{"id":"7be55071-3e2b-47f5-8dc9-675253f6c798","arxiv_id":"2501.06968","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"AAS2RTO is a flexible greedy scheduler that ranks transient candidates by score components, demonstrated to find roughly two SNe Ia per night within half a day of peak brightness for a 1.5-m telescope.","lead":"This paper introduces AAS2RTO, a Python tool that ranks astronomical transients for follow-up by combining broker alerts, lightcurve fits, and telescope visibility. It is designed to help small telescopes catch supernovae at peak brightness in the upcoming LSST survey era.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Peak-timing validation is internal to SALT: the reported 0.5-day yields measure self-consistency, not accuracy against independently determined peak epochs.","rationale":"The reader's weakest_assumption is precisely the load-bearing point. The paper is genuinely useful: AAS2RTO is transparent, modular, publicly available, and its ZTF alert-stream integration is a real step toward LSST-era follow-up. The authors also honestly list the idealizations in the LSST simulation (Sect. 3.4). However, the headline precision and yield numbers are produced by comparing truncated SALT fits with a full SALT fit, and the scheduling factor xpeak depends entirely on t0. Because the mock-LSST test uses SALT-generated fluxes and the input SALT t0 as truth, it cannot catch systematic errors in SALT peak times. This is not a fatal objection to the tool: even a biased t0 can still rank candidates, and the bias may be calibratable. But the specific claim of catching SNe Ia within 0.5 days of true peak requires an external peak estimator. The proposed check is feasible with existing ZTF/TNS data and would settle whether the concern is quantitative or merely theoretical. This supports the reader's conditional recommendation rather than changing it.","tokens_in":29868,"tokens_out":6087,"duration_ms":67652,"concrete_test":"On the ZTF archival sample, select spectroscopically confirmed SNe Ia with dense light curves (at least 20 detections spanning roughly -10 to +30 days relative to maximum) and estimate each peak epoch independently of the AAS2RTO fitting pipeline, for example with SNooPy templates or a non-parametric Gaussian-process/spline fit in the rest-frame B band after cross-filter K-corrections. Then compare these independent peak times with the real-time truncated SALT t0 values that AAS2RTO would have used, and recompute the nightly yield within +/- 0.5 days of the independent peak. If the median offset exceeds 0.3 days or the 68% scatter exceeds 1 day, the claimed precision is internal to SALT rather than absolute, and the reported per-night yield near true peak is overestimated.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing assumption is that the SALT2 zero-phase parameter t0 equals the astrophysical time of peak brightness. In Sect. 3.3.1 the reference value t*_0 is obtained by fitting SALT2 to the full available lightcurve, and the real-time t0 estimates are compared with this same SALT-derived t*_0. The quoted precision of -2.1 to +1.3 days, and the derived N = 2.08 +/- 0.07 SNe per night within 0.5 days of the predicted peak (Sect. 3.3.2, Fig. 9), therefore quantify how well truncated SALT fits agree with the full SALT fit. They do not test whether SALT t0 coincides with the true maximum of the supernova. Since the scoring factor xpeak (Eq. 6) uses t0 to schedule observations, a systematic offset in SALT t0 from host-galaxy reddening, dust, stretch-colour degeneracies, template mismatch, or redshift assumptions would directly shift spectra to the wrong phase. The mock-LSST test in Sect. 3.4 cannot reveal such offsets because its photometry is generated from the same SALT templates and evaluated against the input SALT t0; that input is not an independent peak measurement. Thus the central claim of catching SNe Ia within 0.5 days of true peak brightness is not yet established against an independent peak epoch.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript presents AAS2RTO, an open-source Python tool for prioritizing transient candidates from alert streams (ZTF now, LSST in the future), intended to support spectroscopic follow-up with the Danish 1.54m telescope. The core algorithm is a greedy ranking: each candidate receives a score S = S_base * x_i, where the factors x_i are user-defined and transparent, and the paper describes data ingestion from brokers (fink, ALeRCE, Lasair), ATLAS forced photometry, TNS crossmatching, SALT2/Bazin lightcurve fitting, and site-visibility factors. For the example science case of SNe Ia near peak brightness, the score includes factors xmag, xpeak, xrise and xspan. The paper validates the approach on two years of ZTF/fink archival