{"id":"5c2ccba1-a957-4207-a2e9-5dbb3b0dfc02","arxiv_id":"2608.07163","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"An adaptive link-layer controller improves mean entanglement distribution rate by 14% over optimized static settings in a trace-driven simulation of a 64 km deployed fiber link.","lead":"An adaptive software controller for quantum fiber links tunes pump power and polarization compensation in real time to keep entanglement quality above an application target while maximizing distribution rate. In trace-driven simulations over a 64 km deployed fiber, it beats optimized static settings by 14% without any additional quantum hardware.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The headline gain rests on a compensation model fit to a similar link and validated only on durations; if the model's exit fidelities are wrong, the 14% advantage may not survive.","rationale":"The reader identified the APC model as the weakest assumption, and this is also the most load-bearing concern for the central claim. The claim is a quantitative rate improvement over a 24-hour trace, and every simulation of compensation behavior flows through Algorithm 1. The model is not validated on the target link, is fit only against compensation durations, and is explicitly flagged by the authors as potentially affecting results. The controller's timing policy in Equation 7 uses the modeled tradeoff between rate loss and compensation benefit, and the fixed 1-second compensation duration in Section 5.4 assumes the modeled behavior is accurate. Secondary concerns, such as possible overlap between the training and evaluation windows and the absence of error bars, are real but less decisive because even a clean held-out evaluation would still depend on the APC model. Since the paper provides public code and data and is internally consistent, a conditional verdict remains appropriate. The proposed check would settle whether the 14% margin survives when the simulated compensation response is replaced by empirical APC outcomes.","tokens_in":21383,"tokens_out":10363,"duration_ms":100755,"concrete_test":"Replace Algorithm 1's predicted compensation outcome with the empirical 70-hour APC dataset: for each simulated trigger, sample a realized compensation duration and exit fidelity from the recorded APC runs (or from a bootstrap of those runs) instead of simulating vanilla Riemannian gradient descent. Recompute the 24-hour mean entanglement distribution rate for both the adaptive protocol and the best static protocol under the same drift trace. If the 14% margin is not preserved, or if the adaptive protocol produces Fmin violations that the duration-only fit hides, the headline claim is not supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is a 14% rate improvement computed inside a trace-driven simulator whose most consequential physical submodel is the APC response model of Section 3.3 and Algorithm 1. The model assumes vanilla Riemannian gradient descent with a fitted step size and step duration from Appendix C, and it is fit to 70 hours of data from a similar 62 km link rather than the 64 km link used in evaluation. The reported comparison in Figure 12 checks only the distribution of compensation durations, not the final fidelity with which the APC exits, even though the controller's benefit depends on both how long compensation takes and how much fidelity it restores. The protocol's compensation timing condition in Equation 7 and the fixed 1-second compensation duration in Section 5.4 are optimized against this modeled response. If the real device restores less fidelity within 1 second, or fails more often under high drift, the adaptive controller's rate advantage and its ability to honor the Fmin constraint could both change. The paper itself concedes that deviations from this model may affect results upon deployment (Section 3.3), and no hardware-in-the-loop validation of the adaptive controller on the 64 km link is provided. The quantitative headline therefore rests on an unverified fit.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a classical control layer for single quantum link operation, formulated as a joint optimization problem (Opt-2) over a tunable rate-fidelity source parameter and an uncontrollable, time-varying polarization drift. The authors instantiate the framework on a 64 km deployed fiber in the DC-QNet, using an SPDC entanglement source whose pump power trades fidelity against rate, an empirical polarization drift prediction model built from measured Stokes parameters, and a model of a commercial automated polarization compensator (APC) as noisy Riemannian gradient descent. The adaptive protocol sets pump power conservatively to satisfy a minimum fidelity constraint, schedules fidelity checks by a value-of-information criterion, and triggers fixed-duration polarization compensations. A trace-driven simulator evaluates the protocol over a 24-hour trace and reports a 14% mean entanglement rate improvement over the best optimized static policy, with no offline policy optimization.","tokens_in":21617,"tokens_out":5483,"duration_ms":54558,"significance":"If the quantitative results are reliable, the paper makes a useful contribution: it demonstrates that