{"id":"65249eb4-4626-40d9-8806-753f5c90aa33","arxiv_id":"2607.21657","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A conservation law for epidemic momentum lets a single, well-scaled incidence curve separate the basic reproduction number R0 from the pre-epidemic susceptible fraction.","lead":"By tracking \"epidemic momentum\"—current infections weighted by their remaining ability to transmit—this paper derives a way to estimate, in principle, both pathogen transmissibility and pre-existing immunity from a single epidemic curve. The method works on simulated epidemics, but the headline 1918 Philadelphia numbers (R0≈2.7, ~20% prior immunity) are tied to an assumed death rate and carry no error bars.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"External scale, not the conservation law, is the load-bearing assumption: without a known absolute incidence scale, Eq. (2.10) cannot identify R0 and x− separately, and the Philadelphia z−≈0.20 estimate is a direct function of the assumed CFP.","rationale":"The reader's weakest_assumption exactly matches the point that would sink the abstract's unqualified claim. I checked Eq. (2.10) algebraically: it follows from the stated conservation law and Euler–Lotka tail relation, and the simulation evidence is consistent with the method working when absolute incidence is known. The failure mode is the unknown multiplicative scale. The paper is unusually honest about this, flagging it in §2.4 and §4.2 and providing the CFP lower bound Eq. (3.6), yet the abstract and Philadelphia discussion present R0≈2.7 and z−≈0.20 as estimates from the epidemic curve. Since the empirical numbers are affine functions of the chosen CFP and the admissible CFP range makes z− span nearly the full [0,0.5] interval, the conditional verdict is appropriate; no new concern moves it further. I would keep the verdict at CONDITIONAL, and would ask the authors to either add the CFP sensitivity table or soften the abstract.","tokens_in":15608,"tokens_out":16082,"duration_ms":142528,"concrete_test":"Compute the full CFP sensitivity table for the Philadelphia reconstruction: for CFP = 0.016, 0.0205, 0.025, and 0.031, evaluate Eq. (3.5) and Eq. (2.11) to obtain R0 and z− (holding the deconvolved momentum curve fixed). If z− ranges by more than 0.4 across this interval, the empirical prior-immunity estimate is not identified by the mortality time series alone; a decisive check would then be to require an independent serological or epidemiological constraint on z− and see whether any CFP in Frost's range is consistent. This test directly settles whether the 'single time series' claim holds for the Philadelphia data.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central mathematical step is Eq. (2.10): R0=(1/ŷ)V(1/L−). Given L−=L[g](λ−) and an absolutely scaled incidence curve, this correctly separates R0 from x−. The load-bearing assumption is that ŷ is on the absolute incidence scale. An unknown constant reporting/death scale factor c multiplies every value of the deconvolved incidence and hence multiplies Y and ŷ by c, while leaving λ− and L− unchanged. Equation (2.10) then returns R0/c unless c is known and divided out. The paper acknowledges this in §2.4 and §4.2, but the abstract's 'single epidemic time series' phrasing hides it. In the Philadelphia application the missing scale is supplied by CFP: Eq. (3.5) gives R0≈135×CFP, and Eq. (2.11) gives z−=1−1/(135·CFP·L−). Using L−≈0.464, over the internally admissible range CFP∈[1.6%,3.1%], z− ranges from ≈0 to ≈0.49 and R0 from ≈2.16 to ≈4.19. The point estimate z−≈0.20 is thus essentially an echo of the assumed CFP≃2%, not an independent inference from the time series. This does not invalidate the method for data with known absolute scale, but it makes the empirical headline quantitatively non-robust and requires the advertised claim to be stated as conditional on an external scale.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper claims that epidemic momentum—prevalence weighted by remaining infectious potential—combined with the initial growth rate of an epidemic can identify the basic reproduction number R0 and the pre-epidemic susceptible fraction x− (equivalently prior immunity z− = 1−x−) separately from a single epidemic time series. The method is derived in the renewal-equation framework: Eq. (2.10) gives R0 = (1/ŷ) V(1/L−), where L− is the Laplace transform of the generation-interval distribution at the initial growth rate and ŷ is the peak epidemic momentum; Eq. (2.11) then gives x−. The approach is tested on stochastic SEIR simulations and applied to 1918 Philadelphia influenza mortality. The authors reconstruct incidence from mortality via deconvolution, assume CFP = 2%, and obtain R0 ≈ 2.71 and z− ≈ 0.203. The paper explicitly notes in §2.4 and §4.2 that absolute incidence scale is required, but the abstract and conclusions present the result without this