alerts and on LSST OpSim mock observations, reporting a peak-time precision of -2.1 to +1.3 days (68% interval) and mean yields of N = 2.08 ± 0.07 SNe Ia per night within 0.5 days of the predicted peak (ZTF) and 2.05 ± 0.04 (LSST mock). The central claim is that AAS2RTO can deliver ranked lists of SNe Ia at the right phase for 1.5m-class spectroscopy.","tokens_in":30086,"tokens_out":6907,"duration_ms":70345,"significance":"If the reported yields are taken at face value, AAS2RTO is a practically useful, low-cost component of the LSST follow-up ecosystem, and the transparent multiplicative-factor design and public repository are genuine strengths. The paper is also commendable for stating explicit assumptions and limitations in §3.4, including the idealised nature of the mock LSST alert stream and the assumption that every detection is a valid, correctly classified supernova alert. The main quantitative deliverable—an expected rate of SNe Ia observable near peak—is, however, currently calibrated only against the SALT2 model itself, not against independently measured peak epochs; this limits the strength of the central quantitative claim but does not undermine the software framework. The archival ZTF experiment is internally consistent and provides a reproducible benchmark that will be useful to the community.","major_comments":[{"comment":"The reference peak t*_0 is obtained by fitting SALT2 to the full available lightcurve, while the real-time t0 used in the xpeak factor (Eq. 6) comes from the same SALT2 model fit to truncated data. The quoted 68% interval of -2.1 to +1.3 days therefore measures the agreement between truncated and full SALT2 fits, not the accuracy of the predicted peak relative to the true maximum of the supernova. A systematic offset in SALT2 t0 (from template mismatch, host-galaxy extinction, stretch–colour degeneracy, or redshift assumptions) would shift the scheduled observations away from true peak even when the truncated fit agrees with the full fit. The paper's own Fig. 6 shows a median offset of about -2 days at phase -6 days, so the real-time schedule is systematically early relative to the nominal peak; this propagates directly into the yield N = 2.08 ± 0.07 if 'within 0.5 days' is interpreted as true peak. I request either an external validation against independent peak epochs (e.g., high-cadence photometric maxima from other surveys or template-independent estimates) or a rewording of the claims to state that the quoted precision is an internal SALT consistency measure.","section":"§3.3.1, Eq. (6), Figs. 6–7"},{"comment":"The LSST mock evaluation is generated using SALT3 templates and compared against the simulation-input SALT t0, so it cannot test the systematic offset identified in §3.3.1. In addition, the mock assumes that every photometric detection is a valid alert, is correctly classified as a supernova, and is not affected by host-galaxy masking; the authors acknowledge these limitations in §3.4, but they are load-bearing for the extrapolated LSST yield of 2.05 ± 0.04. I recommend presenting the LSST numbers as an end-to-end test of the pipeline's self-consistency under LSST cadence, not as a prediction of the rate of observations at true peak brightness, or adding an external validation that does not rely on the same SALT templates used for prediction.","section":"§3.4, Figs. 10–11"}],"minor_comments":[{"comment":"The caption reads 'An sketch'; it should read 'A sketch'.","section":"Fig. 1 caption"},{"comment":"There is a duplicated article in 'such as the the one suggested in Bazin et al. (2011)'.","section":"§3.1"},{"comment":"As written, xmag = 10^(0.5*(18.5-m)) is positive for all magnitudes, yet the text states that candidates with m > 18.5 mag are 'flagged so that their final score will be negative'. Please clarify whether the implementation applies a separate sign flag or modify Eq. (5) to include the exclusion.","section":"Eq. (5)"},{"comment":"The magnitude limit is stated inconsistently: §3 gives a faint limit of i < 18.5 mag, Eq. (5) uses the latest ZTF g/r detection, and the Conclusions state r > 18.5 mag. Please harmonize the notation and specify which band is used for the 18.5 mag limit.","section":"§3 and Conclusions"},{"comment":"The phrase 'maximum model g measurement' is ambiguous; it would be clearer to say 'brightest model-predicted g-band magnitude'.","section":"Fig. 7 caption"},{"comment":"The behaviour of the saturation function for xvis/A << 1 is described as 'approaches zero', which is true but could be stated more informatively as '∩ xvis ≈ xvis in that limit', as the text later notes.","section":"Appendix B, Eq. (B.2)"}],"recommendation":"major_revision","confidential_remarks":"The paper fits the scope of A&A methods and the software is released with a public repository, which is a clear strength. The main quantitative claim about catching SNe Ia within 0.5 days of true peak should be either softened to an internal-consistency statement or supported by an independent peak-epoch validation; this is a fixable issue within the scope of a revision. I see no citation or attribution concerns."