{"id":"efac7544-a5ce-4e8e-a2ab-7e1772a21866","arxiv_id":"2507.15830","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"A new open-source pipeline, AutoWISP, automates calibration, astrometry, photometry, and detrending for DSLR images, achieving sub-percent precision in a Sony camera case study.","lead":"AutoWISP is a new software pipeline that turns ordinary DSLR camera photos of the night sky into precise, three-color light curves of stars, with a case study showing about one percent or better precision. It aims to let citizen scientists with modest cameras do exoplanet follow-up and variable-star science that currently needs professional telescopes.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Sub-percent MAD after EPD/TFA is computed in-sample; a holdout test is needed to rule out overfitting.","rationale":"The reader identifies generalization to citizen-science data as the weakest assumption. I agree that is an important gap, but I see a more immediate load-bearing issue: the sub-percent precision statistic itself may be optimistically biased because the detrending models are fit and evaluated on the same light curves. If EPD/TFA overfit, then even the case-study dataset does not demonstrate sub-percent precision. This matters because the abstract presents sub-percent precision as an achieved result, not as a generalizable potential. The paper provides no cross-validation, no injected-signal recovery test, and no explicit statement that target stars are excluded from the TFA template set. The open-source code and external TESS comparison for five eclipsing binaries are genuine strengths, but they do not by themselves validate the numerical precision claim. The manual curation issue from Section 4.1 compounds the concern, since the in-sample result benefited from hand-picked good frames. The proposed holdout cross-validation is cheap and decisive: if the out-of-sample MAD is comparable to the in-sample value, the claim is credible; if it degrades, the claim needs qualification or revision. I therefore maintain the reader's CONDITIONAL verdict: the paper is a useful contribution, but the headline precision figure requires independent validation before acceptance as stated.","tokens_in":22044,"tokens_out":7372,"duration_ms":82307,"concrete_test":"Split the 2017 Sony-α7R II dataset by dither pattern (D25 vs D26) or by a contiguous date boundary; fit magnitude fitting, EPD, and TFA coefficients on the first half and apply the fitted corrections unchanged to the second half. Compute the MAD for Gaia G<8 bright stars in the held-out half. If the median hold-out MAD exceeds about 10.9 mmag (1% flux), the in-sample sub-percent precision claim is not supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim of sub-percent photometric precision rests on the MAD scatter reported in Figures 5-8 after applying magnitude fitting, EPD, and TFA. These three detrending steps are fitted to the same light curves on which the MAD is then computed, so the residual scatter is not an out-of-sample measure. Section 2.7.2 describes TFA as subtracting a linear least-squares fit of template light curves from each star, but the paper does not state that the target star is excluded from the template set, nor does it bound the number of templates relative to the number of epochs. If a star appears in its own template set, its post-TFA residual is artificially near zero; even without that, a sufficiently rich template set can absorb photon noise and slow astrophysical variability, biasing the MAD low. Section 5.2 further notes that the best result was obtained without PSF/PRF modeling, so the quoted precision depends on a particular configuration and on manually curated frames (Section 4.1). Because the abstract claims 'sub-percent photometric precision' without qualification, and no cross-validation, injected-signal recovery, or error bars on the MAD are provided, the headline precision figure is not yet demonstrated to be robust even for this dataset.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"AutoWISP is a Python pipeline for automated reduction of wide-field DSLR color images into three-color light curves, extending the authors' earlier AstroWISP tool. The pipeline performs calibration, Gaia-based astrometry, PSF/PRF and aperture photometry, iterative ensemble magnitude fitting, light-curve construction, and EPD/TFA post-processing. The case study uses a Sony α7R II camera with a 135 mm f/2 lens on the HAT10 telescope, covering three fields with two dither patterns and about 19,000 object images. The paper reports per-channel and combined-channel median absolute deviation (MAD) as a function of Gaia G magnitude, claims sub-percent photometric precision, and shows phase-folded light curves of five eclipsing binaries, including color-dependent eclipse depths for TX UMa and GW UMa. The manuscript is primarily a technical description of the software and its validation on a single dataset.","tokens_in":22259,"tokens_out":3967,"duration_ms":41333,"significance":"AutoWISP addresses a real gap: an open, cross-platform, automated pipeline for citizen-science DSLR photometry that handles Bayer masks and yields simultaneous three-color light curves. The paper's strengths include publicly available code on GitHub/PyPI/Zenodo, self-describing HDF5 intermediate products, a modular design that preserves photometry at each processing stage, and a case study that recovers known eclipsing-binary signals, including a color-dependent system. If the sub-percent precision claim survives out-of-sample validation, the tool would be a useful contribution to exoplanet follow-up and stellar variability studies with consumer cameras. However, the precision evidence in the current manuscript is computed on the same data used to fit the detrending models, and the dataset is manually curated and from a single camera/lens combination; the headline claim is therefore not yet established as stated.","major_comments":[{"comment":"The central precision claim is computed on the same light curves used