caveat.","tokens_in":15939,"tokens_out":6358,"duration_ms":52859,"significance":"If the method is valid, it is a conceptually useful extension of the Wallinga–Lipsitch approach: the shape of the epidemic curve, through the maximum of epidemic momentum, carries information beyond the initial growth rate, potentially separating transmissibility from prior immunity. The derivation is transparent, the renewal-equation setting is general, and the stochastic simulation tests are a strength. The paper is also explicit about many of its limitations, including the need for an absolute incidence scale. However, the empirical headline is not robust: the Philadelphia estimates are essentially a rescaling of the assumed case-fatality proportion, and no uncertainty quantification is provided. These issues do not invalidate the method, but they require substantial revision of the claims and presentation.","major_comments":[{"comment":"The point estimates R0 ≈ 2.71 and z− ≈ 0.203 are direct functions of the assumed CFP. Eq. (3.5) gives R0 ≈ 135 × CFP, and with L− ≈ 0.464, Eq. (2.11) gives z− = 1 − 1/(135·CFP·L−). The internally admissible range from Frost, CFP ∈ [1.6%, 3.1%], yields R0 between ≈2.16 and ≈4.19 and z− between ≈0 and ≈0.49. Thus the headline 'about 20% already immune' is not an inference from the time series; it is an echo of the assumed CFP ≃ 2%. The paper acknowledges the scale dependence in §2.4 and §4.2, but the Abstract and Conclusions do not. A sensitivity table or figure over the CFP range, and over the generation-interval and deconvolution assumptions, is needed before the empirical claims can stand.","section":"§3.2, Eqs. (3.5), (3.8), (2.11)"},{"comment":"The paper presents no uncertainty quantification for the Philadelphia estimates. The method depends on deconvolution of mortality, estimation of the initial growth rate, the form and parameters of the generation-interval distribution, and the infection-to-death delay distribution, all of which are uncertain. The text in §4.2 that 'an important statistical challenge will be to develop robust confidence intervals' is not a substitute for reporting them or at least a sensitivity analysis in the current application. The simulation study (Figs. 1 and 2) shows point estimates and scatter, but no coverage or interval estimates; without this, 'estimating' R0 and x− is overstated. At minimum, a bootstrap or profile-likelihood analysis should accompany the empirical results.","section":"§3.2, §4.2"},{"comment":"The central claim as stated—that R0 and x− can be inferred 'from a single epidemic time series'—is too strong. As the authors note in §2.4, the absolute scale of incidence is required to identify ŷ. Without a known scale factor, an arbitrary multiplicative constant changes R0 and x− linearly, so the two quantities are not separately identifiable from a relative incidence curve. The abstract and conclusions should explicitly say 'absolutely scaled incidence' or 'given a known incidence scale.' Otherwise the reader may reasonably infer that mortality counts or other relative time series suffice, which is not the case.","section":"Abstract and §4.4"}],"minor_comments":[{"comment":"The notation is confusing: ι(τ) is used both for incidence and for cumulative incidence in 'Writing ι(τ) = ∫ ...'. Use a different symbol (e.g., I(τ) or C(τ)) for cumulative incidence.","section":"Eq. (2.6)"},{"comment":"The appearance of 'xi' without subscripts is ambiguous; use x_i or x− consistently.","section":"Figure 2 caption"},{"comment":"Reference [33] is 'In preparation.' If the nonlinear-incidence extension is not available, it should be cited as a personal communication or removed; otherwise the reader cannot verify the claim in §4.3.","section":"§4.1 and Ref. [33]"},{"comment":"Appendix C is said to be identical to [5, Appendix F]. Consider citing the original rather than reproducing it, or at least make the duplication explicit in the main text.","section":"Appendix C"},{"comment":"The lower bound R0 ≥ 2.16 is scale-independent and useful, but the paper could also note that no upper bound is available from the method without an assumed CFP; this reinforces the need for a sensitivity analysis.","section":"§3.2"}],"recommendation":"major_revision","confidential_remarks":"The methodological core is a consequence of the authors' own unpublished preprint [5]. The present manuscript would be stronger if it included a self-contained derivation of the conservation law, or at least a clear statement that the result is proved in [5] and is not being introduced here for the first time. The most important issue for the editor is the gap between the strong abstract claims and the conditional nature of the empirical results: the Philadelphia estimates are a function of the assumed CFP, and the paper should be revised to present them as illustrative and conditional on that assumption, with sensitivity analysis."