software-only physical-layer control can improve near-term quantum link throughput, and it packages the problem in a hardware-agnostic abstraction (Opt-0/Opt-1/Opt-2) that could generalize to other source and compensation technologies. The authors should be credited for releasing simulator source code and trace data, for comparing against an upper bound with zero drift, and for identifying the practical overheads (probe and compensation downtime). However, the headline 14% figure is a simulated outcome whose most consequential submodel is the APC response model, and the manuscript explicitly concedes that deviations from this model may affect the results upon deployment (Section 3.3). The main contributions are therefore significant conditional on additional validation.","major_comments":[{"comment":"The central quantitative claim depends on the modeled exit fidelity of the APC, but the model is validated only against the distribution of compensation durations, not against the final polarization fidelity achieved. Figure 12 compares histograms of compensation time, and Appendix C reports only a visual fit. The controller's benefit, however, comes from both the duration of compensation and the fidelity it restores: Equation 7 and the fixed 1-second compensation policy determine when to compensate, while the resulting Fpol after compensation determines whether the Fmin constraint can be honored and how much rate is recovered. The APC model is also fitted to 70 hours of data from a similar 62 km link rather than the 64 km link used in evaluation, and its algorithm is proprietary, so the paper assumes vanilla Riemannian gradient descent. Given the paper's own caveat that deviations from this model may affect results upon deployment, the manuscript should provide direct validation of modeled exit fidelities against the real APC, or in lieu of hardware validation, a sensitivity analysis over the APC model parameters (eta, Delta, failure rates) showing that the 14% improvement is robust.","section":"Section 3.3, Algorithm 1, Appendix C, Figure 12"},{"comment":"The polarization drift prediction model is trained on a 48-hour tracking experiment described in Section 3.2, while the evaluation trace in Section 6.1 spans three days. It is not stated whether the evaluation days are disjoint from the training period. If the same trace is used for both fitting the drift model (including the 0.1-conservative quantile used in Equation 4) and evaluating the protocol, the reported rate improvement may be substantially inflated by in-sample prediction. The manuscript should state explicitly whether the training and evaluation periods are disjoint, and if they are not, the evaluation should be repeated on held-out drift data or with a cross-validated protocol.","section":"Section 3.2 vs. Section 6.1"},{"comment":"The headline comparison is a single 24-hour trace without confidence intervals or day-to-day variability. Figure 7 reports one mean rate for each protocol on one trace, and the static baselines are optimized on the same trace used for comparison. Because polarization drift is highly environment-dependent and the paper itself notes that the best static protocol for one day can be very different for another, the 14% and 16% improvements should be accompanied by a multi-day or bootstrapped analysis, or at least a clear statement that these are single-trace results. As written, the quantitative gain is not shown to be statistically reliable.","section":"Section 6.2, Figure 7"}],"minor_comments":[{"comment":"The sentence fragment 'Pump power is updated every 100ms. attenuator that sets the pump power to the desired value.' appears to be missing a verb and should be rewritten for clarity.","section":"Section 6.1"},{"comment":"There is a typo in 'the the Washington DC metropolitan quantum network' on the first page of Section 2.4; it should read 'the Washington DC metropolitan quantum network.'","section":"Section 2.4"},{"comment":"The statement that 'No parameter choices are required for the adaptive algorithm' is too strong. The protocol requires choosing delta = 0.10, the compensation duration c = 1 s, the model step size eta, and the step duration Delta, all of which affect performance. These are not tuned offline for the policy, but they are free parameters and should be described as such.","section":"Section 5.2, Section 5.4, Section 7.1"},{"comment":"The sign and semantics of the noise term Delta * d(t) in Algorithm 1 should be specified more precisely. As written, the term can either help or hinder the gradient descent depending on the sign of the drift rate, and it is not clear whether d(t) is the signed rate of change of the drift angle or an unsigned magnitude. This affects both the model fit and the interpretation of Figure 12.","section":"Algorithm 1 and Section 3.3"},{"comment":"The claim that setting Ftarget = 1.0 and T_timeout = c 'ensures that the compensation routine will take time c' relies on the assumption that the APC never reaches unit fidelity before the timeout. The paper argues that continuous drift makes Fpol < 1.0 at all times, but occasional early termination would violate the constant-duration assumption used in Equation 7 and Appendix E; this edge case should be