{"id":"4648cbf1-38c9-4c3c-92e5-5f204d37f220","arxiv_id":"2608.05824","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":7,"one_line_summary":"A single analytical chain from crystal kinetics to SiPM microcell recovery yields closed-form scintillation pulses and a 100 ps timing bound, with a fitted saturation model preferred by AIC on high-amplitude pulses.","lead":"The paper builds one mathematical model connecting how LYSO crystals emit light, how that light travels to a silicon photomultiplier, and how the detector's microcells recover, to predict pulse shape and timing limits. It also derives the widely used two-exponential pulse shape as an approximation of this deeper chain, and tests the saturation part on digitized Na-22 pulses.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Claimed recovery of the bi-exponential model is not quantitatively consistent with the reported fits: Eq. (25) predicts a ~44 ns decay tail, while Appendix AD reports τ_fit_d≈118 ns, and no LTI mechanism in Eqs. (22)-(27) can lengthen the slowest pole to that value.","rationale":"I read the paper in good faith and checked the main mathematical chain. The dynamic occupancy solution (14) does satisfy the ODE (13), the integration-by-parts current expression (19) is consistent, and the EMG convolution (24) is correct. The numerical verifications in Section III are internal consistency checks and are not the concern. The load-bearing weakness is the mapping from the asymptotic theory to the experimental fitting model. The reader correctly noted that the fitted decay time is not explained by the model's slow pole, and that the proxy substitution in Appendix AF tests a phenomenological saturating ODE. My stress-test sharpens this into a quantitative contradiction: Eq. (25), with the stated representative parameters, predicts a dominant ≈44 ns exponential tail; the reported fit value ≈118 ns cannot be produced by any LTI convolution with the fast SiPM poles or by Gaussian TTS smoothing. Because this discrepancy touches the central claim that the conventional bi-exponential is recovered as a controlled reduction, it is more load-bearing than the fit-proxy issue alone. However, the discrepancy is in the empirical validation and parameter mapping, not in the internal algebra of the forward derivation, so the appropriate verdict remains conditional rather than reject. I therefore do not change the reader's conditional verdict.","tokens_in":55354,"tokens_out":9544,"duration_ms":104713,"concrete_test":"Recover the fitted (τ_fit_d, τ_fit_r) for the 100 medium-amplitude pulses and compare against a control simulation: generate Eq. (25) with the Section III parameters, convolve through the same numerical pipeline, and fit the 5-parameter bi-exponential (38). If the control fit returns a decay constant near 44 ns rather than ~118 ns, the claimed reduction does not reproduce the data. As a second check, fit the medium-amplitude cohort directly with Eq. (25) with τeff, τd, τp1, τp2, and σ_TTS fixed to independently characterized values and only amplitude and t0 free; if the fit cannot reach the observed tail or the residual AIC deteriorates sharply, the model is not validated as an end-to-end forward predictor.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that the conventional bi-exponential pulse model 'emerges as a controlled asymptotic reduction' is not sustained by the numbers reported in the validation. In Section II-E, Eq. (27) is obtained from Eq. (25) by taking τ_r→0 and applying a dominant-pole approximation, so the late-time decay is set by τeff=τ/(1−εaη)=44.4 ns with the representative parameters of Section III. A convolution with the single-cell response (16), whose poles are 0.5, 1.5, and 4 ns, cannot shift the dominant tail to a slower time constant: LTI convolution of exponentials preserves the slowest pole, and the Gaussian TTS kernel (7) does not alter the exponential tail. Appendix AD nonetheless states that the fitted template (38) has τ_fit_d≈118 ns and τ_fit_r≈36 ns, with the former attributed to 'broadening' of τeff and the latter to TTS and electronic bandwidth. Neither attribution follows from the stated model: a Gaussian width would need to be on the order of tens of nanoseconds to turn a 0.5–4 ns pole into a 36 ns rise, far beyond σ_TTS=40 ps or a 16 GHz front end. The validation therefore fits the phenomenological template (38) and the proxy ODE (47), where β absorbs α·k_trans·C_gen·κ_EMG (AF.3), rather than the first-principles linear kernel (25). The AIC comparison may show that one empirical saturating envelope beats another, but it does not establish that the model's controlled reduction is the bi-exponential actually observed, which is an essential part of the central claim.