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a solid, honest engineering paper. The headline is that AAS2RTO is a flexible, broker-agnostic prioritization layer for small-telescope follow-up, and the example science case (SNe Ia near peak) is worked through with real ZTF alerts and mock LSST data. The tool itself is transparent, publicly available, and the authors are commendably upfront about its limitations, including the fact that the greedy dispatch idea is not new.\n\nWhat is genuinely new: the specific factor design for this science case, the SALT t0 convergence analysis on two years of ZTF alerts, and the mock-LSST yield estimates. The paper also does a decent job of situating AAS2RTO among existing schedulers (Steele & Carter, SONG, GOTO, SOXS, BTSbot) and it does not oversell the algorithm.\n\nThe main soft spot is the peak-timing validation. In Sect. 3.3.1, 'truth' is the SALT2 fit to the full lightcurve, and real-time estimates are compared to that same model. So the reported -2.1 to +1.3 day precision, and the N=2.08 per night within 0.5 days of predicted peak, are measures of how well truncated SALT fits agree with the full SALT fit. They are not a check against independently measured peak epochs. The mock-LSST test has the same issue, since the photometry is generated from the same SALT templates. That is a real limitation, but I don't think it is fatal: the paper's central claim is that the tool can rank candidates, and the ranking is based on a reasonable proxy. Still, if the authors claim spectra will be caught near true peak, they should validate t0 against spectroscopically or well-sampled photometric peak times.\n\nOther soft spots are minor: the LSST mock is idealized (no non-Ia contamination, no host galaxy masking, perfect classification), though the authors acknowledge this; the factor choices (xpeak width, amplitude) are hand-set with no sensitivity analysis; and the paper would be easier to reproduce with a pinned commit and processed test data.\n\nWho is this for: anyone planning LSST transient follow-up with a small telescope, and anyone thinking about how to structure a flexible prioritization tool. It deserves a serious referee; I would send it to peer review with a request for the independent t0 validation and reproducibility artifacts. The math is straightforward and appears sound; the citation pattern is appropriate.","headline":"Solid engineering paper whose headline yields rest on a self-consistent SALT validation; recommend peer review with a request for external peak-time checks.","tokens_in":30737,"tokens_out":3865,"would_cite":true,"duration_ms":38882,"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 simple multiplicative score can rank LSST transient alerts well enough that a 1.5-metre telescope catches about two type Ia supernovae per night within half a day of predicted peak brightness.","keywords":["transient follow-up","LSST","Zwicky Transient Facility","type Ia supernovae","candidate prioritisation","greedy scheduling","SALT lightcurve fitting","spectroscopic observations"],"falsifier":"Replay AAS2RTO on archival ZTF alerts for SNe Ia whose peak epochs are known independently (for instance from high-cadence forced photometry interpreted with a different lightcurve model, or from spectroscopic velocity evolution), and count the fraction of ranked candidates that fall within half a day of those independent peak epochs; a catch fraction well below the reported value would show that the internal-consistency assumption is the limiting factor.","tokens_in":29569,"feed_emoji":"🔭","tokens_out":8108,"duration_ms":73739,"temperature":0.7,"pith_summary":"With LSST expected to issue roughly ten million alerts per night, no single small telescope can spectroscopically follow everything; this paper introduces AAS2RTO, a Python prioritisation tool that reduces each candidate to one score and re-ranks the list whenever new alerts arrive. The score is a product of independent factors chosen by the user, so the same engine can serve very different science cases. For the worked example of catching type Ia supernovae near peak brightness with a 1.5-metre telescope, the paper reports that SALT2 lightcurve fits predict the peak epoch to within about $-2.1$ to $+1.3$ days (central 68%), and that replaying archival ZTF alerts yields an average of $\\bar{N}=2.08\\pm0.07$ candidates per night within half a day of the predicted peak. A simulation fed with mock