addressed.","section":"Section 5.4"}],"recommendation":"major_revision","confidential_remarks":"The central idea is promising and the paper is clearly written, but the quantitative headline rests on an unvalidated APC model and possibly in-sample drift prediction. I recommend major revision rather than rejection because the missing evidence is identifiable and could be supplied by held-out evaluation, APC exit-fidelity validation or a sensitivity analysis, and multi-day uncertainty quantification. The novelty relative to existing link layer and control work should also be clarified in the revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know this paper is a solid engineering result with a real gap between its clean simulation and its deployment-ready claim. The new thing is the joint control of pump power and polarization compensation scheduling — previous work fixed one or the other, and nobody had framed it as a single optimization problem with an explicit fidelity constraint. The Opt-2 decomposition is clear, the protocol is simple enough to run in real time, and the evaluation against optimized static policies is fair (they actually tune the static parameters rather than strawmanning them). The code and trace are public, which is a real plus.\n\nThat said, the 14% rate improvement is a simulation result built on a compensation model that is the softest link in the chain. The APC model is fitted to 70 hours of data from a similar 62 km link, not the 64 km link used in evaluation, and because the device algorithm is proprietary the authors assume vanilla Riemannian gradient descent. Appendix C validates the model only against compensation durations, not against final fidelity — but the controller's benefit depends on both. The paper itself concedes (Section 3.3) that deviations from this model may affect results on deployment. That is a honest caveat, but it means the headline number is conditional on an unverified fit.\n\nThe other issue is the drift prediction model. It is trained on a 48-hour window and evaluated on a 24-hour trace, and the paper never says those windows are disjoint. The reader flagged this and it is a fair catch: if the evaluation trace overlaps the training window, the prediction model gets in-sample credit and the 14% could be inflated. No error bars are reported anywhere, which makes it hard to tell how much noise is in that number.\n\nThese are fixable. A one-paragraph statement that the evaluation trace is fully held out, plus a sensitivity analysis that varies the APC model's fidelity restoration, would turn this from a conditional into a substantive claim. As written, I would not quote the 14% as a measured gain.\n\nFor whom: quantum networking researchers, especially people working on link-layer control and real deployments, will get genuine value from the framing and the protocol. The upper-bound comparison is honest and the nighttime gap under 8% suggests there is not a huge amount of headroom left, which is itself useful. The paper deserves peer review — with a serious referee who will ask for the held-out statement and the APC sensitivity. I would accept it with that expectation.","headline":"Genuinely new joint control formulation for quantum links, but the 14% headline is a simulation claim that needs a held-out evaluation trace and a better-validated APC model before it should be cited as a deployment gain.","tokens_in":22147,"tokens_out":1760,"would_cite":true,"duration_ms":17000,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P45","81P40"],"pacs":["03.67.Hk","03.67.Bg"],"model":"deepseek-v4-flash","headline":"A classical software controller that predicts polarization drift and adaptively tunes pump power and compensation timing can raise mean entanglement distribution rate on a 64 km deployed fiber link by 14% versus optimized static policies…","keywords":["quantum networks","entanglement distribution","polarization drift","rate-fidelity tradeoff","adaptive link control","fiber-optic quantum links","pump power control","trace-driven simulation"],"falsifier":"Measure the actual compensation durations and resulting fidelities of the commercial APC on the same 64 km link across the same 24-hour window, replace Algorithm 1 in the simulator with the measured response model, and re-run the comparison; if the adaptive protocol's 14% margin over the best static policy shrinks below the static policies' own parameter sensitivity, the central claim fails. A cheaper check is to compare the simulated compensation-time histogram (the paper's Figure 12) against device logs on a live link, since the current fit is judged only visually.","tokens_in":2042,"feed_emoji":"🔗","tokens_out":9789,"duration_ms":112323,"temperature":0.7,"pith_summary":"The paper argues that near-term wide-area quantum links are held back not only by quantum hardware but by how existing control knobs are operated, and that a purely software controller can recover a large share of the lost entanglement rate. It poses link operation as a joint control problem between a controllable rate-fidelity tradeoff (source pump power) and uncontrollable polarization drift, then builds a protocol that sets pump power from a conservative quantile of a predicted polarization fidelity, schedules