{"id":"99d9afec-5b71-424e-a5fa-4c06a1bbe172","arxiv_id":"2509.16817","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A quantum network protocol stack with a Global Entanglement Module, where simulations show a scoring-based adaptive strategy improves entanglement generation rates by about 20% over fixed-tree baselines.","lead":"This paper designs a unified six-layer software architecture, a protocol stack, for future quantum internet networks. It introduces a Global Entanglement Module that keeps nodes informed about available entangled connections, and simulations show adaptive swapping guided by this module improves entanglement generation rates by about 20%.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Scoring's ~20% gain over fixed-tree may come from the separate Early Discarding mechanism, not from GEM-based adaptive execution; the fixed-tree baseline is not described as using the same discard policy.","rationale":"The reader's weakest assumption concerns GEM broadcast scalability and eventual consistency. That is a legitimate limitation, but the manuscript explicitly scopes GEM to near-term, modest-scale networks (e.g., SCY-QNet) and defers hierarchical synchronization to future work, so it is partly self-acknowledged rather than a hidden flaw in the central performance claim. The more load-bearing issue is that the paper's headline quantitative result may be an artifact of an asymmetric comparison. The adaptive strategies include the Early Discarding Strategy, while the fixed-tree baseline is not described as including it. Because proactive discarding is a concrete mechanism that can improve rate independently of global coordination, the reported ~20% improvement could be due to this mechanism rather than to GEM's network-wide view or the Scoring policy. This is a testable experimental confound, and it directly determines whether the strongest claim in the abstract holds. A second supporting issue is the complete omission of values for the scoring weights and cutoff function, which prevents reproduction and leaves room for overfitting; however, the baseline confound is the more fundamental concern. The verdict should remain CONDITIONAL rather than move to REJECT, because the architecture and qualitative findings may survive a corrected comparison, but the numerical headline should be re-evaluated under a fair baseline.","tokens_in":21448,"tokens_out":5162,"duration_ms":51232,"concrete_test":"Re-run the Fixed-Swapping-Tree baseline with the identical Early Discarding strategy used by the adaptive policies — same T_cutoff(d), same depth-dependent f(d), same age bookkeeping and memory handling — and re-measure generation rates under the default 100-node Waxman setting. If fixed-tree + early discarding closes the rate gap to within simulation noise, the headline 20% improvement should be attributed to the discard heuristic rather than to GEM-based adaptive execution.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central quantitative claim in §VI is that the Scoring strategy improves entanglement generation rate by ~20% over a globally optimal fixed-tree baseline. But the adaptive strategies are 'each applied in conjunction with the Early Discarding Strategy' (§VI, Our Algorithms), whereas the Fixed-Swapping-Tree baseline is described only as the DP-optimal tree from [11], with no indication that it receives the same depth-dependent cutoff T_cutoff(d) = f(d)·L_target defined in §IV-E. Early discarding removes aged intermediate EPs before they can be consumed by swaps; this can independently improve throughput by avoiding wasted swaps and freeing memory, independent of any GEM-based global coordination. If the fixed-tree baseline does not include this mechanism, the reported 20% improvement conflates the adaptive swapping policy with an auxiliary aging heuristic, so the claim that GEM/scoring-based adaptive execution delivers the improvement is not established by the experiments as described. This is reinforced by the fact that the scoring weights α, γ, δ, β1, β2 in Algorithm 2 and the cutoff function f(d) are never reported, leaving the strategy underspecified and potentially tuned to the evaluated baselines.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a six-layer quantum network protocol stack with a cross-layer Global Entanglement Module (GEM) that maintains a distributed, eventually consistent view of active entanglement metadata. The stack separates offline planning from real-time execution, and the Swapping/Fusion Layer adapts to the stochastic availability of EPs using a set of lightweight policies (youngest, oldest, longest-hop, shortest-hop, and score-based), together with a proactive early-discarding rule. The main quantitative claim is that the score-based strategy improves entanglement generation rates by about 20% over a globally optimal but non-adaptive fixed-tree baseline, and more than doubles the rate of a connectionless hop-by-hop baseline. The paper also describes support for predistributed entanglement, purification, and multipartite states, and evaluates the framework in NetSquid on Waxman random networks