{"id":"c1b9616f-143f-486d-9ee9-1a1ec20d1026","arxiv_id":"2509.05128","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A six-parameter fit of a nova's brightness decline separates recurrent from classical P-class novae and flags V2860 Ori as a possible hidden repeater.","lead":"Astronomers fit three straight lines to the fading light of 30 nova explosions, then used the six fitted numbers to separate known repeat-exploding novae from one-time ones. The same numbers back the already-suspected KT Eri as a repeater and point to V2860 Ori as a possible future repeater worth checking.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"V2860 Ori is a training-set object; its 'prediction' as a recurrent nova is in-sample and not validated by leave-one-out, unlike KT Eri.","rationale":"The reader identified the purity of the classical-nova training labels as the weakest assumption. That concern is real: if some of the 22 P-class CNe are unrecognized RNe, the boundary is biased. However, the paper's own example of KT Eri shows an awareness of leave-one-out validation for that object, yet no such validation is reported for V2860 Ori, which is the paper's only new predictive claim. This is a more immediate and more specific flaw: it does not depend on external label contamination, only on the internal logic of the fitting procedure. The concrete test is straightforward and directly settles whether the V2860 Ori suggestion has any out-of-sample basis. The hidden treatment of censored/missing values is also concerning and should be clarified, but the in-sample circularity is the primary blocker. If the leave-one-out check for V2860 Ori passes, the claim becomes credible; if it fails, the paper's main novelty dissipates, even though the separation of the known samples might still hold. Therefore the verdict remains conditional on this additional analysis, matching the reader's overall assessment but for a different reason.","tokens_in":12865,"tokens_out":6287,"duration_ms":68601,"concrete_test":"Refit the logistic regression exactly as described (same features, same solver) after removing V2860 Ori from the training set. Use the resulting coefficients to compute F for V2860 Ori; determine which side of the boundary it falls on. Repeat this leave-one-out procedure for every object (all 22 CNe and all 14 RN outbursts) to assess boundary stability and report the leave-one-out cross-validated classification accuracy. If V2860 Ori moves to the CN side when excluded, the published near-boundary placement is an artifact of in-sample fitting and the paper's candidate claim fails; if it remains on the RN side, the claim is supported.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The paper's central claim includes a new candidate: V2860 Ori, which the classifier places near the RNe/CN boundary and is suggested as a possible recurrent nova (Section 4.2 and Conclusions). However, V2860 Ori is one of the 22 classical novae used to fit the logistic regression boundary. Its position relative to the fitted hyperplane is therefore an in-sample result, not a prediction. The paper explicitly excludes KT Eri when deriving its boundary (Section 4.3) but does not do the same for V2860 Ori. With a small sample (22 CNe, 14 RN outbursts), a single training point can exert strong leverage on the fitted coefficients; the near-boundary placement may be an artifact of its own label. The manuscript also fails to state how missing m3 values ('none' for 6 CNe and 3 RN outbursts) and lower-limit α2 values (marked '>') were handled in the logistic regression, which further undermines confidence that the reported coefficients are stable. The most load-bearing, testable concern is therefore the in-sample status of V2860 Ori: without a leave-one-out refit, the 'prediction' is circular.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper addresses the problem of identifying whether a P-class nova light curve belongs to a recurrent nova (RNe) or a classical nova (CNe) using only the optical light-curve morphology. The authors fit three-segment piecewise-linear models to AAVSO visual-band light curves of 22 CNe and 14 outbursts of 8 RNe, extracting six parameters (magnitude offset, two time-scaling factors, three decline slopes). They report correlations of these parameters with white-dwarf mass and mass-accretion rate, then train a linear logistic-regression classifier on five of the parameters. The classifier assigns KT Eri to the RNe side even when KT Eri is excluded from the fit, and the authors note that V2860 Ori lies near the decision boundary and may be a recurrent nova. The paper concludes that a simple linear classifier can separate P-class RNe from CNe based solely on the light curve.","tokens_in":13109,"tokens_out":6756,"duration_ms":71324,"significance":"If robust, this would be a valuable, cheap diagnostic from a single outburst and would connect light-curve morphology to underlying system parameters. The KT Eri leave-one-out check is a useful sanity test, and the proposed nova super-remnant search for V2860 Ori is a falsifiable prediction. However, the validation is currently too thin: no cross-validated accuracy, no treatment of missing/censored parameters, and the V2860 Ori statement