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a fully analytical forward model of LYSO-SiPM pulse formation that couples finite thermalization, recursive self-absorption, depth-dependent optical transit-time spread (TTS), SiPM microcell occupancy and recovery, and multi-exponential single-cell current response. In the linear regime the model yields closed-form exponentially modified Gaussian (EMG) superpositions (Eq. 25); in saturation it gives state-dependent integral solutions (Eqs. 14, 19). The paper further derives a Poisson Fisher-information lower bound on coincidence timing resolution, reporting about 100 ps FWHM for a representative 511-keV LYSO-SiPM configuration (Eq. 37). It claims that the conventional bi-exponential pulse model emerges as a controlled reduction of this cascade, and it presents an experimental Na-22 validation in which a 12-parameter saturation model is favored by AIC over an 11-parameter ringing-corrected bi-exponential template in 100/100 high-amplitude and 98/100 medium-amplitude pulses.","tokens_in":55801,"tokens_out":3882,"duration_ms":45132,"significance":"If the central claim were fully substantiated, the paper would provide a valuable unifying framework: pulse shape, saturation distortion, variance envelopes, and timing limits would follow from one analytical cascade, with the ubiquitous bi-exponential fit model derived rather than assumed. The algebraic derivation in Sections II and the appendices is careful and largely self-consistent, with several non-trivial checks (photon-number conservation, the tau_rec to infinity static limit, the EMG normalization, and explicit singularity resolutions). The paper is also unusually transparent about its own limitations, including the binary-recovery dead-time bias, the Campbell variance bound, and the Gaussian TTS restriction. However, the experimental validation and the claimed recovery of the empirical bi-exponential model contain load-bearing inconsistencies: the fitted decay constant is roughly 118 ns while the model's effective decay pole is 44.4 ns, and the AIC comparison is affected by a convergence filter and by fitting the saturation coupling parameter beta rather than testing an independently fixed prediction.","major_comments":[{"comment":"The claim that the conventional bi-exponential pulse model 'emerges as a controlled asymptotic reduction' is not quantitatively supported by the reported fits. With the representative parameters of Section III, Eq. (27) predicts a late-time decay governed by tau_eff = 44.4 ns, yet Appendix AD reports tau_fit_d about 118 ns. No LTI mechanism in the model can lengthen the slowest pole: convolving with the single-cell response (16), whose poles are 0.5, 1.5, and 4 ns, or with the Gaussian TTS kernel (7), preserves the slowest exponential pole rather than moving it to 118 ns. The attribution in Appendix AD to 'SiPM recovery and readout bandwidth' is not a mechanism within Eqs. (22)-(27), and a 16 GHz front end or 40 ps TTS cannot produce a 36 ns fitted rise time. The paper needs either to add and justify an explicit slow mechanism (e.g., a slow scintillation component or a long recharge-related pole) that enters the linear kernel, or to substantially restate the claim as a shape-family reduction without parameter identification.","section":"Section II-E and Appendix AD; Eq. (25)-(27) vs. Table I"},{"comment":"The abstract's statement of 'validation on 10,000 directly digitized Na-22 pulses' overstates the analysis: the detailed model comparisons are performed on a random 100-pulse sample and two 100-pulse cohorts selected for amplitude, not on the full 10,000-pulse dataset. More seriously, the fitting procedure in Section IV-C applies a convergence filter that excludes dynamic-model fits whose RMSE exceeds 1.5 times that of the ringing-corrected model and replaces them until each cohort contains 100 'converged' fits. The reported 100/100 AIC win count therefore does not describe the full high-amplitude cohort as acquired; it describes a selected subset after excluding failures. The authors should report the raw cohort sizes, the number of excluded pulses, and the AIC comparison both with and without the replacement procedure, and should adjust