{"id":"65a30ddf-3608-4ea9-9817-1b78952b752e","arxiv_id":"2601.17043","paper_version":2,"verdict":"REJECT","confidence":"MODERATE","novelty_score":3.0,"correctness_risk":"high","formal_verification":"none","parameter_count":3,"one_line_summary":"A proposal to treat detector deadtime as missing data in a recursive IIR-style trigger state; the claimed performance advantage is unsupported by any describable simulation.","lead":"This preprint proposes replacing binary coincidence triggers with a recursive filter that keeps a 'coherence state' alive during detector deadtime. It claims improved event recovery in simulations, but provides no simulation details, data, or code to support that claim.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"RMC's recovery advantage is unverified: efficiency curves in Figs. 5–6 lack matched false-alarm rates, threshold Γ, and a defined baseline, so the gain may be higher false positives from IIR persistence.","rationale":"The reader's REJECT verdict is well-founded. The paper's only evidence for the central claim is a simulation study that is not reproducible as described. My concern sharpens the reader's weakest-assumption point: the missing control is not just simulation realism, but the fundamental trigger-evaluation control of matched false-alarm rates. Without this, Figs. 5–6 cannot distinguish genuine recovery of correlated high-multiplicity events from an IIR filter's increased acceptance of background. The paper honestly lists limitations, and the algorithmic core is a standard, mathematically correct IIR filter, but that does not validate the empirical superiority claim. No code, data, or formal verification is provided. The proposed matched-operating-point test would settle whether the claimed advantage is real; until it is run, the central claim remains unsupported. Therefore the reader's REJECT verdict should stand unchanged.","tokens_in":8802,"tokens_out":3256,"duration_ms":39420,"concrete_test":"Re-run the Section 4 simulation with a fully specified event generator and liveness model (e.g., Poisson signal and background rates, independent per-channel deadtime probability p_d). For each p_d, choose RMC's threshold Γ so that its false-trigger rate on background-only data equals the baseline coincidence trigger's false-trigger rate (specifying the baseline's coincidence window and multiplicity). Then recompute event-recovery efficiency at these matched operating points. If RMC's efficiency advantage shrinks or disappears, the central claim fails. Repeat the comparison with a correlated-background process (e.g., Hawkes) to test whether RMC's memory inflates false triggers.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim rests on the simulation comparison in Section 4, especially Figs. 5–6, which show RMC's event-recovery efficiency rolling off more slowly than 'baseline coincidence logic' as deadtime probability increases. But efficiency alone is not a meaningful trigger metric: any acceptance rule can achieve high efficiency by lowering its threshold, and the useful comparison is at a fixed false-alarm rate. The paper never specifies the background/noise rate, the threshold Γ used for RMC, the coincidence window/multiplicity parameters for baseline, or the operating point (false-trigger rate) of either. This matters because RMC (Eq. 2.3 with A=ρI) is an exponential smoother: during deadtime, x[k]=ρx[k−1], so both signal-correlated state and accumulated background fluctuations persist. With ρ=0.85–0.95, a long-memory accumulator will naturally raise the chance that noise crosses any fixed Γ, making RMC appear to 'recover' events that are actually false positives. Section 6 concedes that correlated backgrounds, afterpulsing, and calibration drifts are absent, exactly the effects that would penalize a recursive accumulator. Without matched false-alarm control or a fully specified simulation, the claimed 'superior robustness' is not established.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes Recursive Manifold Coherence (RMC), a trigger-logic framework that replaces binary coincidence windows with a first-order linear recursion x[k] = Ax[k-1] + Bu[k] over validated charge/timing features u[k] = L[k]y[k], where L[k] encodes channel liveness. With the recommended choice A = rho I, the state decays smoothly during deadtime instead of resetting, so the scalar trigger score G[k] = phi(x[k]) can remain above threshold even when direct coincidence chains are broken. The authors claim, based on 'simulation studies representative of large optical detector arrays' in Section 4, that RMC recovers event-level coherence for high-multiplicity topologies and degrades more gracefully than baseline coincidence logic as deadtime probability increases. Section 2 presents the update law and an information-geometric interpretation; Section 3 discusses implementation complexity and firmware compatibility; Section 6 concedes that correlated backgrounds, afterpulsing, and calibration drifts are not