is based on a training-set point. With a sample of only 36 data points and 5 predictors, the risk of overfitting is non-trivial. The paper's core idea is promising and worth publishing after these load-bearing issues are addressed.","major_comments":[{"comment":"V2860 Ori is a training-set object, not an independent prediction. In §4.2 the authors state that 'V2860 Ori is lying at the very boundary of the CNe and RNe' and interpret this as 'indicative of the possibility of its recurrence.' However, V2860 Ori is one of the 22 CNe used to fit the logistic-regression boundary (Table 1; §2). Its coordinate relative to the fitted hyperplane is an in-sample geometric statement, not an out-of-sample classification. The paper performs a leave-one-out refit for KT Eri (§4.3) but not for V2860 Ori. With 22 CN training points, a single point can exert substantial leverage on the small linear model. Please provide a leave-one-out or bootstrap assessment of the boundary position for V2860 Ori, and report whether it remains on the RNe side when its own label is withheld. This is needed to support the concluding prediction.","section":"§4.2 / Fig. 3 / Table 1"},{"comment":"Missing and censored parameters are not accounted for in the classifier. Table 2 lists m3 = 'none' for six CNe (V838 Her, V368 Sct, V1229 Aql, DD Cir, V444 Sct, V1368 Cen) and three RN outbursts (CI Aql 2000, IM Nor 2002, V745 Sco 2014), and alpha2 is reported only as a lower limit for several objects (e.g., V838 Her '> 1.143', V1229 Aql '> 26.125'). The text does not explain how these enter the LogisticRegression in §4.2. Since the classifier uses {alpha1, alpha2, m1, m2, m3}, missing values must have been dropped, imputed, or otherwise handled; the choice can change the fitted coefficients and the boundary. This is not a cosmetic issue: it affects the reproducibility and stability of the central separation. State the treatment explicitly and, ideally, show that the conclusions are robust to it (e.g., by excluding rows with missing m3 and refitting).","section":"§3.1 / Table 2 / §4.2"},{"comment":"No quantitative validation of the classifier is reported. Section 4.2 gives the fitted coefficients but no accuracy, completeness, false-positive rate, or confusion matrix, and no cross-validation. With 36 data points and 5 predictors (and some with missing entries), the reported RNe/CNe separation in Fig. 3 could be optimistically biased. The KT Eri leave-one-out check is only one point. Please report leave-one-out cross-validated classification metrics (or at least a confusion matrix and balanced accuracy), and a bootstrap or permutation assessment of the boundary. This is necessary to support the claim that the method 'successfully distinguishes' RNe from CNe.","section":"§4.2"},{"comment":"The training labels are assumed to be clean, but the paper itself cites evidence against that assumption. The 22 CNe are taken from Schaefer (2022) as ground truth (§2). Pagnotta & Schaefer (2014), cited in the Introduction, showed that recurrent novae can masquerade as classical novae, and the same concern applies to the present sample. If a few of the 'CN' training objects are actually unrecognized RNe, the fitted hyperplane is systematically biased. The authors should at least discuss the direction and magnitude of this risk; a concrete test would be to remove the most CN-like borderline objects and re-fit, or to treat labels as uncertain in the logistic regression.","section":"§2 / §4.2"}],"minor_comments":[{"comment":"The text describes p = 0.09 and p = 0.10 as 'statistically significant.' Please state the adopted significance threshold or phrase these as marginal correlations.","section":"§4.1"},{"comment":"The decomposition after the logistic-regression equation has sign inconsistencies: with a1 = 1.28, a2 = -1.04, and a6 = -5.05, the time-scaling component should be 1.28*alpha1 - 1.04*alpha2 - 5.05, but the text prints -1.28*alpha1 + 1.04*alpha2 + 5.05. Please correct or clarify.","section":"§4.2"},{"comment":"Please define 'none' and '>' in the caption, and give the units of m1/m2/m3 and alpha1/alpha2. Also specify whether the alpha2 lower limits are used as point values in the analysis.","section":"Table 2"},{"comment":"There are several typographical errors, e.g., 'address this--the presence' and 'thisthe' in the Abstract/Introduction. A careful proofread is needed.","section":"General"},{"comment":"The pwlf Python library is used but no reference or version is given. Please cite it or provide enough algorithmic detail to reproduce the breakpoint fits.","section":"§3.1"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The short version: the six-parameter piecewise-linear description of P-class nova light curves plus a linear classifier is a genuinely new combination, and the KT Eri leave-one-out check is a real, honest test. But the V2860 Ori 'possibility of recurrence' is not a prediction: V2860 Ori is in the training set, so its position near the boundary is expected, not independent. Also, the paper never says how it handled m3 = none and lower-limit alpha2 values in the logistic regression, which matters for trusting the reported coefficients.