of 50–200 nodes.","tokens_in":21839,"tokens_out":4843,"duration_ms":48469,"significance":"If the main quantitative claim is substantiated, the architectural contribution is meaningful: separating planning from execution and maintaining a network-wide entanglement view through GEM is a useful step toward practical quantum network stacks, and the modular support for predistribution, purification, and multipartite states goes beyond many prior proposals. The paper is also explicit about the near-term scope, citing modest network sizes and eventual consistency via broadcast. The NetSquid evaluation is a genuine attempt to compare multiple adaptive policies against non-adaptive and connectionless baselines. However, the load-bearing quantitative comparison currently has a confounding baseline, undisclosed algorithm parameters, and no statistical support, so the central 20% claim is not yet established as reported.","major_comments":[{"comment":"The text states that all five adaptive strategies are 'each applied in conjunction with the Early Discarding Strategy' (§VI, Our Algorithms), while the Fixed-Swapping-Tree baseline is described only as the DP-optimal tree from [11] with no indication that it also uses the depth-dependent cutoff T_cutoff(d)=f(d)·L_target from §IV.E. Early discarding can independently increase throughput by freeing memory and avoiding swaps with aged EPs. The reported ~20% gain of Scoring over Fixed-tree therefore conflates the adaptive swapping policy with the early-discarding heuristic. The fixed-tree baseline must be run with the same early-discard policy, or an ablation isolating early discarding from scoring must be provided.","section":"§VI, Our Algorithms and Prior Algorithms Compared"},{"comment":"The scoring strategy depends on five weighting parameters α, γ, δ, β1, β2 (Algorithm 2) and on the depth-dependent scaling function f(d) used in the early-discarding cutoff. None of these values are reported anywhere in §VI or the appendix. Without them, the experiments are not reproducible, and the comparison could depend on parameter choices that favor Scoring over the baselines. Please report the exact parameter values and, ideally, a sensitivity analysis showing that the ~20% margin is stable across reasonable parameter ranges.","section":"Algorithm 2 and §VI Parameter Values"},{"comment":"Each data point is described as one 100-second NetSquid simulation, but the number of independent runs, seeds, confidence intervals, or statistical tests is never stated. Entanglement generation and swapping are stochastic, and the reported performance differences—especially the ~20% improvement over fixed-tree—need error bars or a seed analysis to be credible. Without this, it is impossible to tell whether the observed ordering of strategies is robust or within simulation noise.","section":"§VI, Simulation Setting"},{"comment":"The conclusion that a distributed GEM is architecturally necessary is based on the 'centralized module scheme', which is implemented with the oldest adaptive strategy rather than the scoring strategy or a centralized variant of the same policy. The poor performance of this centralized scheme could be due to the strategy choice, the lack of the same early-discard rule, or an unfavorable message-query rate, rather than to centralization per se. A fair comparison should use the same adaptive policy and same early-discard policy in both centralized and distributed settings.","section":"§VI, Centralized Module scheme"}],"minor_comments":[{"comment":"Typesetting and spacing issues: 'aGlobal Entanglement Module' appears in the abstract and elsewhere, and 'entanglementsynchronization' appears in §III-C. These should be corrected.","section":"Title/Abstract and §III-C"},{"comment":"The term 'Early Discarding Strategy' is used in §VI but the formal subsection is titled 'Proactive Strategy for Discarding Decohered EPs'. Please use consistent terminology so the reader can identify the mechanism.","section":"§IV.E"},{"comment":"The text says f(d) produces an 'exponentially decaying function' but no explicit family or parameterization is given. Even if the exact values are deferred, please define the functional form (e.g., f(d)=a·b^d with specified constants) or cite the source of the curve.","section":"§IV.E, Figure 4"},{"comment":"The execution-monitor thresholds (N_min, T_min, ΔT, T_susp, δr, δf, ηr, ηf) are listed as inputs in Algorithm 1 but no numerical values are provided in the evaluation section. If any of these were used in the NetSquid experiments, they should be reported.","section":"Algorithm 1 and §VI Parameter Values"},{"comment":"The figure captions do not identify which curves correspond to each strategy. Adding a legend or clarifying the line styles would make the results much easier to interpret.","section":"Figures 6–9"}],"recommendation":"major_revision","confidential_remarks":"The paper's comparisons rely heavily on the authors' own prior algorithms as baselines (swapping trees, predistribution), which is legitimate, but the undisclosed scoring weights and cutoff function increase the risk that the favorable comparison is an artifact of parameter selection rather than an architectural advantage. I would ask the editor to require full parameter disclosure and ideally a release of the simulation scripts or a detailed pseudocode appendix before reconsideration."