modeled.","tokens_in":9158,"tokens_out":3984,"duration_ms":48122,"significance":"If the performance claim were established, the proposal would offer a simple, bounded-memory alternative to coincidence-based triggers, with an implementation cost as low as O(m) per sample when A = rho I. The algebraic recursion is correct as a linear filter, and the paper usefully identifies that stateless binary logic discards correlated evidence during deadtime. However, the central demonstration is not supported. The simulation study is unspecified, no false-alarm-rate control is provided, and the 'deadtime bridging' behavior is definitionally built into the exponential-smoothing update law. The geometric apparatus (Eq. 2.2) is never used in the evaluation. As a result, the claimed 'superior robustness' is not established, and the contribution currently reduces to an exponential smoother with a decorational information-geometry layer.","major_comments":[{"comment":"The central performance claim rests entirely on the simulation comparison, but the simulation is not specified. There is no generative model for true events, no noise or background rate, no detector response function, no liveness/deadtime process, no definition of the baseline 'coincidence logic', no trigger threshold Gamma, and no false-alarm or trigger-rate operating point. Efficiency alone is not a valid trigger metric: any acceptance rule can achieve high efficiency by lowering its threshold. Because RMC with rho in [0.85, 0.95] is a long-memory accumulator, it will also accumulate background fluctuations; without matched false-alarm rates, the slower efficiency roll-off in Figs. 5-6 may simply reflect a higher false-trigger rate.","section":"Section 4, Figs. 5-6"},{"comment":"The claimed 'recovery' of coherence through deadtime is definitional, not empirical. When u[k] is approximately zero, Eq. (2.3) reduces to x[k] = A x[k-1], so with A = rho I the state decays exponentially by construction. The deadtime-bridging visualization (Fig. 1 and Fig. 3) is therefore a plot of the update law, not evidence that RMC 'recovers event-level coherence'. To substantiate the claim, the authors need an independent definition of the true event process and a detection rule whose probability of detection is evaluated at a fixed false-alarm probability.","section":"Section 2.3, Remark 2.1 and Section 4.2"},{"comment":"The limitations section concedes that spatially heterogeneous noise, correlated backgrounds, afterpulsing, and time-dependent calibration drifts are not modeled. These are exactly the effects that would penalize a recursive accumulator with fixed persistence, because they create correlated or slowly varying input that the IIR memory would retain. The absence of these effects in the simulation, combined with the lack of matched false-alarm control, makes the claim of 'superior robustness against data fragmentation' in the abstract unsubstantiated for realistic detector conditions.","section":"Section 6"},{"comment":"The information-geometric metric G_ij(theta) in Eq. (2.2) and the scoring functional phi in Eq. (2.4) are never instantiated in the evaluation. The simulation results in Figs. 5-6 depend only on the linear recursion with A = rho I; the manifold geometry does not enter the update or the decision rule. Either the geometric component should be operationalized and tied to the reported results, or the paper should be repositioned as a study of recursive persistence for deadtime handling, without the manifold-coherence framing.","section":"Sections 2.2, 2.4, and 4"}],"minor_comments":[{"comment":"The notation 'dTheta^2 = G_ij(theta) dtheta^i dtheta^j' is nonstandard; usually one writes ds^2 or explicitly sums over repeated indices. This should be corrected for clarity.","section":"Eq. (2.2)"},{"comment":"The x-axis label is inconsistent: Fig. 5 uses 'Deadtime probability' in [0,0.7], while Fig. 6 uses 'deadtime probability' in [0,50] (apparently percent). The y-axis 'Event recovery efficiency' is undefined; a definition in the text or caption is needed.","section":"Fig. 5 and Fig. 6"},{"comment":"The plotted 'uncertainty' is not defined. Since the recursion is deterministic, the uncertainty envelope presumably comes from an external model, but the text does not specify how it is computed.","section":"Fig. 3"},{"comment":"The caption contains a typo: 'weights / exp(¡1 2 d 2 M)' appears to be a character-encoding artifact; should read something like 'exp(-(1/2) d_M^2)'.","section":"Fig. 2 caption"},{"comment":"Reference [23] is a self-citation to an arXiv preprint whose connection to the present work is not explained; if it is background, it should be cited in context, not listed without discussion.","section":"References"}],"recommendation":"reject","confidential_remarks":"The manuscript has the structure of a framework proposal but claims a simulation-based demonstration. As a referee, I cannot verify the headline result because the simulation is not specified and the comparison lacks operating-point control. More fundamentally, the 'recovery' behavior is a direct consequence of the chosen A = rho I update; showing an exponential decay is not evidence for the paper's central claim. The information-geometric apparatus is not used in the evaluation. Even a substantial revision would need to replace Section 4 with a fully specified study including false-alarm matching and realistic noise models, and would need to either operationalize or de-emphasize the manifold framing. Given the current state, publication as a serious instrumentation paper is not warranted."