{"id":"874a0297-abb3-4a6f-bb1c-6bb04eca090b","arxiv_id":"2504.12955","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Constrained rewiring of supplier-customer links cuts modeled systemic risk by 16 to 50 percent across six firm-level supply chain subnetworks without changing firms' production functions.","lead":"Systemic risk in firm-level supply chain networks can be lowered by 16 to 50 percent through rewired supplier-customer links while keeping firms' production nearly unchanged. The result suggests that real supply networks carry extra risk that targeted policy incentives could remove.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The claim 'without reducing production output' is not established: the rewiring algorithm permits ±20% out-strength deviations, and no post-hoc check rules out output-driven risk reduction.","rationale":"The reader identified the NACE 3-digit substitutability assumption as the weakest point. That is a serious concern, and the paper candidly acknowledges it. However, the output-tolerance issue is more directly load-bearing for the headline claim: the abstract promises mitigation 'without reducing the production output of firms,' yet the algorithm permits up to 20% output deviation and the final output changes are never reported. This is an internal inconsistency with the central claim, not merely an external modeling choice, and it is checkable from the authors' own simulation outputs. If the 16-50% reductions vanish under exact output preservation, the main policy conclusion collapses. The NACE proxy, by contrast, affects the feasibility set and could bias in either direction, and the authors argue their restrictions may underestimate the reduction. Both concerns warrant conditions, but the output-preservation check is the one that should be run first. Since the reader's verdict was already CONDITIONAL, my analysis reinforces that verdict rather than changing it.","tokens_in":28982,"tokens_out":4724,"duration_ms":50829,"concrete_test":"Re-run all six rewiring experiments with the out-strength constraint tightened to exactly 0% deviation (e.g., weight-conserving full swaps that preserve every node's out-strength exactly), and compare the final average-ESRI reductions against Table I. Also report the distribution of relative out-strength changes in the current final configurations. If the 16-50% reductions persist under exact output preservation, the headline claim stands; if they shrink or vanish, the reported mitigation is partly or wholly an artifact of permitted output reduction.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim is that rewiring reduces systemic risk by 16-50% without reducing firms' production output. The optimization, however, explicitly allows each firm's out-strength to deviate by up to 20% from its empirical value (main text 'Constraints for the rewiring algorithm'; SI S3.A). The ESRI calibration equates a firm's production output with its out-strength (SI S2: 'the production output that each firm i is able to sustain ... is exactly equal to the summed volume of its sales transactions, the node out-strength'). Thus the algorithm permits output reductions of up to 20%, and the paper never reports the realized out-strength changes in the final risk-mitigated configurations. Because ESRI uses out-strength both as the loss weight and in the denominator, a configuration that shifts or reduces measured output can lower average ESRI mechanically, without any genuine topological improvement. The beta = 0 control does not resolve this: it also operates under the same 20% tolerance, so it only shows that unbiased rewiring does not reduce ESRI, not that the biased reductions are output-neutral. If the final configurations contain firms at 80% output, the abstract's 'without reducing production output' is false, and the reported mitigation may be an artifact of the allowed output slack.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper asks whether systemic risk in supply chain networks can be substantially reduced by rewiring supplier-customer links while preserving firms' production outputs. Using the ESRI measure of firm-level systemic risk, the authors apply a Metropolis-Hastings simulated-annealing algorithm to rewire links within six subnetworks of Ecuadorian and Hungarian supply chains, constrained to preserve firms' in- and out-degrees per industry sector and to keep out-strength within 80-120% of its empirical value. They report average ESRI reductions of 16-50%, attribute the reductions to meso-scale topological features, and argue that real supply chain networks carry unnecessarily high systemic risk. A beta=0 control shows that unbiased random rewiring does not reduce average ESRI, supporting the role of the risk bias.","tokens_in":29185,"tokens_out":5308,"duration_ms":53633,"significance":"If the result holds, the paper provides a novel and important demonstration that supply chain topology alone has a large effect on systemic risk and that network rewiring, which occurs naturally at high rates, could be steered by policy. The optimization methodology is coherent, and the beta=0 control is a good check that the reduction is not simply an artifact of any random rewiring. The paper also makes a useful step by applying the approach to both weighted and unweighted networks and to two countries. However, the central claim of output neutrality is currently not empirically verified, and the feasibility of the rewiring is grounded in a coarse industry classification. If these issues are resolved, the paper would be a valuable contribution to the