"run_record":null,"basis":{"kind":"not_recorded","source":null,"chars_given":null,"chars_total":null},"receipt":null,"falsifier":"The regime-A existence claim is falsifiable at the level of pair statistics: fix a concrete parameter set satisfying the divergence condition (for instance $N=2$, $c_k=2^k$, $\\lambda_k=1$ for all $k$) and a time $t>0$, and compute in the $L$-truncated models $\\Omega_2^L$ the probability that two Haar-random individuals share a most recent common ancestor within time $t$; if this probability is zero, or fails to converge as $L\\to\\infty$, then Theorem 2.1(a)(iii) is false and the limit in (2.3) cannot be the claimed $U$-valued process. Conversely, the uniqueness half of Theorem 2.7 could be settled by exhibiting two distinct solutions of the $(\\Pi, L^{\\mathrm{gen}}_{\\Omega_N^\\infty}, \\delta_{U_0})$-martingale problem for one initial state, which the duality argument claims is impossible.","supporting_citations":[],"lane_accuracy":{"status":"not yet measured","lane":{"model":"deepseek/deepseek-v4-flash","input_kind":null},"url":"https://pith.science/accuracy"},"review_version":1}