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Brief take: the paper's only real content is Eq. (2.3) with A = ρI, a first-order IIR filter, applied to deadtime as missing data. That's a standard tool, and the paper says so. The 'manifold coherence' framing is mostly decoration. The claimed advantage over coincidence logic rests on simulation Figures 5–6, but there is no event generator, no noise model, no threshold Γ, and no false-alarm matching. Without that, the efficiency curves just show that a recursive accumulator with ρ≈0.9 remembers old evidence — which is the definition of the filter's persistence. So the central claim is unverified, not wrong.\n\nWhat the paper does well: it's clearly written, the recursive update is stated exactly, and Section 6 honestly lists what's missing (correlated backgrounds, afterpulsing, calibration drifts). The idea of using exponential smoothing for deadtime bridging is not crazy; it's just not new. Ref [27] covers intermittent observations, and exponential smoothing is textbook. I don't see a new derivation or a new algorithm.\n\nSoft spots in proportion: the biggest problem is the simulation. Figure 5 shows RMC 'remains stable' across ρ, but there's no mention of false-trigger rate. A higher ρ also holds onto background fluctuations longer, so the apparent recovery may simply be more false positives. The paper never gives the threshold Γ or the baseline coincidence parameters. This is not a minor omission; it's the entire empirical case. Also, the 'information metric' G_ij in Eq. (2.2) is never instantiated; the framework says it could be Fisher information, but the actual update never uses it. The metric is inert. The deadtime-bridging visualization is just the filter's impulse response, not evidence.\n\nIs the central argument coherent? Yes, in the narrow sense: given A = ρI and u = 0, x decays — that follows. And a decaying state can keep a trigger above threshold. But that's built into the update, not discovered. The paper calls it 'recovery' and 'superior robustness,' which oversells a linear filter's memory.\n\nWho this is for: someone who wants a simple, cheap way to soften deadtime in a trigger might take the idea and test it themselves. But they'd need to specify the operating point and compare false-alarm matched. As a paper, it doesn't meet that bar.\n\nRecommendation: I would not send this to peer review in its current form. It needs a real, reproducible simulation study with a defined baseline, threshold, and false-alarm control, or a hardware prototype. If the authors come back with that, the kernel is worth a proper look.","headline":"RMC is an IIR filter in a manifold costume; the only empirical evidence is a simulation that isn't specified, so the central claim doesn't land.","tokens_in":9651,"tokens_out":2607,"would_cite":false,"duration_ms":27253,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["07.05.Hd","29.40.-n"],"model":"deepseek-v4-flash","headline":"This paper claims that a simple recursive filter, applied to a low-dimensional coherence state, can preserve event-level trigger information across detector deadtime where binary coincidence logic fails.","keywords":["deadtime","pile-up","distributed trigger","information geometry","recursive state estimation","software-defined triggering","coincidence logic","neutrino observatory"],"falsifier":"Run RMC against a coincidence trigger on a benchmark generator with an independent signal process—e.g., Poisson-distributed hits with wideband timing jitter and per-channel random deadtime not correlated with the signal—and measure efficiency versus deadtime probability. If RMC's advantage narrows to the persistence time of ρ and vanishes when ρ is tuned to the deadtime duration, the proposed 'manifold coherence' mechanism is not the operative driver.","tokens_in":8663,"feed_emoji":"⚛️","tokens_out":3506,"duration_ms":36017,"temperature":0.7,"pith_summary":"The paper argues that detector deadtime need not erase trigger-level evidence. It replaces binary coincidence logic with a low-dimensional 'coherence state' updated by a first-order recursion (an exponential smoother), so during non-live intervals the state decays smoothly instead of resetting. In simulations with increasing deadtime probability, the recursion's event-recovery efficiency rolls off far slower than standard coincidence logic, recovering high-multiplicity events whose coincidence chains are broken. The authors frame