literature on supply chain resilience and systemic risk.","major_comments":[{"comment":"The abstract's claim 'without reducing the production output of firms' is not established by the reported analysis. The algorithm explicitly allows out-strength (identified with production output in SI S2) to deviate by up to 20%, and the manuscript never reports the realized out-strength changes in the risk-mitigated configurations. Because average ESRI is a production-weighted loss, a configuration that reduces the out-strength of high-ESRI firms can lower average ESRI mechanically, independently of topology. The beta=0 control does not resolve this because it also operates under the same 20% tolerance; it only shows that unbiased rewiring does not reduce ESRI. Please report the actual out-strength distributions after rewiring and, ideally, recompute ESRI with out-strengths fixed at their empirical values to verify that the reduction is topological.","section":"Constraints for the rewiring algorithm and SI S2, S3.A"},{"comment":"The rewiring space is defined by NACE 3-digit substitutability: firms in the same NACE group are assumed to be freely interchangeable suppliers or customers. The authors correctly acknowledge the limitation, noting the error could go in either direction. However, the central policy conclusion that real topologies 'carry unnecessarily high levels of systemic risk' rests on this assumption. Without a sensitivity analysis at a coarser or finer industry classification, or a more detailed justification of within-NACE substitutability, the magnitude of the mitigation (16-50%) cannot be interpreted as an achievable reduction in a real economy. Please qualify the abstract's claim to 'under the NACE proxy' or provide additional validation.","section":"SI S3.A and Discussion (NACE proxy limitation)"},{"comment":"Each beta value and the simulated annealing run are reported as a single trajectory. Since the algorithm is stochastic and the authors themselves note the risk of getting trapped in local minima (e.g., beta=3200 in Fig. 3a), the reported reductions may depend on initialization and random seed. Please report results across multiple independent runs and state the number of seeds and dispersion (e.g., min/max or interquartile range) for the final average ESRI values. This is important for interpreting the 16-50% range as a robust property of the rewiring process.","section":"Figure 3, Figure S4, and Table S2"}],"minor_comments":[{"comment":"The statement that '10,000 steps roughly correspond to one update for each supply link, equivalent to less than two years of real-world rewiring rates' assumes the algorithm's swap moves are comparable to observed annual link turnover. Please clarify how the two quantities are mapped.","section":"Results (rewiring steps interpretation)"},{"comment":"The paper calls the beta=0 rewiring a 'configuration model', which is not the standard configuration model (which randomizes edges preserving the degree sequence). Here the rewiring preserves degrees per NACE sector pair and avoids multi-edges in unweighted networks; consider using a term like 'random link-swap model' to avoid confusion.","section":"Table I and main text"},{"comment":"The abstract reports '16-50%' but the exact reductions in Table I are 16.1%, 33.3%, 42.4%, 43.7%, 50.0%, and 18.3%. Please state the range consistently, e.g., '16-50%' with a footnote or table reference.","section":"Abstract and Results"},{"comment":"The Introduction describes ESRI as 'the fraction of the total production of the economy affected by the failure of that firm', while the Methods defines it as the fraction of total production lost. Please align the wording to avoid ambiguity about whether the contribution is a conditional loss or an effect measure.","section":"Introduction and Materials and Methods (ESRI)"},{"comment":"The caption says 'the crustaceans and soft drinks subnetworks are highlighted by red and green nodes'. Given the complexity of the network plot, consider a color-blind-accessible palette or a separate panel with the subnetworks isolated for visibility.","section":"Figure 1 caption"}],"recommendation":"major_revision","confidential_remarks":"The paper fits the journal's scope and the optimization approach is well designed. The most pressing issue for the editor is that the headline claim of 'without reducing production output' is not supported until the authors document the realized out-strength changes in the final configurations. The NACE proxy limitation is acknowledged but deserves a similar degree of scrutiny. I recommend major revision with a request for the output-neutrality check and robustness of the optimization runs."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this paper applies a simulated-annealing rewiring algorithm to six real firm-level supply-chain subnetworks from Ecuador and Hungary, and reports 16-50% reductions in average ESRI. The optimization scaffolding is solid, and the beta=0 control is genuinely informative: random rewiring under the same constraints leaves average risk at the empirical level. But the abstract's central claim, that this happens 'without reducing the production output of firms,' is not supported by the paper's own constraints. The algorithm explicitly allows each firm's out-strength to deviate by up to 20% from its empirical value, and the ESRI model equates out-strength with production output. The authors never report the realized out-strength changes in the final mitigated networks, so the risk reduction may be partly mechanical: shrinking measured output can lower ESRI without any real topological improvement. The beta=0 control does not resolve this, because it also operates under the same 20% tolerance; it only shows that unguided rewiring does not reduce ESRI, not that the guided reductions are output-neutral.