\n\nWhat's actually new: Bode (1987) used decline timescales and amplitudes and failed on T Pyx; the six-parameter parameterization with α1, α2, m1, m2, m3 captures the plateau shape in a way that visibly separates most CNe from RNe in Figure 3. The correlations with white dwarf mass (m1, m2, m3 positive; α1, α2 negative) are sensible and supported by the expected physics, though the sample is small and the p-values have multiple-comparison issues. KT Eri's placement on the RN side, with the boundary refit without it, is a fair in-sample test and lines up with the super-remnant detection, so that part is credible.\n\nSoft spots, in rough order:\n\n1. The V2860 Ori statement. The abstract and conclusions call it a 'possibility'/'prediction'. Since V2860 Ori is one of the 22 classical novae used to fit the classifier, its position is an in-sample geometry fact, not an out-of-sample prediction. The KT Eri check shows how to do this properly; they just didn't do it for V2860 Ori. Easy fix: re-fit with V2860 Ori held out and report where it lands.\n\n2. Missing data and censored values. Table 2 has m3 = none for 9 rows and alpha2 as '>' for several more. The paper doesn't say how these entered the logistic regression. If they were dropped or imputed, the coefficients could shift. This needs to be stated.\n\n3. Multiple outbursts of the same nova (RS Oph x5, U Sco x2, T Pyx x2) are treated as independent samples. That inflates the effective sample size and can bias the decision boundary toward the more frequently observed objects. A clustered CV or at least a per-nova split would be more convincing.\n\n4. The training labels come from Schaefer (2022), and Pagnotta & Schaefer (2014) showed that unrecognized RNe lurk among 'classical' novae. That doesn't kill the result, but it means the boundary is only as good as the labels.\n\nWho it's for: anyone working on nova demographics, WD mass/ accretion rate censuses, or SN Ia progenitor searches. It's a useful screening tool, not a finished method. It deserves peer review — the flaws are fixable and the core idea is sound enough to be worth referee time. I'd accept if the authors address the in-sample issue and the missing-data handling.","headline":"A six-parameter light-curve classifier that plausibly separates P-class CNe from RNe and correctly places KT Eri; V2860 Ori is an in-sample boundary flag, not a prediction.","tokens_in":13673,"tokens_out":3537,"would_cite":true,"duration_ms":33789,"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":"A linear classifier built from five optical light-curve shape parameters can identify recurrent novae from a single outburst, the paper argues.","keywords":["recurrent novae","classical novae","P-class light curves","light-curve morphology","white dwarf mass","accretion rate","KT Eri","V2860 Ori"],"falsifier":"A deep H-alpha imaging search around V2860 Ori for an extended nova super-remnant—the shell frequent eruptions are expected to sweep out—would directly test the central prediction: a clear detection would confirm the recurrent classification, while a sensitive null detection would undercut the boundary's placement of this object. More generally, a newly confirmed recurrent nova with a recurrence interval under 100 years that falls on the classical side of the fitted boundary would falsify the current separation.","tokens_in":12697,"feed_emoji":"🌟","tokens_out":5411,"duration_ms":54592,"temperature":0.7,"pith_summary":"Recurrent novae are usually recognized only after a second eruption, which can take decades to arrive. This paper asks whether the shape of a single P-class optical light curve—one with a plateau—already carries enough information to tell a recurrent nova from a classical one. The authors model each outburst with six parameters: a magnitude offset, three segment slopes, and two time scalings relative to a reference nova. They then show that recurrent and classical novae occupy separated regions of this parameter space, and they encode the boundary with a simple linear classifier. If correct, the method would let astronomers flag recurrence candidates from one well-observed outburst, and it would make KT Eri a recurrent nova on light-curve morphology alone, with V2860 Ori as a testable borderline case.","feed_headline":"Six light-curve parameters separate recurrent from classical novae","feed_subtitle":"A linear boundary on plateau slopes and time scales flags KT Eri as recurrent and targets V2860 Ori for follow-up.","key_machinery":"The central object is the five-parameter vector x = {α1, α2, m1, m2, m3}, where m1, m2, m3 are the slopes of the three linear segments of a P-class light curve and α1, α2 are the durations of the pre-plateau and plateau phases scaled to the reference nova V1974 Cyg. A piecewise-linear fit identifies the breakpoints and slopes; a logistic-regression hyperplane F(x) = 1.28α1 − 1.04α2 − 14.90m1 + 272.02m2 + 113.15m3 − 5.05 then separates recurrent from classical novae. The same parameters show significant Pearson correlations with white dwarf mass and accretion rate, connecting the empirical light-curve morphology to the physical drivers of recurrence.","core_discovery":"The paper establishes that P-class recurrent and classical novae are separable using only the morphology of their optical light