the abstract and conclusions accordingly.","section":"Abstract and Section IV; Tables I-II and fitting procedure"},{"comment":"The experimental saturation comparison tests an empirical proxy model rather than the first-principles ODE (13). In Eq. (47), the driving term beta g(t-t0) replaces alpha r_ph(t;z), and Appendix AF explicitly states that beta absorbs alpha, k_trans, C_gen, and a shape-matching factor kappa_EMG. The fitted beta values (0.052 ns^-1 versus 0.023 ns^-1) are then presented as evidence for the ODE's amplitude-dependent saturation, but this is circular: fitting the coupling parameter and confirming that it increases with amplitude validates the chosen empirical envelope, not the first-principles coupling. The paper should either calibrate beta independently from the forward model, or explicitly frame the AIC result as a comparison of two empirical fitting families and remove the claim that it provides 'strong statistical evidence that the saturation ODE captures physical structure'.","section":"Section IV-C and Appendix AF; Eq. (47) and Eq. (AF.3)"}],"minor_comments":[{"comment":"The phrase 'validation on 10,000 directly digitized Na-22 pulses' should be changed to reflect that the source dataset contains 10,000 pulses while the statistical model comparisons use selected 100-pulse cohorts.","section":"Abstract"},{"comment":"The sentence describing the full 10,000-pulse acquisition as 'the source dataset for this section' is clear, but the paper should state explicitly that Tables I and II report only on the sampled cohorts, so that readers do not infer full-dataset statistics.","section":"Section IV-A"},{"comment":"The ringing-corrected model conflates optical TTS and electronic bandwidth into a single Gaussian width sigma, while Appendix AD later assigns the fitted rise time tau_fit_r about 36 ns to these effects. A brief quantitative justification of how sigma and tau_fit_r relate to the stated 40 ps TTS and 16 GHz bandwidth would help the reader evaluate this approximation.","section":"Section IV-B, Eq. (40)"},{"comment":"The 'Best AIC count' rows should include the number of pulses excluded by the convergence filter for each cohort; otherwise the 100/100 and 98/100 counts are ambiguous.","section":"Table II"},{"comment":"The mapping tau_fit_r <- tau_SiPM_d is inconsistent with the stated numerical value: Section III sets tau_d = 0.5 ns, yet the fitted rise time is reported as about 36 ns. This discrepancy should be acknowledged and explained in terms of the actual fitted parameter values.","section":"Appendix AD, Step 4"}],"recommendation":"major_revision","confidential_remarks":"The theoretical machinery is substantial and likely worth publishing after substantial revision, but the central 'controlled reduction to the bi-exponential model' claim and the headline validation statistics both need to be reworked. If the authors cannot reconcile the 118 ns fitted decay with the 44.4 ns model pole, the paper should be recast as a forward-model framework with separate empirical validation, rather than as a derivation of the observed bi-exponential pulse."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The theoretical machinery in this paper is solid and worth reading; the central claim about recovering the empirical bi-exponential pulse is not supported by the paper's own numbers.\n\nWhat's actually new: the authors chain the standard pieces—Marano's single-cell response, the static occupancy closure, Borel crosstalk, Gaussian TTS, and Poisson Fisher information—into one analytical forward model. The closed-form EMG solutions, the exact integral for the recovery ODE, and the CRLB analysis are all useful and, as far as I can tell, algebraically correct. The appendices are unusually thorough; this is a serious modeling effort.\n\nThe soft spots are real and load-bearing. With their representative parameters, tau_eff = 44.4 ns, and Eq. (27) makes the linear pulse tail decay with that pole. Gaussian TTS convolution cannot slow an exponential tail, and the single-cell poles (0.5-4 ns) are faster, not slower. Yet Appendix AD reports fitted decay constants of about 118 ns. There is no LTI mechanism in Eqs. (22)-(27) that produces a 118 ns tail from a 44.4 ns slowest pole. The claim that the fitted decay time is the intrinsic tau_eff 'broadened' is simply wrong in LTI convolution, which preserves the slowest pole. So the paper's flagship 'controlled reduction' story collapses when you check the time constants.