fidelity checks by a value-of-information criterion, and triggers fixed-duration compensations when the instantaneous rate falls below the period average. In trace-driven simulation on 64 km of deployed metropolitan fiber, the adaptive protocol beats the best static policy by 14% mean entanglement rate over 24 hours, lands within 21% of the zero-drift upper bound, and requires no offline policy optimization. The practical point is that this gain comes from software over existing hardware, so it could be deployed near-term on any link exposing the same three abstractions: a rate-fidelity frontier, a drift model, and a compensation primitive.","feed_headline":"Adaptive software control boosts quantum link rates by 14%","feed_subtitle":"A drift-predicting controller tunes pump power and compensation timing to beat static policies on a 64 km fiber.","key_machinery":"The load-bearing object is the Opt-2 feedback loop: a joint control problem that couples the source-detector operating point to the compensation schedule. It is instantiated through three replaceable interfaces, namely a calibrated rate-fidelity frontier (here, SPDC pump power, where higher power raises rate but multipair emission degrades fidelity), an empirical drift-prediction model that turns past Stokes-vector drift into a belief distribution over the polarization fidelity $F_{\\mathrm{pol}}$, and a compensation model, Algorithm 1, which simulates the commercial compensator as a noisy Riemannian gradient descent $\\theta_{k+1} = \\theta_k - \\frac{\\eta}{2}\\sin\\theta_k$ with fitted step size $\\eta = 0.062$ and step duration $\\Delta = 27.7$ ms. The control law reduces to two threshold rules that carry the argument: pump power keeps the product $F_{\\mathrm{sd}}F_{\\mathrm{pol}}$ above $F_{\\min}$ with probability $1-\\delta$, and compensation fires when the rate curve crosses the running period average, which the paper proves (Appendix E) maximizes the period's average rate whenever the rate is decreasing.","core_discovery":"On the paper's own terms, the discovery is that a wide-area link's polarization drift can be treated as a partially predictable disturbance, and a classical controller that (i) chooses source pump power so that $\\Pr[F(t) < F_{\\min}] \\le \\delta$ with $\\delta = 0.10$ using a drift-prediction model, (ii) initiates a fidelity check only when the expected throughput gain $\\Lambda_1 - \\Lambda_2$ covers the downtime cost $r(F_{\\mathrm{pol}}^\\delta) t_{\\mathrm{check}}$, and (iii) starts a fixed 1-second polarization compensation exactly when the instantaneous rate meets the period-average rate $r(t) \\le \\bar{r}(t)$, achieves 14% higher mean entanglement distribution rate over a 24-hour trace than any optimized static configuration of the same five control variables. The controller needs no quantum hardware changes and no offline policy search; its inputs are the fidelity target, a calibrated drift model, and the source-detector rate-fidelity curve. In the same simulations the protocol tracks a continuously varying fidelity demand on the fly, and it stays within 21% of the theoretical no-drift rate ceiling over a day (under 8% at night), which the authors read as bounding the headroom left for better prediction or compensation hardware.","pith_inferences":["If the 14% margin survives a true deployment, the paper's implicit claim is that physical-layer classical control is a cheaper performance lever than hardware upgrades; a field deployment that compares the adaptive controller against the same static baselines on a live link is the direct test.","The value-of-information criterion for fidelity checks is a general probe-and-compensate scheduling principle that could transfer to other costly measurements on quantum links, such as clock synchronization checks, dispersion monitoring, or quantum-memory calibration.","The fixed 1-second compensation policy is a deliberate truncation of a diminishing-returns tail; an untested alternative is letting the timeout scale with the predicted drift rate, which could buy fidelity headroom at night, where the gap to the upper bound is already under 8%.","The $\\delta$-conservative pump power rule is an empirical chance constraint; a calibrated drift model would allow it to be recast as a rigorous chance-constrained controller with formal safety guarantees, which the paper does not attempt."],"forward_implications":["Deployed wide-area quantum links can gain roughly 14% mean entanglement rate purely from a software layer over 24 hours, with the margin rising to 16% at night when drift is slower and falling to 6% during the day.","The protocol adapts instantaneously to changing application fidelity demands, including continuously varying ones, whereas each static policy must be re-tuned for every new $F_{\\min}$ value.","The adaptive controller operates within 21% of the zero-drift upper bound over a full day and within 8% at night, with only about 3.4% of night time spent on probes and compensation, so the remaining room for better prediction or compensation is modest.","Because the controller is defined over hardware-agnostic interfaces, the same protocol applies to other sources (four-wave mixing, quantum memories), other encodings, and