\n\nWhat is new and worth crediting: the application to supply chains with explicit production constraints is a real step beyond the earlier interbank rewiring papers, and the data are real nationwide VAT networks, even if the subnetworks are deliberately selected to be rewirable. The paper also does a thorough job showing that standard network metrics (clustering, diameter, reciprocity) do not explain the risk reduction, which is a useful negative result pointing to meso-scale structure.\n\nWhere it is soft: the output-neutrality gap is the main issue. The NACE 3-digit product proxy is a close second: the entire feasible rewiring set depends on the assumption that firms in the same sector can substitute for each other, which the authors acknowledge is a simplification, and the error could cut either way. I would also like to see error bars or multiple runs; only one trajectory per beta is shown, and no code or data are public.\n\nIf the authors either enforce strict output preservation or report the realized out-strength distributions and show that the mitigation holds for near-neutral configurations, I would buy the result. As it stands, the 16-50% numbers are conditional, not established.\n\nI would send this to a serious referee, but I would make output-neutrality verification a condition of acceptance. It is a paper worth arguing with, not a desk reject.","headline":"A clever optimization study whose headline claim about output-neutral risk reduction is undermined by the unenforced 20% output tolerance.","tokens_in":29735,"tokens_out":3392,"would_cite":false,"duration_ms":33502,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Rewiring supplier–customer links in real supply networks can cut systemic risk by 16–50 percent while leaving firm production intact.","keywords":["supply chain networks","systemic risk","ESRI","network rewiring","risk mitigation","simulated annealing","firm-level production networks","NACE classification"],"falsifier":"Re-run the rewiring on a supply-chain dataset with actual product-level transaction records, replacing NACE 3-digit codes with the real products exchanged; if the 16–50% risk reduction shrinks or vanishes, the sector-substitutability proxy is the source of the result. A cheaper check is to repeat the analysis using NACE 4-digit or product-category constraints and compare the achievable reduction.","tokens_in":28747,"feed_emoji":"🔗","tokens_out":7857,"duration_ms":69796,"temperature":0.7,"pith_summary":"This paper asks whether the systemic risk carried by real supply chain networks can be reduced by rewiring supplier–customer links instead of by holding inventories or adding redundant suppliers. The authors claim it can: a constrained Monte Carlo link-swapping algorithm applied to six national subnetworks lowers the average Economic Systemic Risk Index by 16–50% while preserving each firm's input quantities per product and keeping output close to its empirical level. If the claim is right, the topology of observed supply chains is not forced by production technology, and a meaningful share of systemic risk could be removed through targeted rewiring incentives rather than costly redundancy.","feed_headline":"Rewiring supply links can cut systemic risk by 16–50%","feed_subtitle":"Constrained link swaps preserve production output while removing avoidable risk from real supply networks.","key_machinery":"The key machinery is the Economic Systemic Risk Index (ESRI) combined with a constrained link-swapping Metropolis-Hastings algorithm. ESRI assigns each firm a value equal to the total production lost when that firm fails, using generalized Leontief production functions with essential and non-essential inputs; the network-level risk is the average $\\langle \\mathrm{ESRI} \\rangle$ over all firms. The rewiring step swaps the endpoints of two directed links that share the same ordered pair of NACE 3-digit activities, preserving the per-sector input and output structure; weighted links are swapped when similar or split when very different, and each firm's out-strength is kept within 80–120% of its empirical value. Simulated annealing raises the acceptance temperature parameter $\\beta$ so the system accepts higher-risk moves less often, driving the network toward configurations with lower average risk while escaping local minima.","core_discovery":"The paper's central claim is that the topology of real supply chain networks carries a large, removable component of systemic risk. Using the average Economic Systemic Risk Index (ESRI)—the expected fraction of economy-wide output lost when a firm fails—as the objective, the authors show that a Metropolis-Hastings link-swapping algorithm, run under constraints that preserve each firm's input quantities per product and keep output within 20% of its empirical value, lowers average ESRI by 16–50% in six national subnetworks (Ecuador's crustacean and soft-drinks chains, weighted and unweighted, and Hungary's food and automotive chains, unweighted). Unbiased rewiring leaves risk close to the empirical level, while risk-biased rewiring converges to lower plateaus; the empirical networks therefore resemble random configurations rather than risk-minimizing ones. The authors read this as evidence that observed supply-chain topologies are suboptimal with respect to systemic risk and that the margin could be