LSST observations gives $2.05\\pm0.04$ candidates per night within half a day of the true peak. The point is that a transparent, cheap ranking rule can keep a small telescope scientifically competitive in the LSST era.","feed_headline":"Two supernovae per night caught within half a day of peak","feed_subtitle":"As LSST floods the sky with alerts, a transparent score ranks which transients deserve a small telescope's spectra.","key_machinery":"The load-bearing object is the AAS2RTO scoring function, a product of user-defined factors $S=S_{\\rm base}\\prod_i x_i$, embedded in a greedy dispatch loop that ingests broker alerts, pre-filters candidates, fits models, computes scores, and emits ranked lists per observatory. For the example science case the factors are $x_{\\rm mag}$ (brightness relative to the limiting magnitude), $x_{\\rm peak}$ (Gaussian proximity in time to the SALT2 peak estimate $t_0$), $x_{\\rm rise}$ (fraction of brightening detections), $x_{\\rm span}$ (a logistic penalty on transients older than the roughly 19-day rise time), and $x_{\\rm vis}$ (a normalised measure of how soon the candidate sets below the minimum altitude). The SALT2 lightcurve fit supplies a single zero-phase parameter $t_0$ on which the peak factor depends; all factors multiply into one rankable number, and the ranked list is recomputed on every alert-arrival loop.","core_discovery":"The paper's central claim is that prioritising transient candidates with a transparent multiplicative scoring function is enough to make rapid spectroscopic follow-up of LSST alerts feasible on a small telescope. Each candidate is scored as $S=S_{\\rm base}\\prod_i x_i$, where every factor $x_i$ encodes one observed or modelled property: brightness, proximity to the predicted peak, whether the lightcurve is still rising, age of the transient, and remaining visibility from the observing site. For the type Ia supernova example, the peak-timing factor $x_{\\rm peak}$ is a Gaussian peaked at the latest SALT2 zero-phase estimate $t_0$ with width $\\sigma=1$ day, so candidates are ranked by how close they are to the predicted peak and how soon they will set. Tested by replaying two years of ZTF alerts through the ranking loop, the paper finds that the SALT2 $t_0$ estimate at the true peak has a central-68% spread of $-2.1$ to $+1.3$ days, yielding $\\bar{N}=2.08\\pm0.07$ candidates per night within half a day of the predicted peak; the same machinery applied to simulated LSST data yields $2.05\\pm0.04$ candidates per night within half a day of the true peak. The authors present these numbers as a demonstration of usable yield for follow-up, not as a measurement of the supernova rate.","pith_inferences":["Editorial extension: if SALT2 $t_0$ carries a systematic offset for real SNe Ia (for instance from host-galaxy reddening, dust, or peculiar velocities), the real-time catch fraction would be lower than the reported internal-consistency value; a calibration against independently measured peak epochs could quantify and correct this.","Editorial extension: the same score design could be applied to other time-critical transients, such as kilonovae or tidal disruption events, by replacing the SALT peak factor with a model appropriate to those lightcurves; the visibility factor already generalises.","Editorial extension: a deep-learning predictor of peak epoch could be inserted as one factor, reserving full SALT fits for the few candidates that reach the top of the list, which would cut per-alert computation as LSST alert rates climb.","Editorial extension: the appendix's proposed saturation of the unbounded visibility factor, $\\hat{x}_{\\rm vis} = \\min(x_{\\rm vis}, A)$, would prevent a setting source from dominating the ranking solely because it is near the horizon."],"forward_implications":["If the claimed precision holds, a small single-object spectrograph can expect about two type Ia supernovae per night within half a day of predicted peak brightness, which is enough to sustain a modest spectroscopy program without large-facility time.","Because the score is recomputed whenever new alerts arrive, the ranked list tracks a transient's evolution and naturally drops candidates as they pass the peak.","The same factor framework transfers to other science cases: new criteria are added as factors, and candidates failing any factor are excluded or rejected rather than manually triaged.","The reported $\\bar{N}=2.08\\pm0.07$ is roughly half the volumetric SN Ia rate expected in the ZTF footprint, which the paper attributes to quality cuts and missed or poorly sampled lightcurves rather than to misclassification.","Replaying the ranking on mock LSST alerts gives nearly the same per-night catch rate as the ZTF replay, suggesting the yield is set mainly by alert cadence and