other compensation mechanisms by swapping in recalibrated models.","The Opt-2 formulation exposes a per-link feasible set of rate-fidelity operating points that higher layers can exploit, and its online decisions are $O(1)$ computations, so the protocol fits inside 100 ms control loops."],"supporting_citations":[{"why":"Supplies the public simulator and the 100 ms Stokes-parameter trace data from the 64 km link on which every reported rate is computed.","marker":"[8]"},{"why":"Describes the commercial automated polarization compensation device and its prior entanglement distribution over deployed fibers; the paper's compensation model is built around this device.","marker":"[10]"},{"why":"Provides the commercial four-wave-mixing source characterization showing the same rate-fidelity tradeoff, which supports the hardware-agnostic control abstraction.","marker":"[11]"},{"why":"Establishes the link-layer rate-fidelity control precedent (Opt-0) and the fidelity-requirement rationale that the paper's protocol builds on.","marker":"[12]"},{"why":"Gives the fidelity formula $F_{\\mathrm{pol}} = \\frac{1}{2}(1+\\cos\\theta)$ that converts measured drift angles into the polarization fidelity used throughout the controller.","marker":"[23]"},{"why":"Characterizes the DC-QNet metropolitan testbed whose 64 km LTS-to-NIST fiber provides the drift trace and deployment context for the simulations.","marker":"[33]"},{"why":"Provides the state-of-the-art static protocol baseline, with fixed APC parameters ($F_{\\mathrm{trigger}} = 0.98$, $F_{\\mathrm{target}} = 0.99$, $T_{\\mathrm{timeout}} = 55$ s) and constant pump power, that the adaptive protocol must outperform.","marker":"[43]"}],"fun_headline_variants":["Software-only tweak lifts quantum link rates 14%","Drift-aware control boosts quantum entanglement rate 14%","14% faster entanglement, no new hardware: adaptive control","Adaptive pump-power control beats static quantum link policies","Adaptive control delivers 14% quantum rate gain on deployed fiber"],"cache_read_input_tokens":24320,"weakest_assumption_plain":"The simulated model of the commercial polarization compensator, Algorithm 1 with fitted step size $\\eta = 0.062$ and step duration $\\Delta = 27.7$ ms, faithfully represents the real device on this link; the paper assumes vanilla Riemannian gradient descent because the device's algorithm is proprietary and states that deviations from the model may affect results upon deployment.","fun_headline_variants_meta":{"raw":{"variants":["Software-only tweak lifts quantum link rates 14%","Drift-aware control boosts quantum entanglement rate 14%","14% faster entanglement, no new hardware: adaptive control","Adaptive pump-power control beats static quantum link policies","Adaptive control delivers 14% quantum rate gain on deployed fiber"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001577,"raw_usage":{"total_tokens":6317,"prompt_tokens":993,"completion_tokens":5324,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":609,"completion_tokens_details":{"reasoning_tokens":5242}},"tokens_in":609,"tokens_out":5324,"duration_ms":33536,"temperature":1.0,"reasoning_tokens":5242,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T14:28:14.118002+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the actual compensation durations and resulting fidelities of the commercial APC on the same 64 km link across the same 24-hour window, replace Algorithm 1 in the simulator with the measured response model, and re-run the comparison; if the adaptive protocol's 14% margin over the best static policy shrinks below the static policies' own parameter sensitivity, the central claim fails. A cheaper check is to compare the simulated compensation-time histogram (the paper's Figure 12) against device logs on a live link, since the current fit is judged only visually.","supporting_citations":[{"cited_title":"Rate-Fidelity Control for Wide-Area Quantum Links","cited_arxiv_id":null,"evidence_quote":"Supplies the public simulator and the 100 ms Stokes-parameter trace data from the 64 km link on which every reported rate is computed."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Describes the commercial automated polarization compensation device and its prior entanglement distribution over deployed fibers; the paper's compensation model is built around this device."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the commercial four-wave-mixing source characterization showing the same rate-fidelity tradeoff, which supports the hardware-agnostic control abstraction."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the link-layer rate-fidelity control precedent (Opt-0) and the fidelity-requirement rationale that the paper's protocol builds on."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Characterizes the DC-QNet metropolitan testbed whose 64 km LTS-to-NIST fiber provides the drift trace and deployment context for the simulations."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the state-of-the-art static protocol baseline, with fixed APC parameters ($F_{\\mathrm{trigger}} = 0.98$, $F_{\\mathrm{target}} = 0.99$, $T_{\\mathrm{timeout}} = 55$ s) and constant pump power, that the adaptive protocol must outperform."}],"review_version":2}