\n\nThe validation also overreaches. The abstract says 'validation on 10,000 directly digitized pulses,' but the statistical comparison uses 300 pulses. The convergence filter replaces failed fits until each cohort reaches 100 converged fits, which is cherry-picking without knowing the failure rate. The AIC test between the 12-parameter dynamic model and the 11-parameter ringing-corrected model is a weak discriminator, especially since the dynamic model takes the fitted bi-exponential template as its input; beta absorbs a pile of unmeasured factors, so the comparison is one empirical envelope against another. Reporting that fitted beta scales with amplitude as evidence for the ODE is circular.\n\nThe math and CRLB work should be preserved; the experimental validation and the 'recovery' claim need major rework or a significant reframing. Readers in the ToF-PET and scintillator modeling community will find the EMG and CRLB sections useful, but I would not cite the bi-exponential-recovery claim in its current form. This paper deserves refereeing: it's a real contribution with identifiable, fixable problems. Send it to reviewers, but expect major revision.","headline":"Solid theory; the 'controlled reduction' to the bi-exponential is contradicted by the paper's own fitted decay times, and the validation overclaims the data.","tokens_in":56291,"tokens_out":6415,"would_cite":false,"duration_ms":64480,"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":"Coupled thermalization, optical transit, and SiPM occupancy yield closed-form EMG pulses and a ~100 ps Fisher-information timing bound, with the bi-exponential pulse recovered as a controlled limit.","keywords":["LYSO","SiPM","scintillation pulse modeling","exponentially modified Gaussian","microcell saturation","coincidence timing resolution","Fisher information","time-of-flight PET"],"falsifier":"A DOI-tagged pencil-beam experiment on the same 3.9×3.9×20 mm³ LYSO-SiPM module would settle the transport-and-timing chain: if the measured pulse broadening and coincidence timing as a function of interaction depth do not follow the predicted $\\sqrt{\\sigma_0^2 + k_{\\mathrm{disp}} z}$ scaling and the Fisher-information floor, the cascade model is falsified. A direct single-photon measurement of the light-pulse arrival profile would separately test whether the bi-exponential proxy $g(t)$ matches the true $r_{\\mathrm{ph}}(t)$ shape.","tokens_in":55147,"feed_emoji":"⚡","tokens_out":10362,"duration_ms":90667,"temperature":0.7,"pith_summary":"Existing detector models treat scintillation kinetics, optical transport, SiPM response, and timing statistics as separate problems. This paper argues that a single forward cascade—finite thermalization of the excitation, depth-dependent optical transit-time spread, and microcell activation with recovery—governs both the macroscopic pulse shape and the attainable coincidence timing. In the linear regime the model yields closed-form exponentially modified Gaussian pulses; under saturation it gives state-dependent integral solutions; and the ubiquitous bi-exponential pulse model is derived as a controlled asymptotic reduction of the same cascade. Fitting the dynamic saturation model to 10,000 directly digitized Na-22 pulses, the authors report AIC wins over a matched bi-exponential baseline in 100/100 high-amplitude and 98/100 medium-amplitude pulses. Coupling the dynamic triggering rate to compound Poisson statistics gives a Fisher-information coincidence timing lower bound of about 100 ps FWHM for a reference 511-keV LYSO-SiPM configuration.","feed_headline":"One cascade explains LYSO-SiPM pulses and sets a 100 ps floor","feed_subtitle":"Bi-exponential pulse fits become a derived limit of the full cascade, with saturation verified on 10,000 pulses.","key_machinery":"The carrying object is a three-stage forward cascade: a bi-exponential photon-generation profile $Y_{\\mathrm{mod}}(t)$ formed by convolving the thermalization cascade with the self-absorption-renormalized decay; a Gaussian optical transit-time-spread kernel $f_{\\mathrm{TTS}}(t;z)$ whose width and mean grow with depth of interaction; and a SiPM microcell occupancy ODE whose linear limit is a convolution with the multi-exponential single-cell response. The two identities that make the derivation work are the integration-by-parts reduction of the rate convolution into state-dependent integral forms (Eqs. 17 and 19) and the EMG convolution identity (Eq. 24), which converts each causal