captured through market-based incentives.","pith_inferences":["If the NACE substitutability premise holds, comparable mitigation margins should appear in other national supply networks; that is testable wherever firm-level transaction data exist.","The sector-pair swap constraint is a double-edged proxy: real products may be specialized within a NACE class (over-estimating feasible swaps) or may cross sector boundaries (under-estimating them), so the 16–50% range is an estimate, not a fixed bound.","The paper's own observation that a small set of swaps likely produces most of the reduction suggests the rewiring result could be turned into a targeted intervention: identify the few critical links and adjust incentives around them.","The similar reductions found in unweighted networks imply that countries with only presence/absence transaction data can still pursue topology-based mitigation without waiting for transaction volumes."],"forward_implications":["Real supply networks sit near the risk level of unbiased random rewiring, so the systemic risk they carry is not a fundamental property of production alone; it can be lowered by choosing different links.","Risk can be cut by 16–50% across six subnetworks while keeping each firm's input mix per product fixed and its total output within 20% of the empirical value.","The reduction is concentrated among the riskiest firms: the top of the ESRI profile shrinks by half or more, with only a few firms becoming riskier.","Standard network metrics—degree, clustering, reciprocity, diameter, and component sizes—do not explain the improvement, pointing to meso-scale structures such as a 'systemic risk core'.","Because observed supply-link turnover is already high, modest market-based incentives could push networks toward lower-risk configurations without forcing firms to change technology or output."],"supporting_citations":[{"why":"defines the ESRI risk measure and the cascade-propagation model that the whole analysis optimizes","marker":"[21]"},{"why":"documents the high empirical turnover of supply links that makes rewiring feasible in practice","marker":"[27]"},{"why":"shows how targeted incentives such as a systemic-risk transaction tax reduce risk in financial networks, the policy template transferred here","marker":"[28]"},{"why":"provides the earlier network-optimization benchmark showing about 70% systemic-risk reduction in interbank exposure networks","marker":"[30]"},{"why":"supplies empirical evidence on firm-level production network structure, including the over-expression of reciprocity","marker":"[17]"},{"why":"demonstrates that firm-level network models capture propagation effects that sector-level aggregation misses, motivating the subnetwork analysis","marker":"[20]"},{"why":"defines the NACE 3-digit classification used as the product proxy in the rewiring constraints and ESRI computation","marker":"[32]"}],"fun_headline_variants":["Rewiring supply links cuts systemic risk 16–50% without output loss","Real supply networks carry avoidable systemic risk","Link swaps trim supply chain risk by up to half","Systemic risk drops 16–50% with network rewiring","Supply chain rewiring reduces risk without cutting output"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The result rests on the assumption that firms with the same NACE 3-digit code are interchangeable suppliers and customers for a given sector pair, and that the ESRI model's essentiality and substitutability parameters faithfully describe how failures propagate; if real products within a sector are specialized, the feasible rewiring set and the computed risk reductions are partly artifacts.","fun_headline_variants_meta":{"raw":{"variants":["Rewiring supply links cuts systemic risk 16–50% without output loss","Real supply networks carry avoidable systemic risk","Link swaps trim supply chain risk by up to half","Systemic risk drops 16–50% with network rewiring","Supply chain rewiring reduces risk without cutting output"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000818,"raw_usage":{"total_tokens":3594,"prompt_tokens":971,"completion_tokens":2623,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":587,"completion_tokens_details":{"reasoning_tokens":2542}},"tokens_in":587,"tokens_out":2623,"duration_ms":16434,"temperature":1.0,"reasoning_tokens":2542,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T12:18:21.035640+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-run the rewiring on a supply-chain dataset with actual product-level transaction records, replacing NACE 3-digit codes with the real products exchanged; if the 16–50% risk reduction shrinks or vanishes, the sector-substitutability proxy is the source of the result. A cheaper check is to repeat the analysis using NACE 4-digit or product-category constraints and compare the achievable reduction.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"provides the earlier network-optimization benchmark showing about 70% systemic-risk reduction in interbank exposure networks"},{"cited_title":"Lafond, P","cited_arxiv_id":null,"evidence_quote":"supplies empirical evidence on firm-level production network structure, including the over-expression of reciprocity"},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"demonstrates that firm-level network models capture propagation effects that sector-level aggregation misses, motivating the subnetwork analysis"},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"defines the NACE 3-digit classification used as the product proxy in the rewiring constraints and ESRI computation"}],"review_version":1}