this as a shift from hard vetoes to stateful, missing-data-aware triggering, with a fixed per-sample computational cost suitable for real-time pipelines.","feed_headline":"One recursive filter rescues events lost to detector deadtime","feed_subtitle":"Binary coincidence logic fragments under detector deadtime; a smoothed coherence state decays gracefully and re-locks when hits resume.","key_machinery":"The recursive coherence state: a first-order IIR/exponential-smoothing recursion x[k] = Ax[k−1] + Bu[k], where A = ρI provides bounded exponential memory. Liveness indicators Li[k] gate observations so deadtime becomes suppressed input rather than a reset, causing the state to decay with controlled uncertainty. A correlation-based metric (Fisher information, covariance, or correlation-weighted approximation) defines distances in feature space and supplies the weighting for the state update.","core_discovery":"The central claim is that treating deadtime as missing data in a stateful estimator—rather than as a hard veto—preserves event-level coherence. The update law x[k] = Ax[k−1] + Bu[k], with A = ρI giving exponential memory, propagates correlation evidence across gaps, and the scalar coherence score G[k] = φ(x[k]) can exceed threshold even when direct coincidence is broken. Simulation results show RMC retains higher efficiency than coincidence logic as deadtime probability rises, and the deadtime-bridging visualization shows the state re-locking to the truth once observations resume.","pith_inferences":["The mechanism is formally an IIR smoother, so the claimed advantage may reduce to choosing a persistence time constant longer than the deadtime gap; a fair baseline should include a stateful (smoothed) coincidence trigger to isolate the contribution of the coherence geometry itself.","The state decay rate could be calibrated to known deadtime statistics, turning the trigger into a continuous live-fraction estimator and connecting this work to standard deadtime-correction theory.","A testable extension would be to run the recursion on real data with afterpulsing or correlated backgrounds—explicitly absent in this study—to see whether state-dependent persistence or low-rank extensions are needed.","Because the recursion is linear and scalar, it could be interpreted as a matched filter for signal persistence, suggesting that the detector-agnostic claim may be better phrased as a tunable-persistence trigger rather than a manifold-geometric one."],"forward_implications":["Trigger decisions can be made with bounded memory and fixed per-sample cost, compatible with FPGA fixed-point and software-based pipelines.","Events with high multiplicity and extended topology survive partial deadtime where coincidence triggers lose them.","Deadtime becomes observable as uncertainty growth rather than silent data loss, giving operators a continuous liveness diagnostic.","The same recursion works across detector geometries because it only requires compact charge/timing features and liveness flags.","With an appropriate persistence parameter, trigger efficiency degrades gracefully with deadtime probability rather than sharply collapsing."],"fun_headline_variants":["Geometric filter recovers events lost to detector deadtime","Coherence state bridges detector deadtime gaps","Stateful trigger logic survives detector deadtime","Manifold coherence beats binary triggers under deadtime"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The claimed advantage rests on the unstated simulation being a fair representation of real detector deadtime and signal correlation; the paper gives no event generator, background model, or baseline specification, and its own limitations section admits correlated backgrounds, afterpulsing, and calibration drifts are absent.","fun_headline_variants_meta":{"raw":{"variants":["Geometric filter recovers events lost to detector deadtime","Coherence state bridges detector deadtime gaps","Stateful trigger logic survives detector deadtime","Manifold coherence beats binary triggers under deadtime"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000165,"raw_usage":{"total_tokens":1077,"prompt_tokens":722,"completion_tokens":355,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":466,"completion_tokens_details":{"reasoning_tokens":295}},"tokens_in":466,"tokens_out":355,"duration_ms":3913,"temperature":1.0,"reasoning_tokens":295,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T09:23:59.584321+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run RMC against a coincidence trigger on a benchmark generator with an independent signal process—e.g., Poisson-distributed hits with wideband timing jitter and per-channel random deadtime not correlated with the signal—and measure efficiency versus deadtime probability. If RMC's advantage narrows to the persistence time of ρ and vanishes when ρ is tuned to the deadtime duration, the proposed 'manifold coherence' mechanism is not the operative driver.","supporting_citations":[],"review_version":1}