exponential kernel into an exponentially modified Gaussian—an exponential tail convolved with a Gaussian—when the transit kernel is applied. These identities let the model stay closed form in the linear regime and reduce to the bi-exponential pulse in the $\\tau_r\\to 0$, dominant-pole limit.","core_discovery":"The central discovery is that the full LYSO-SiPM detection chain—scintillation generation with finite thermalization, recursive self-absorption, depth-dependent Gaussian optical transit spread, microcell occupancy and recovery, and a multi-exponential single-cell current response—can be written as one causal LTI cascade. Its linear-regime output is a superposition of exponentially modified Gaussian terms: convolving each exponential pole with the Gaussian transit kernel produces $\\operatorname{EMG}(t;\\mu_z,\\sigma_z,\\tau)$, and Eq. (25) assembles the poles into a closed-form pulse. The dynamic recovery ODE for the busy-microcell population, Eq. (13), admits a state-dependent integral solution (14), and the macroscopic current in saturation takes the integral form (19), which reduces exactly to the static binomial occupancy current as $\\tau_{\\mathrm{rec}}\\to\\infty$. In the instantaneous-thermalization and dominant-pole limits the kernel collapses to the bi-exponential shape (27), giving a first-principles origin for the empirical pulse model. Coupling the same dynamic triggering rate to non-stationary compound Poisson statistics produces current-variance envelopes and a Poisson Fisher-information bound on coincidence timing, evaluated at about 100 ps FWHM for the reference 511-keV configuration.","pith_inferences":["Editorial inference: if the ~100 ps floor is robust, further coincidence-timing gains for this class of detectors must come from light channels that bypass the slow scintillation tail, such as Cherenkov or prompt emission, because the bound is set by the rising-edge slope of the dynamic triggering rate.","Editorial inference: the same cascade structure should transfer to other scintillator–SiPM pairs (GAGG, BGO, plastic) by re-parameterizing the time constants and cell-response poles; a testable prediction is that the AIC advantage of the saturation model over the bi-exponential grows with the ratio of peak photon flux to microcell count.","Editorial inference: the validation's proxy assumption could be checked directly by single-photon counting the light pulse with a fast photodetector; if the measured arrival profile deviates measurably from the bi-exponential template, the fitted saturation coupling $\\beta$ would need reinterpretation beyond $\\alpha k_{\\mathrm{trans}} C_{\\mathrm{gen}}$."],"forward_implications":["Pulse-fitting and sparse-sampling reconstruction can use the EMG superposition or the saturation ODE instead of an empirical bi-exponential template, gaining a physical parametrization of optical spread, microcell density, and recovery time.","At higher deposited energy the instantaneous microcell occupancy grows, so the model predicts amplitude-dependent peak suppression and tail modification; the AIC comparison on high- and medium-amplitude pulses supports this prediction.","The Fisher-information construction turns detector parameters—light yield, effective decay time, transit-time spread, PDE, dark count, crosstalk—directly into a coincidence timing floor, so the ~100 ps bound can be recomputed for any LYSO-SiPM geometry without Monte Carlo.","The DOI-dependent transport parameters make the same forward model a basis for waveform-based depth-of-interaction estimation, and the framework supplies a joint Fisher-information benchmark for timing and DOI inference."],"supporting_citations":[{"why":"Supplies the self-absorption/re-emission cascade logic used to derive the effective scintillation decay envelope $\\tau_{\\mathrm{eff}}$.","marker":"[4]"},{"why":"Provides the analytical multi-exponential single-photoelectron current response used as the SiPM impulse kernel.","marker":"[5]"},{"why":"Gives the static binomial microcell occupancy model that the dynamic recovery ODE extends.","marker":"[6]"},{"why":"Provides the microcell RC-recovery phenomenology and partial-recharge physics for the recovery term.","marker":"[12]"},{"why":"Establishes the Fisher-information lower bound on scintillation timing that the Poisson CRLB construction builds on.","marker":"[24]"},{"why":"Supplies the Borel crosstalk cascade and excess noise factor used in the compound-Poisson variance