{"id":"396f706c-9654-44ae-b816-cedba485a49a","arxiv_id":"2507.11159","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Diverse real-world networks show universal Horton-style scaling in their nested community trees, and a similarity-based connection model reproduces these exponents.","lead":"This paper reports that hierarchical communities in diverse real-world networks, from brains to road systems, follow the same statistical scaling laws, and a simple model where similar nodes connect more often reproduces those laws. A smart generalist might read it to see whether self-similar communities within communities are a universal organizing pattern across biology, infrastructure, and society.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Universal exponents may be an artifact of the forced recursive GN-bisection tree: with no null-model or alternative-algorithm comparison, the claimed universality is not yet established.","rationale":"The reader's weakest assumption identifies the same load-bearing gap: the binary tree is not validated as the true hierarchy, and no null model or algorithm robustness check is provided. I agree with that reading. The paper has genuine strengths: openly available datasets, released code, and a transparent statement that ECO and DDI are excluded because they lack structural self-similarity. But the headline claim of universal exponents depends on showing that the measured scaling is not produced by the forced-bisection procedure itself or by generic random-tree statistics. Because the recursive GN algorithm always yields a binary tree, the Horton-Strahler exponents could be dominated by the algorithm's geometry rather than by the networks' organizing principles. The concrete test above would settle this: if degree-preserving random graphs produce the same gamma_b interval, the empirical universality claim collapses; if they do not, the conditional acceptance can be upgraded. Until that test is run, the current CONDITIONAL verdict is appropriate, so I recommend UNCHANGED.","tokens_in":23544,"tokens_out":5885,"duration_ms":88311,"concrete_test":"Run the exact recursive GN-bisection plus Horton-Strahler pipeline from Materials and Methods on (i) 100 configuration-model random graphs with the same degree sequence as each of the eight real networks, and (ii) random binary trees with matched numbers of leaves and depth distributions. Compute gamma_b and gamma_eta_d; if the mean exponents and 90% intervals overlap the real-network values (gamma_b = -0.53 +/- 0.04, gamma_eta_d = -0.39 +/- 0.04), the claimed universality is not specific to real networks. Additionally, repeat the real-network analysis using a modularity-based hierarchical dendrogram (e.g., Clauset-Newman-Moore) instead of GN bisection; if gamma_b shifts by more than the reported +/- 0.04, the exponent is an artifact of the tree construction.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central empirical claim rests on interpreting the binary tree produced by recursive Girvan-Newman bisection as the true multi-scale organization of each network, then measuring Horton-Strahler scaling on that tree (Materials and Methods, 'Binary tree representation'; Table I). But this pipeline returns a binary tree for every connected network, and it imposes the very bifurcating hierarchy that is later characterized by Horton-Strahler orders. The paper provides no null-model baseline: no degree-preserving randomized graphs, no random binary trees, and no alternative hierarchical community-detection method. This matters because the authors themselves cite Ref. [68], which shows that Horton laws emerge generically in random self-similar trees; absent a comparison, the similarity of exponents across the eight chosen networks does not discriminate a universal organizing principle from a generic property of recursive bisection trees. The exclusion of ECO and DDI in Supplementary S1 strengthens this worry: those networks also obey Horton's law of branch numbers, but with exponents far outside the claimed universal range, so the claimed interval is partly defined by which networks are kept. A direct null-model test is therefore the most load-bearing missing check.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper claims that hierarchical community structure in eight diverse real-world networks obeys universal Horton-Strahler scaling laws, with exponents such as γb ≈ -0.53 ± 0.04 for log10(b_h) vs h (Fig. 2, Tables I and II). It then introduces a 'status model' in which the probability that a new node connects to an existing node is inversely proportional to the difference in node statuses (Eq. 1, referred to as Eq. 5 in the text), and reports that this model reproduces the same scaling exponents. The manuscript further argues that the underlying self-organizing principle is the minimization of within-community status diversity at every organizational scale, formalized through continuous entropy and an 'organizational entropy' E(h) (Eqs. 2 and 3), in line with Haken's principle. The paper includes supplementary analyses of additional networks, alternative status distributions, non-growing model variants, and publicly archived code.","tokens_in":23786,"tokens_out":4216,"duration_ms":55070,"significance":"If established, the claimed universality would be a notable contribution to the network-science literature, potentially linking local similarity-based attachment to the nested community geometry of biological, social, and infrastructure networks. The manuscript has clear strengths: the model is simple and transparent, the authors provide archived code, they include surrogate analyses and non-growing variants, and they explicitly discuss networks that do not fit the universal pattern. However, the central empirical claim currently lacks a null-model baseline, the exclusion of two networks from the universal range is not justified by a quantitative criterion, and the entropy-minimization principle is only demonstrated in a model where it is essentially built into the attachment rule. These issues are load-bearing for the paper's main conclusions, so the manuscript requires substantial additional work.","major_comments":[{"comment":"The hierarchical tree for every network is constructed by repeatedly removing the highest edge-betweenness edge until each component splits into exactly two (Materials and Methods). This procedure produces a binary tree for any connected graph, and the claimed universal exponents are then measured on that tree. The manuscript provides no comparison with degree-preserving randomized graphs, random binary trees, or alternative hierarchical community-detection algorithms. The authors themselves cite Ref. [68], which shows that Horton laws emerge generically in random self-similar trees, so without such a baseline the similarity of γb across the eight chosen networks does not discriminate a universal organizing principle from a generic property of the forced recursive-bisection construction. I recommend adding null-model tests (e.g., configuration-model random graphs and random binary trees with matched sizes) and at least one alternative hierarchical detection method, and reporting the resulting exponents.","section":"Materials and Methods, 'Binary tree representation'; Fig. 2(b), Table I"},{"comment":"The supplementary material shows that ECO and DDI also satisfy Horton's law of branch numbers, with γb = -0.83 and -1.03, far outside the claimed universal range. The paper excludes these networks because they have 'limited organizational scales' and a poor γηd fit, but this is a post hoc selection: the universal interval in Table I is computed after removing exactly the networks that violate it. To support the universality claim, either include ECO and DDI in the reported dispersion or in a formal outlier test, or state a quantitative, pre-specified criterion for exclusion (e.g., minimum number of organizational scales and minimum R² for γηd) and demonstrate that the main conclusions are robust to that criterion.","section":"Supplementary S1, Table S1"},{"comment":"The claim that nodes 'self-organize' so that within-community status diversity is minimized at every scale is only demonstrated in the status model, where the link probability is by construction inversely proportional to status difference (Eq. 1). The surrogate tests in Fig. 5(a) randomly permute statuses within fixed communities and show that the original assignment has lower entropy, but this is a direct consequence of the model's attachment rule: communities detected by edge-betweenness on such a network will be grouped by status. The manuscript should state this circularity explicitly and provide either a falsifiable prediction of the minimization principle that is not already encoded in the attachment rule, or a test on a real-world network with an independently measured node attribute that is not the one used to define the links.","section":"Section 'A general self-organizing principle...', Eqs. (2)-(3), Fig. 5"},{"comment":"Per-network scaling exponents are quoted to two decimals with R² values but without confidence intervals, and the 'standard error' in Table I is the dispersion across the eight networks, not the uncertainty of each individual fit. For the model, the exponents are obtained from a single simulated network (N = 1000 in Fig. 4), so finite-size and realization effects are unreported. The universality claim would be much stronger if the authors provided per-network fit uncertainties and an ensemble of model realizations (at least 10-100 networks per parameter set) with reported mean and standard deviation of each exponent.","section":"Tables I and II"}],"minor_comments":[{"comment":"The model's link probability appears as Eq. (1) in the main text but is repeatedly referred to as 'Eq. 5' (e.g., in the 'Universal scaling laws' section and in Supplementary S4). Please renumber or correct the references.","section":"Equation numbering throughout"},{"comment":"The gray surrogate lines and the reported slope β = 1.0 are described only qualitatively; the reader cannot tell how many surrogate realizations were averaged, whether the slope is the mean over realizations, and what the spread across realizations is. Please provide these details.","section":"Fig. 5(a) caption and text"},{"comment":"The statement that the large block structures in the ECO and DDI adjacency matrices 'do not represent communities but are just remnants of the community detection algorithm' is asserted without quantification. A quantitative measure of community strength, such as modularity or a comparison to a random-graph baseline, would support this claim.","section":"Supplementary S1, ECO/DDI discussion"},{"comment":"The term 'organizational entropy' E(h) is defined as a mean surplus entropy relative to a fully randomized status assignment, but the subsequent statement that E(h) 'represents the information needed to describe the orderliness of the structure' is not justified by the definition and should be reworded or supported by a derivation.","section":"Eq. (3) and surrounding text"}],"recommendation":"major_revision","confidential_remarks":"The paper has a promising core, but the missing null-model baseline and the post hoc exclusion of ECO and DDI are the main barriers to accepting the universality claim. The entropy-minimization principle also needs to be presented with its built-in circularity made explicit and with a genuinely falsifiable test. If these controls are added, the paper could become a solid contribution; in its current form, the claims are considerably stronger than the evidence."