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's the short version: this paper does something real—it puts a complete protocol stack on the table, with a clean separation between offline planning and real-time execution, and a Global Entanglement Module (GEM) that gives nodes a shared, best-effort view of entanglement resources. That fills a recognized gap in the quantum networking literature, and the authors are honest about the limits: broadcast-based eventual consistency, with hierarchical synchronization left to future work. The NetSquid implementation is a useful artifact, and benchmarking against fixed-tree and connectionless baselines is a sensible design.\n\nThe soft spots are in the evaluation, and they are load-bearing for the quantitative claim. The stress-test note is right: the adaptive strategies are all 'applied in conjunction with the Early Discarding Strategy' (§VI, Our Algorithms), but the Fixed-Swapping-Tree baseline from [11] is not described as using the same depth-dependent cutoff. Early discarding of aged intermediate EPs can improve throughput independently of any GEM-based adaptation. So the reported ~20% gain over the fixed tree may conflate the adaptive swapping policy with an auxiliary aging heuristic. The paper doesn't tell us whether the baseline had the same discard policy.\n\nRelated, the scoring strategy is under-specified. The weight parameters α, γ, δ, β1, β2 in Algorithm 2 and the depth-dependent cutoff f(d) are never defined numerically. That prevents reproducibility and leaves the door open to tuning. There are also no error bars, no seed counts, and no released code. Each data point is a single 100-second simulation per the evaluation section; a few independent runs would tell us whether the ~20% is real or noise. These are fixable, but as written the experiments don't support the strength of the headline claim.\n\nI don't think this is fatal. The architecture and the qualitative findings—scoring beats simple heuristics; distributed GEM beats centralized—are plausible and worth testing. I would send it to peer review, but with an expectation of major revisions: report the parameters, add the early-discarding baseline (or remove it from all conditions), and run multiple seeds. A reader working on quantum network protocol design will get value from the stack description and the evaluation structure, even if the numbers need to be re-taken. I wouldn't cite the quantitative result in my own work until those numbers are re-established, but I'd cite the architecture.","headline":"A genuinely integrative quantum network stack that deserves a real referee, but the headline ~20% gain is not yet established because the fixed-tree baseline may not use the same early-discard mechanism and the scoring parameters are unpublished.","tokens_in":22230,"tokens_out":3585,"would_cite":false,"duration_ms":30565,"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":"The paper proposes the first quantum-network protocol stack that separates offline planning from real-time execution, using a distributed Global Entanglement Module (GEM) to coordinate adaptive entanglement swapping, and reports a roughly 2","keywords":["quantum network protocol stack","global entanglement module","adaptive entanglement distribution","swapping trees","fidelity","pre-distributed entanglement","multipartite entanglement","quantum network simulation"],"falsifier":"Simulate the same stack on a network an order of magnitude larger (e.g., 500-1000 nodes) or with a classical-message loss rate above zero; the scoring strategy's ~20% rate advantage over the fixed-tree baseline should shrink or reverse if the GEM's global view is the load-bearing mechanism.","tokens_in":21411,"feed_emoji":"⚛️","tokens_out":4670,"duration_ms":40757,"temperature":0.7,"pith_summary":"The paper argues that a quantum network needs a full protocol stack analogous to the classical Internet, and that the missing piece is a module that maintains a consistent network-wide view of entangled pairs. It proposes a six-layer stack with a Global Entanglement Module (GEM) that synchronizes metadata about every active entanglement across nodes via broadcast updates, separating offline planning (swapping trees) from real-time adaptive execution. The central claim is that this separation, guided by GEM, lets lightweight local policies react to stochastic generation outcomes and decoherence, and that a scoring-based policy improves entanglement generation rates by about 20% over a globally optimal but non-adaptive fixed-tree plan, and more than doubles the rate of connectionless hop-by-hop routing. If correct, this gives a practical architecture for scalable quantum networks that naturally supports pre-distributed entanglement, purification, and multipartite state generation.","feed_headline":"Adaptive quantum network stack beats fixed-tree plans by 