peak-timing accuracy rather than by survey depth."],"supporting_citations":[{"why":"Supplies the SALT2 empirical lightcurve model whose zero-phase parameter $t_0$ is the peak-time estimate used by the scoring factor $x_{\\rm peak}$.","marker":"Guy et al. (2007)"},{"why":"Provides the SALT fitting implementation used to fit lightcurves and recover $t_0$ in both the ZTF replay and the LSST simulation.","marker":"Barbary et al. (2016)"},{"why":"Supplies the broker alert stream and classification annotations used for the archival ZTF test, and the reported average of six days between first classification and peak.","marker":"Möller et al. (2021)"},{"why":"Provides the SALT2 revision and the DES-based distributions of stretch and colour used to generate the mock LSST supernova lightcurves.","marker":"Taylor et al. (2021)"},{"why":"Provides the Pantheon+ stretch-luminosity and colour-luminosity parameters and parameter distributions used to assign absolute magnitudes in the LSST simulation.","marker":"Brout et al. (2022)"},{"why":"Supplies the OpSim baseline survey schedule that determines the simulated LSST visit timestamps, filters, and depths.","marker":"Delgado et al. (2014)"},{"why":"Provides the volumetric SN Ia rate used to draw the number of mock supernovae in the LSST simulation.","marker":"Dilday et al. (2008)"},{"why":"Provides the local SN Ia volumetric rate used to compare the expected ZTF yield with the per-night candidate counts from the replay.","marker":"Frohmaier et al. (2019)"},{"why":"Supplies the Milky Way extinction map applied when fitting SALT2 models to real ZTF lightcurves.","marker":"Schlegel et al. (1998)"}],"fun_headline_variants":["AAS2RTO prioritizes transients, catching 2 SNe Ia per night near peak","Two SNe Ia per night near peak from replaying ZTF alerts","AAS2RTO: transparent scores rank transients for rapid 1.5m follow-up","Replaying ZTF and LSST alerts, greedy scoring finds ~2 SNe Ia nightly","For LSST's flood, a transparent rank picks 2 SNe Ia per night for spectra"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The reported peak-timing precision is measured against the best SALT2 fit to the full lightcurve itself, so if that fitted peak is systematically offset from the true brightness peak for real supernovae (for example because of host-galaxy reddening, dust, or peculiar velocities), the number actually caught within half a day of peak would be smaller than the reported $\\bar{N}=2.08\\pm0.07$ per night.","fun_headline_variants_meta":{"raw":{"variants":["AAS2RTO prioritizes transients, catching 2 SNe Ia per night near peak","Two SNe Ia per night near peak from replaying ZTF alerts","AAS2RTO: transparent scores rank transients for rapid 1.5m follow-up","Replaying ZTF and LSST alerts, greedy scoring finds ~2 SNe Ia nightly","For LSST's flood, a transparent rank picks 2 SNe Ia per night for spectra"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001161,"raw_usage":{"total_tokens":4931,"prompt_tokens":1196,"completion_tokens":3735,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":812,"completion_tokens_details":{"reasoning_tokens":3618}},"tokens_in":812,"tokens_out":3735,"duration_ms":24583,"temperature":1.0,"reasoning_tokens":3618,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T20:49:16.526766+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Replay AAS2RTO on archival ZTF alerts for SNe Ia whose peak epochs are known independently (for instance from high-cadence forced photometry interpreted with a different lightcurve model, or from spectroscopic velocity evolution), and count the fraction of ranked candidates that fall within half a day of those independent peak epochs; a catch fraction well below the reported value would show that the internal-consistency assumption is the limiting factor.","supporting_citations":[{"cited_title":"2016, SNCosmo: Python li- brary for supernova cosmology, Astrophysics Source Code Library, record ascl:1611.017","cited_arxiv_id":null,"evidence_quote":"Provides the SALT fitting implementation used to fit lightcurves and recover $t_0$ in both the ZTF replay and the LSST simulation."},{"cited_title":"2014, in Society of Photo- Optical Instrumentation Engineers (SPIE) Conference Series, V ol","cited_arxiv_id":null,"evidence_quote":"Supplies the OpSim baseline survey schedule that determines the simulated LSST visit timestamps, filters, and depths."},{"cited_title":"A., et al","cited_arxiv_id":null,"evidence_quote":"Provides the volumetric SN Ia rate used to draw the number of mock supernovae in the LSST simulation."},{"cited_title":"E., et al","cited_arxiv_id":null,"evidence_quote":"Provides the local SN Ia volumetric rate used to compare the expected ZTF yield with the per-night candidate counts from the replay."}],"review_version":1}