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The core claim here is worth taking seriously: if you know the absolute scale of incidence, then the initial growth rate plus peak epidemic momentum separates R0 from prior immunity. The derivation of Eq. (2.10) is clean and I checked the substitutions—they are consistent with the renewal equation and the cited conservation law. The simulation tests support the method under the stated assumptions, and the comparison against the Wallinga-Lipsitch shortcut makes the practical problem clear. This is a genuine step beyond previous identifiability results: Bergström et al. proved non-identifiability under under-reporting, and here the authors show what extra information (peak momentum) is needed. That is a real contribution.\n\nThe paper is also honest about its limitations. Section 2.4 states that the absolute incidence scale is required; Section 4.2 flags the lack of robust confidence intervals; Section 4.3 notes the mass-action and fixed-transmission assumptions. I don't think the authors are overselling the method.\n\nThe soft spot is exactly what the stress-test note says: the empirical headline is a direct function of an external normalization. The abstract's phrasing \"single epidemic time series\" hides the fact that the scale must be known. In the Philadelphia application, Eq. (3.5) gives R0 ≈ 135×CFP, and the z− ≈ 0.20 estimate is essentially the assumed 2% CFP echoed back. Over Frost's own range of 1.6% to 3.1%, z− swings from near zero to nearly half. So the point estimates are not quantitatively robust, even though the method itself is sound. I would have liked to see this sensitivity made explicit in the main text, rather than apparent only after working through the equations.\n\nAlso worth flagging: no error bars, the deconvolution and generation-interval choices are reasonable but add uncertainty, and the code is promised but not shipped. The lower bound R0 ≥ 2.16 is scale-free and useful, and the internal inconsistency they point out in Mills et al. is a nice catch.\n\nThis paper deserves a serious referee. The method is new, the derivation is reproducible, and the simulation evidence is relevant. The revision should focus on rephrasing the claims to make the scale dependence unavoidable, adding sensitivity analyses on CFP and generation interval, and shipping the code. I would bring it to a reading group, and I would cite the method even if the Philadelphia application stays as an illustration.","headline":"The momentum method is real and the derivation checks out, but the Philadelphia numbers are hostage to the assumed CFP, so the paper's value is the method, not the point estimates.","tokens_in":16511,"tokens_out":1148,"would_cite":true,"duration_ms":28520,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["92D30"],"pacs":[],"model":"deepseek-v4-flash","headline":"A conservation law for 'epidemic momentum' lets a single epidemic time series yield separate estimates of the basic reproduction number R0 and pre-epidemic immunity.","keywords":["epidemic momentum","prior immunity","basic reproduction number","renewal equation","conservation law","generation interval","1918 influenza","epidemic inference"],"falsifier":"In an outbreak where pre-epidemic immunity is independently measured (e.g., by serology) and the reporting fraction is known, compute the momentum-based estimate of the susceptible fraction; if it disagrees with the independent measurement beyond the expected simulation error, one of the conservation law's supporting assumptions—mass-action transmission, known generation-interval distribution, or known incidence scale—is violated.","tokens_in":15398,"feed_emoji":"🦠","tokens_out":9866,"duration_ms":72723,"temperature":0.7,"pith_summary":"This paper establishes that fitting an epidemic model to an incidence curve need not conflate pathogen transmissibility with pre-existing population immunity. It introduces a conserved quantity called epidemic momentum—prevalence weighted by each infected person's remaining potential to transmit—whose peak value, combined with the initial growth rate, separates the basic reproduction number R0 from the pre-epidemic susceptible fraction. The authors derive an exact formula, R0 = V(1/L-)/y-hat, where V(u)=u-1-ln u, L- is the integral transform of the generation-interval distribution at the initial growth rate, and y-hat is the peak momentum. They validate the method on stochastic simulations and apply it to the autumn 1918 influenza wave in Philadelphia, estimating R0 ≈ 2.7 and about 20% prior immunity. If correct, this removes a long-standing ambiguity: researchers can estimate true transmissibility and baseline immunity from the same observed curve, provided the absolute scale of incidence is known.","feed_headline":"One epidemic curve yields R0 and prior immunity separately","feed_subtitle":"Reanalyzing Philadelphia's 1918 wave suggests R0 ≈ 2.7 