20%","feed_subtitle":"A distributed entanglement view lets lightweight scoring-based swaps double the rate of connectionless routing.","key_machinery":"Global Entanglement Module (GEM) — a cross-layer distributed data structure that keeps a best-effort, network-wide, near-real-time record of all active entanglements (endpoints, fidelity, age, usage) via timestamped broadcast updates. It is the carrier of the argument: by giving every node a consistent view of entanglement availability, it turns adaptive swapping from a local heuristic into a coordinated network-wide decision process, enabling the scoring-based strategy to balance immediate benefit, opportunity loss, and memory relief in real time.","core_discovery":"The paper's central discovery is that the bottleneck in quantum network protocol design is not a single layer but the absence of a cross-layer, network-wide view of entanglement resources. GEM is a distributed data structure—each node holds a replica tracked with timestamps and synchronized by broadcast—that records for each active entangled pair its endpoints, fidelity, age, and intended use. With this view, the Swapping/Fusion Layer can replace fixed swapping orders with real-time choices: a scoring function weighs immediate benefit (route-completion progress), opportunity loss (foregoing alternative swaps that the offline plan predicted), and memory pressure, then picks the swap with the","pith_inferences":["The broadcast-based eventually-consistent GEM is the scaling bottleneck; a hierarchical synchronization mechanism (which the paper defers to future work) or subscription-based updates would be the natural next step, and its absence likely caps the stack at networks of a few hundred nodes.","The scoring function's weighting parameters and the depth-dependent age cutoff embody implicit design choices; one could test whether adaptively tuning those weights online further widens the reported 20% gap.","The same GEM abstraction could extend to multipartite adaptive execution, which the paper leaves open, potentially carrying the rate gains to GHZ and graph-state distribution.","The paper's 'globally optimal' comparison is an offline optimum computed under the same expected parameters; a stronger test would compare against an online policy that also adapts the route, not just the swapping order."],"forward_implications":["If the stack is right, quantum networks can be built with a modular, Internet-like architecture where new adaptive execution policies are drop-in replacements.","The ~20% rate gain over a globally optimal fixed tree shows that real-time adaptation is not a compromise but an improvement even over optimal static plans.","The two-fold gain over connectionless approaches suggests that a shared global view is more valuable than purely local routing, at least in the simulated regime.","Support for predistribution, purification, and multipartite generation in one stack means applications (distributed quantum computing, GHZ distribution) can be built on a single unified service interface.","The failure of the centralized GEM (worse than fixed-tree) implies that distributing the module is architecturally necessary, not just an optimization."],"fun_headline_variants":["Global entanglement view lets quantum swaps adapt in real time, +20% rate","Quantum protocol stack's global view picks swaps on the fly, beating static plans","GEM: a network-wide entanglement map that doubles connectionless throughput","Scoring-based swap heuristics lift quantum entanglement rates 20% vs fixed tree"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The distributed GEM relies on broadcast updates over reliable classical channels; if broadcast overhead, staleness, or classical-channel unreliability grows with network size, the global view on which adaptive execution depends degrades, and the paper's evaluation only covers modest random networks, deferring hierarchical synchronization to future work.","fun_headline_variants_meta":{"raw":{"variants":["Global entanglement view lets quantum swaps adapt in real time, +20% rate","Quantum protocol stack's global view picks swaps on the fly, beating static plans","GEM: a network-wide entanglement map that doubles connectionless throughput","Scoring-based swap heuristics lift quantum entanglement rates 20% vs fixed tree"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001064,"raw_usage":{"total_tokens":4276,"prompt_tokens":699,"completion_tokens":3577,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":443,"completion_tokens_details":{"reasoning_tokens":3495}},"tokens_in":443,"tokens_out":3577,"duration_ms":22270,"temperature":1.0,"reasoning_tokens":3495,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T16:03:52.617702+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate the same stack on a network an order of magnitude larger (e.g., 500-1000 nodes) or with a classical-message loss rate above zero; the scoring strategy's ~20% rate advantage over the fixed-tree baseline should shrink or reverse if the GEM's global view is the load-bearing mechanism.","supporting_citations":[],"review_version":1}