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this paper gives us a careful measurement of Horton-Strahler exponents on recursively bisected community trees across eight real-world networks, plus a simple similarity-based attachment model that reproduces them. The empirical core is worth a look, but the headline universality claim is not yet established because the pipeline always produces a binary tree and no null-model or alternative-detection check is reported.\n\nWhat's actually new and good: the survey itself. Eight diverse networks (brain, protein, gene, collaboration, animal, social, infrastructure) showing tightly clustered scaling exponents is a real observation, and the authors also go beyond Horton's law of branch numbers to check fixed-depth scaling, which is a harder test of structural self-similarity. The code and data are openly available, and the supplement is candid: they test different status distributions, varying link density, non-growing versions of the model, and they report the ECO and DDI networks that do NOT fit the universal range. That transparency earns credit. The Horton exponents from real networks are genuinely independent empirical outputs.\n\nNow the soft spots. The biggest is the missing null model. The authors cite Ref. [68] showing that random self-similar trees obey Horton laws, and they cite it approvingly, yet they never compare their trees to randomized graphs, random binary trees, or even a degree-preserving shuffle. Without that, it is entirely possible that gamma_b ≈ -0.53 is a generic artifact of recursive Girvan-Newman bisection rather than a signature of an organizing principle. The post hoc exclusion of ECO and DDI deepens this worry: those networks obey Horton's law of branch numbers too, just with different exponents, so the claimed universal interval is partly defined by which networks are kept.\n\nSecond, the \"self-organizing principle\" is close to circular. The model's link rule says similar-status nodes connect with higher probability; the surrogate test shows that communities then have reduced status entropy. That is what the construction guarantees. Calling it a discovered principle is a stretch, though the authors might say they are merely quantifying what follows from the rule. The Horton exponents are not circular, so the core empirical result survives this critique.\n\nMinor point: model exponents come with R-squared but no uncertainty across repeated runs. That is fixable with a few simulations.\n\nWho is the paper for? People working on hierarchical modularity, community detection baselines, and fractal network structure. It is a phenomenological description, not a proof of mechanism. I would not cite it for the universality claim until the null-model gap is closed, but I might cite it as a cautionary case study in how detection algorithms shape apparent scaling.\n\nRecommendation: yes, send it to peer review. The data and code are there, the fits are transparent, and the missing checks are exactly what a competent referee can demand. Conditional acceptance with a required null-model and alternative-algorithm comparison would be appropriate.","headline":"A clean, well-documented empirical survey of Horton scaling in hierarchical communities, but the universality claim needs a null model and alternative algorithm before it lands.","tokens_in":24297,"tokens_out":2538,"would_cite":false,"duration_ms":33528,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that nested communities in brains, protein networks, collaborations, and road grids obey one scaling law, reproduced by a similarity-based linking rule.","keywords":["hierarchical community structure","universal scaling laws","Horton-Strahler ordering","self-similarity","self-organization","community detection","status-based link formation","complex networks"],"falsifier":"Run the identical pipeline — iterated Girvan–Newman bisection to a binary tree, then Horton–Strahler counting — on degree-preserving randomized copies of the same eight networks and on random binary trees such as critical Galton–Watson trees; if those random structures also give a branch-count exponent near $-0.5$, the universality is a generic property of forced-bisection trees rather than of the networks. Equally decisive would be recomputing the exponents from an alternative hierarchical community-detection method on the same eight networks and finding that they deviate from $-0.53 \\pm 0.04$.","tokens_in":23373,"feed_emoji":"🕸️","tokens_out":14069,"duration_ms":147028,"temperature":0.7,"pith_summary":"The paper claims that the nested 'communities-within-communities' structure found in diverse real-world networks — mouse brain fibre tracts, E. coli protein interactions, C. elegans gene interactions, scientific collaborations, weaver-bird interactions, mutually liked Facebook pages, and European road infrastructure — follows the same quantitative scaling laws, with the number of branches per organizational level decaying as $\\log_{10} b_h = \\gamma_b h + c$ with $\\gamma_b \\approx -0.53 \\pm 0.04$. It then shows that a minimal generative model, in which a new node links to an existing node with probability inversely proportional to the difference in their intrinsic 'status', reproduces this entire family of exponents whether the status distribution is Gaussian, exponential, or quadratic. The paper interprets the match as evidence for a general self-organizing principle: nodes sort into communities so that the diversity (continuous entropy) of statuses within each community is minimized at every organizational scale, and organizational entropy falls as the scale coarsens, consistent with Haken's principle. If correct, the discovery means one similarity-based linking rule can account for a quantitative structural signature shared by brains, cells, societies, and infrastructure.","feed_headline":"One scaling law governs nested communities across many networks","feed_subtitle":"The same ~−0.53 scaling for nested communities holds from brains to road grids.","key_machinery":"The carrying object is the binary community tree with Horton–Strahler ordering. Communities are detected by iterated Girvan–Newman edge-betweenness removal: each community is split into exactly two children, repeatedly, until single nodes remain, and each community-node receives an order $h$ (leaf communities get $h = 1$; a parent with two children of equal order $h$ gets $h + 1$, otherwise it inherits the larger child order). Horton's law — branch counts decaying geometrically with order, $\\log_{10} b_h = \\gamma_b h + c$ — is the quantitative signature of self-similarity, and the paper's generative mechanism is the status model, in which the connection probability between an incoming node $i$ and an existing node $j$ is $\\pi_{ij} = (1/|S_i - S_j|) / \\sum_k (1/|S_i - S_k|)$, i.e., inverse to status difference. The surrogate-entropy comparison is the device that links the emergent tree to the claimed self-organizing principle of entropy minimization at every scale.","core_discovery":"The central claim is that the binary tree of nested communities obtained by repeatedly bisecting a network is structurally self-similar and obeys universal Horton scaling. Across eight real-world networks, the number of branches $b_h$ of organizational order $h$ satisfies $\\log_{10} b_h = \\gamma_b h + c$ with $\\gamma_b \\approx -0.53 \\pm 0.04$ (bifurcation ratio $R_b \\approx 3.38$), and companion exponents for community count, mean relative link density, mean community size, and fixed-depth attributes ($\\gamma_\\chi$, $\\gamma_\\eta$, $\\gamma_n$, $\\gamma_h$, $\\gamma_{\\chi d}$, $\\gamma_{\\eta d}$, $\\gamma_{nd}$) cluster tightly across the same networks. Because the scaling also holds at fixed hierarchical depth, the paper asserts the stronger property of structural self-similarity, and it shows that a tree that merely obeys Horton's law but is not self-similar fails those fixed-depth tests (as the ECO and DDI networks do). The same exponent family emerges from a model in which the probability of linking two nodes is proportional to the inverse of the difference in their assigned statuses, and surrogate shuffles show that the resulting communities minimize the continuous entropy of statuses within each community at every scale, with organizational entropy decreasing as the scale of description grows.","pith_inferences":["Because the tree is produced by repeatedly bisecting the network until each community splits into exactly two children, the universal exponents may partly reflect properties of the Girvan–Newman bisection procedure itself; rebuilding the trees with other hierarchical detection methods, or from random binary trees, would settle this, and the paper does not report such a comparison.","If the scaling survives a change of tree-construction method, the exponents join degree exponents and fractal dimensions as candidate universal descriptors, and one could ask whether the status model's exponents are fixed-point values of the linking rule or drift continuously with the shape of the status distribution $p(S)$.","The cost reading sketched in the discussion — maintaining links between dissimilar nodes costs more — suggests a testable extension: networks whose link cost grows with status difference should show sharper Horton scaling as the cost grows, and blur toward random-tree behaviour as the cost vanishes.","The entropy-minimization and organizational-entropy results are demonstrated on model networks; extending the same surrogate analysis to the eight real-world networks would test whether the minimization claim holds where only the exponents have been measured."],"forward_implications":["The Horton branch-count exponent $\\gamma_b \\approx -0.53$ (bifurcation ratio $\\approx 3.38$) becomes a quantitative fingerprint that can be measured once a network's hierarchical communities are mapped to a binary tree, allowing direct comparison of brains, protein webs, social systems, and infrastructure.","Networks grown from similarity-based linking reproduce the universal exponents without degree-based preferential attachment, so the status model offers a generating mechanism for synthetic networks with prescribed self-similar hierarchical structure.","The surrogate analysis implies that the observed community structure is the most entropy-minimizing arrangement of node properties at every scale: randomizing statuses within communities always raises the mean within-community status entropy.","Organizational entropy $E(h)$ decreases with organizational scale, which the paper presents as a concrete instance of Haken's principle that order formation lowers the system's remaining degrees of freedom.","Not every network obeys the laws: the ECO and DDI networks are Hortonian but fail the structural self-similarity tests, so the universal scaling is a distinguishing property rather than a tautology of any tree built from any network."],"supporting_citations":[{"why":"Horton's law of branch numbers, the scaling relation the paper verifies on community trees.","marker":"[63]"},{"why":"Strahler's ordering scheme, the indexing rule (Eq. 4) that assigns each community its organizational order h.","marker":"[67]"},{"why":"Girvan–Newman edge-betweenness community detection, the algorithm whose repeated application builds the binary community tree.","marker":"[66]"},{"why":"Guimera et al.'s tree representation and self-similarity analysis of nested communities, the direct antecedent of the Horton–Strahler community-tree approach.","marker":"[40]"},{"why":"Scheidegger's caveat that Horton scaling alone does not imply self-similarity, which motivates the paper's fixed-depth structural self-similarity checks.","marker":"[64]"},{"why":"Kovchegov, Zaliapin, and Foufoula-Georgiou's scaling relations for mean attributes of random self-similar trees, the basis for the angle-bracket attribute laws.","marker":"[68]"},{"why":"Haken's principle that entropy decreases when order emerges, the general self-organizing principle the paper claims underlies the scaling.","marker":"[65]"},{"why":"Boguna et al.'s social-distance attachment model, the closest prior similarity-based link formation mechanism that the status model generalizes.","marker":"[59]"}],"fun_headline_variants":["One scaling law for nested communities across networks","Universal self-similarity found in hierarchical networks","Networks self-organize into scale-free community trees","Horton's law governs nested communities in real networks","Same −0.53 exponent shapes brain, road, and social networks"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the binary tree created by repeatedly splitting every community into exactly two sub-communities with Girvan–Newman edge-betweenness removal represents the network's true multi-scale organization; if a different detection method builds a different tree, the universal exponents could be an artifact of the forced bisection procedure.","fun_headline_variants_meta":{"raw":{"variants":["One scaling law for nested communities across networks","Universal self-similarity found in hierarchical networks","Networks self-organize into scale-free community trees","Horton's law governs nested communities in real networks","Same −0.53 exponent shapes brain, road, and social networks"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000659,"raw_usage":{"total_tokens":3027,"prompt_tokens":970,"completion_tokens":2057,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":586,"completion_tokens_details":{"reasoning_tokens":1980}},"tokens_in":586,"tokens_out":2057,"duration_ms":19750,"temperature":1.0,"reasoning_tokens":1980,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T17:15:39.713621+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the identical pipeline — iterated Girvan–Newman bisection to a binary tree, then Horton–Strahler counting — on degree-preserving randomized copies of the same eight networks and on random binary trees such as critical Galton–Watson trees; if those random structures also give a branch-count exponent near $-0.5$, the universality is a generic property of forced-bisection trees rather than of the networks. Equally decisive would be recomputing the exponents from an alternative hierarchical community-detection method on the same eight networks and finding that they deviate from $-0.53 \\pm 0.04$.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Horton's law of branch numbers, the scaling relation the paper verifies on community trees."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Strahler's ordering scheme, the indexing rule (Eq. 4) that assigns each community its organizational order h."},{"cited_title":"Girvan, M","cited_arxiv_id":null,"evidence_quote":"Girvan–Newman edge-betweenness community detection, the algorithm whose repeated application builds the binary community tree."},{"cited_title":"Guimera, L","cited_arxiv_id":null,"evidence_quote":"Guimera et al.'s tree representation and self-similarity analysis of nested communities, the direct antecedent of the Horton–Strahler community-tree approach."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Scheidegger's caveat that Horton scaling alone does not imply self-similarity, which motivates the paper's fixed-depth structural self-similarity checks."},{"cited_title":"Kovchegov, I","cited_arxiv_id":null,"evidence_quote":"Kovchegov, Zaliapin, and Foufoula-Georgiou's scaling relations for mean attributes of random self-similar trees, the basis for the angle-bracket attribute laws."},{"cited_title":"Haken, Advanced synergetics: Instability hierarchies of self-organizing systems and devices, vol","cited_arxiv_id":null,"evidence_quote":"Haken's principle that entropy decreases when order emerges, the general self-organizing principle the paper claims underlies the scaling."},{"cited_title":"Bogun ´a, R","cited_arxiv_id":null,"evidence_quote":"Boguna et al.'s social-distance attachment model, the closest prior similarity-based link formation mechanism that the status model generalizes."}],"review_version":1}