and that 20% were already immune.","key_machinery":"The central object is epidemic momentum, Y(t), a weighted prevalence in which each infected cohort is weighted by its remaining reproductive potential—the upper tail of the intrinsic generation-interval distribution. For SIR and SEIR models this coincides with total prevalence; in general it is a weighted prevalence. The loop-closing identity is the conservation law C(x,y)=y+(1/R0)V(R0 x) with V(u)=u-1-ln u, whose constant value is the maximum momentum y-hat. Because y-hat is directly observable from the incidence history and the generation-interval distribution, it provides the missing information that turns the composite product R0 x- into separately identifiable R0 and x-.","core_discovery":"The central discovery is that the maximum of epidemic momentum, y-hat, is the constant value of a conserved quantity that stays invariant along any epidemic trajectory governed by the renewal equation with mass-action transmission. At the pre-epidemic endpoint, where momentum is zero and the susceptible fraction is x-, evaluating the conserved quantity yields an exact expression for R0 in terms of the initial growth rate and y-hat alone, with no dependence on x-. The effective reproduction number from early growth then gives x- = 1/(R0 L-), and the final size follows from a symmetric relation at the falling tail via the principal branch of the inverse of z e^z. The paper shows the method rec","pith_inferences":["The same conservation law, applied to the falling tail of an epidemic, should reproduce the R0 and x- estimates obtained from the rising tail; using both tails could yield a consistency check that does not require the absolute incidence scale when both tails are observed.","Because the inferred prior immunity depends linearly on the absolute incidence scale, the method could be inverted to estimate an unknown reporting fraction: the scale that makes the implied x- agree with an independent serosurvey would be an estimate of ascertainment.","Since the momentum peak is reached early in the outbreak and before the peak of delayed observations like mortality, the approach is a candidate for real-time estimation of both transmissibility and baseline immunity as an epidemic unfolds."],"forward_implications":["Epidemic time series can identify both R0 and pre-epidemic immunity, not just their product.","For the 1918 Philadelphia autumn wave, the method gives R0 ≈ 2.7 and z- ≈ 0.2, indicating substantial pre-existing immunity, plausibly from the spring herald wave.","Earlier estimates of R0 in Philadelphia that assumed fixed prior immunity are internally inconsistent with that assumption under the same generation-interval distribution, once the momentum relation is applied.","When the absolute incidence scale is unknown, the initial growth rate alone still gives a lower bound R0 ≥ 1/L- (≈ 2.16 for Philadelphia), independent of the reporting fraction.","Peak momentum occurs after peak incidence, so provisional estimates of R0 and prior immunity can be formed before the observed series peaks, especially when incidence is reconstructed from delayed outcomes."],"fun_headline_variants":["One curve yields both R0 and prior immunity","Conservation law separates R0 and immunity","Momentum invariant yields R0 and prior immunity","R0 and prior immunity from a single curve","Single time series disentangles R0 from immunity"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The time series must be measured on an absolute per-capita scale, because the peak momentum used in the analysis scales linearly with that scale; without it the method cannot separately fix transmissibility and pre-existing immunity.","fun_headline_variants_meta":{"raw":{"variants":["One curve yields both R0 and prior immunity","Conservation law separates R0 and immunity","Momentum invariant yields R0 and prior immunity","R0 and prior immunity from a single curve","Single time series disentangles R0 from immunity"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001293,"raw_usage":{"total_tokens":5114,"prompt_tokens":740,"completion_tokens":4374,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":484,"completion_tokens_details":{"reasoning_tokens":4303}},"tokens_in":484,"tokens_out":4374,"duration_ms":24194,"temperature":1.0,"reasoning_tokens":4303,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T09:31:22.264455+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"In an outbreak where pre-epidemic immunity is independently measured (e.g., by serology) and the reporting fraction is known, compute the momentum-based estimate of the susceptible fraction; if it disagrees with the independent measurement beyond the expected simulation error, one of the conservation law's supporting assumptions—mass-action transmission, known generation-interval distribution, or known incidence scale—is violated.","supporting_citations":[],"review_version":1}