curves. Each P-class light curve is fit as three linear segments—pre-plateau, plateau, and post-plateau—yielding slopes m1, m2, m3 and two time-scaling factors α1, α2 relative to a reference nova. In this parameter space, recurrent novae have systematically higher decline rates and shorter plateau time scalings, consistent with their higher white dwarf masses and accretion rates. The authors construct a linear decision boundary from these five parameters and show that it cleanly separates the 8 recurrent novae from the 22 classical novae in their sample. The boundar","pith_inferences":["Extension beyond the paper: the same five-parameter description could be tried on S-class recurrent novae such as T CrB and V3890 Sgr, whose smooth declines lack a plateau, to see whether a generalized morphology space keeps the two classes separated.","Extension beyond the paper: the distance of a nova from the decision boundary might be usable as a priority ranking for follow-up monitoring, archival plate searches, and super-remnant imaging.","Extension beyond the paper: with a larger sample, the plane could be calibrated to produce recurrence-time estimates rather than binary class labels, if the mapping from morphology to white dwarf mass and accretion rate becomes tight enough."],"forward_implications":["A single well-sampled P-class outburst can flag a nova as a recurrence candidate without waiting for a second eruption.","KT Eri's recurrent status follows from its light curve morphology independently of the multi-diagnostic argument in earlier work.","V2860 Ori becomes a concrete prediction: deep searches for a nova super-remnant around it should reveal one if the classification is right.","The measured correlations tie slopes and time scalings to white dwarf mass and accretion rate, opening a route from light-curve shape to underlying stellar parameters."],"supporting_citations":[{"why":"Supplies the sample of P-class classical and recurrent novae, their distances, and the classification used as training labels.","marker":"Schaefer (2022)"},{"why":"Defines the P-class light-curve taxonomy and provides the literature breakpoint values adopted when automated fits diverge.","marker":"Strope et al. (2010)"},{"why":"Establishes the statistical association between plateau light curves and recurrent novae, and warns that recurrent novae can masquerade as classical novae.","marker":"Pagnotta & Schaefer (2014)"},{"why":"Provides the white dwarf masses and mass accretion rates used for the correlation analysis.","marker":"Shara et al. (2018)"},{"why":"Supplies the theoretical framework connecting high white dwarf mass and accretion rate to short recurrence timescales.","marker":"Nomoto et al. (2007)"},{"why":"Presents the case for KT Eri being a recurrent nova and provides the light-curve data used in the analysis.","marker":"Schaefer et al. (2022)"},{"why":"Reports the observed nova super-remnant around KT Eri, supporting the recurrent classification that the paper recovers independently.","marker":"Shara et al. (2024)"}],"fun_headline_variants":["Five light-curve parameters separate recurrent from classical novae","Plateau morphology distinguishes recurrent novae from classical","Light-curve model flags KT Eri as recurrent nova","Parametric analysis of light curves identifies recurrent novae","Plateau shape predicts V2860 Ori as recurrent nova candidate"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The 22 training novae labeled classical are assumed to be genuinely non-recurrent; if some are actually recurrent novae that have not yet been seen to repeat, the fitted boundary shifts and the verdicts on KT Eri and V2860 Ori no longer follow.","fun_headline_variants_meta":{"raw":{"variants":["Five light-curve parameters separate recurrent from classical novae","Plateau morphology distinguishes recurrent novae from classical","Light-curve model flags KT Eri as recurrent nova","Parametric analysis of light curves identifies recurrent novae","Plateau shape predicts V2860 Ori as recurrent nova candidate"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000308,"raw_usage":{"total_tokens":1650,"prompt_tokens":849,"completion_tokens":801,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":593,"completion_tokens_details":{"reasoning_tokens":722}},"tokens_in":593,"tokens_out":801,"duration_ms":8708,"temperature":1.0,"reasoning_tokens":722,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T05:35:06.298889+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A deep H-alpha imaging search around V2860 Ori for an extended nova super-remnant—the shell frequent eruptions are expected to sweep out—would directly test the central prediction: a clear detection would confirm the recurrent classification, while a sensitive null detection would undercut the boundary's placement of this object. More generally, a newly confirmed recurrent nova with a recurrence interval under 100 years that falls on the classical side of the fitted boundary would falsify the current separation.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the statistical association between plateau light curves and recurrent novae, and warns that recurrent novae can masquerade as classical novae."}],"review_version":1}