model.","marker":"[18]"},{"why":"Provides experimental timing-resolution limits and modern SiPM parameters used for the ~100 ps comparison.","marker":"[27]"},{"why":"Documents the EMG pulse fits whose dominant temporal poles the asymptotic reduction identifies.","marker":"[16]"}],"fun_headline_variants":["Bi-exponential pulses now a derived limit, not a guess","From cascade to pulse: LYSO-SiPM waveform in one hit","100 ps floor revealed by unifying LYSO-SiPM model","One forward model sets timing floor at ~100 ps FWHM","Why bi-exponential fits work: a unified cascade answer"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The validation's load-bearing premise is that the fitted bi-exponential template, after Gaussian smoothing, faithfully represents the true photon arrival rate at the SiPM; if the shape-matching correction $\\kappa_{\\mathrm{EMG}}$ deviates from 1, the fitted saturation coupling $\\beta$ absorbs that mismatch and the AIC comparison validates the proxy model rather than the first-principles occupancy ODE.","fun_headline_variants_meta":{"raw":{"variants":["Bi-exponential pulses now a derived limit, not a guess","From cascade to pulse: LYSO-SiPM waveform in one hit","100 ps floor revealed by unifying LYSO-SiPM model","One forward model sets timing floor at ~100 ps FWHM","Why bi-exponential fits work: a unified cascade answer"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000139,"raw_usage":{"total_tokens":1244,"prompt_tokens":1122,"completion_tokens":122,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":738,"completion_tokens_details":{"reasoning_tokens":50}},"tokens_in":738,"tokens_out":122,"duration_ms":2347,"temperature":1.0,"reasoning_tokens":50,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T22:48:42.274543+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A DOI-tagged pencil-beam experiment on the same 3.9×3.9×20 mm³ LYSO-SiPM module would settle the transport-and-timing chain: if the measured pulse broadening and coincidence timing as a function of interaction depth do not follow the predicted $\\sqrt{\\sigma_0^2 + k_{\\mathrm{disp}} z}$ scaling and the Fisher-information floor, the cascade model is falsified. A direct single-photon measurement of the light-pulse arrival profile would separately test whether the bi-exponential proxy $g(t)$ matches the true $r_{\\mathrm{ph}}(t)$ shape.","supporting_citations":[{"cited_title":"Understanding and simulating SiPMs,","cited_arxiv_id":null,"evidence_quote":"Provides the microcell RC-recovery phenomenology and partial-recharge physics for the recovery term."},{"cited_title":"Novelγ- and X-ray scintillator research: on the emission wavelength, light yield and time response of Ce3+ doped halide scintillators,","cited_arxiv_id":null,"evidence_quote":"Supplies the self-absorption/re-emission cascade logic used to derive the effective scintillation decay envelope $\\tau_{\\mathrm{eff}}$."},{"cited_title":"Accurate Analytical Single-Photoelectron Response of Silicon Photomultipliers,","cited_arxiv_id":null,"evidence_quote":"Provides the analytical multi-exponential single-photoelectron current response used as the SiPM impulse kernel."},{"cited_title":"Characterization of Silicon Photomulti- pliers for PET Imaging,","cited_arxiv_id":null,"evidence_quote":"Gives the static binomial microcell occupancy model that the dynamic recovery ODE extends."},{"cited_title":"The lower bound on the timing resolution of scintillation detectors,","cited_arxiv_id":null,"evidence_quote":"Establishes the Fisher-information lower bound on scintillation timing that the Poisson CRLB construction builds on."},{"cited_title":"Analytical models of probability distribution and excess noise factor of solid state photomultiplier signals with crosstalk,","cited_arxiv_id":null,"evidence_quote":"Supplies the Borel crosstalk cascade and excess noise factor used in the compound-Poisson variance model."},{"cited_title":"Experimental time resolution limits of modern SiPMs and TOF-PET detectors exploring different scintillators and Cherenkov emission,","cited_arxiv_id":null,"evidence_quote":"Provides experimental timing-resolution limits and modern SiPM parameters used for the ~100 ps comparison."},{"cited_title":"A comprehensive model to predict the timing resolution of SiPM-based scintillation detectors: Theory and experimental validation,","cited_arxiv_id":null,"evidence_quote":"Documents the EMG pulse fits whose dominant temporal poles the asymptotic reduction identifies."}],"review_version":1}