to fit the detrending models. Magnitude fitting (§2.5) builds the reference from the same frames, EPD (§2.7.1) fits polynomials to each light curve, TFA (§2.7.2) derives templates from the same field, and the channel-combining weights in §4.2 are chosen to minimize the MAD on the same light curves. The MAD values in Figures 5-8 are therefore in-sample residuals; they do not by themselves demonstrate that the pipeline reaches sub-percent precision on unseen or new observations. Please provide a holdout or cross-validation test, such as fitting EPD/TFA on one subset and evaluating MAD on an independent subset, or injecting synthetic signals and checking the recovered amplitude and scatter, and quote the resulting MAD with error bars.","section":"Section 4.2, Figures 5-8"},{"comment":"The TFA description does not state whether the target star is excluded from the template set or how the number of templates compares with the number of epochs. If a star contributes to its own template basis, its post-TFA residual is artificially suppressed; even without self-inclusion, a template set that is large relative to the number of epochs can absorb photon noise and slow astrophysical variability. Please specify the template-selection rule, the maximum number of templates used per light curve, and any safeguard that ensures the target is held out.","section":"Section 2.7.2"},{"comment":"The quoted precision is derived from one camera/lens combination on one mount, with frames manually curated each night (Section 4.1: 'We manually checked each night's images to determine which frames were worthy of processing'), and with the best configuration found to ignore the PSF/PRF model because the PSF was highly variable. This makes it difficult to assess whether sub-percent precision generalizes to typical citizen-science observations. Please quantify the effect of the manual frame selection, for example by reporting what fraction of frames was rejected and how the MAD changes under a less stringent selection, and state the number of nights, frames per field, and stars per magnitude bin that enter each MAD point.","section":"Sections 4.1 and 5.2"},{"comment":"The abstract's 'sub-percent photometric precision' is unqualified. The figures show MAD for a limited magnitude range, with combined-channel MAD near 10 mmag for bright stars in some fields but not all; the precision floor is not characterized, and the scintillation-limit estimate in Section 4.2 is explicitly approximate. Please state explicitly the magnitude range and the per-field/per-dither conditions under which sub-percent precision (MAD < 10 mmag) is achieved, and provide an uncertainty estimate for each MAD value.","section":"Abstract and Section 4.2"}],"minor_comments":[{"comment":"The text contains several typographical errors, including 'recieve' in Section 2.1.2 and inconsistent citation formatting for the AAVSO reference ('AAV 2023' in Section 1.1).","section":"Throughout"},{"comment":"Period values differ between the text and figure captions for GG Dra (2.2695484 d versus 2.269548 d) and for TX UMa (3.06333(2) d in the text versus 3.063337 d in the caption); please harmonize these values.","section":"Figures 9-13 and text"},{"comment":"The caption states that yellow points show 'before EPD (only magnitude-fitting)', but the colors are defined only in the caption; adding a legend or labels directly in the figure would improve readability.","section":"Figure 5"},{"comment":"The paper does not state how many eclipsing binaries were found in the catalog cross-match or how the five presented systems were selected; a sentence on the search completeness and selection criteria would help.","section":"Section 4.3"},{"comment":"Several references are incomplete or non-standard, such as 'AAV 2023', 'mission team', and the HDF Group entry; these should be formatted consistently with the journal style.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is a software description with a strong headline precision claim. The main risk is that the quoted sub-percent precision is an in-sample statistic after magnitude fitting, EPD, and TFA, with no holdout validation; this is fixable within the scope of the paper by adding a cross-validation or signal-injection analysis. I do not see grounds for rejection, but the abstract currently overstates what is demonstrated. The paper fits the journal's scope, and the public availability of the code is a genuine asset."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: AutoWISP is a real, usable software contribution, and the case study is credible up to a point. The sub-percent precision claim, however, is not yet demonstrated as robust: the MAD values come from light curves after magnitude fitting, EPD, and TFA were fit to those same light curves. No cross-validation, held-out data, or error bars on the scatter are provided, and the TFA section doesn't say whether the target star is excluded from the template set. That's a fixable issue, but it's load-bearing for the abstract's claim.\n\nWhat's genuinely good: the code is public, the pipeline is modular and documented, and the authors show a complete reduction from raw DSLR frames to detrended light curves. The TESS comparison for five eclipsing binaries is a nice external validation of the shapes and periods, and the three-color light curves of TX UMa and GW UMa show a clear wavelength dependence that would be hard to get with CCDs. The paper is also honest about the manual frame curation and about the PSF modeling failure (Section 5.2), which I trust more than a glossy success story.\n\nSoft spots, in proportion: first, the precision floor is uncharacterized—scintillation and saturation limits are discussed, but the measured MAD combines photon noise, detrending systematics, and possible overfitting. Second, the core photometry (PSF/PRF, aperture) is in an unpublished AstroWISP paper, so a referee can't check the most important algorithmic steps. Third, the dataset is a single camera/lens combination from one site, manually curated; the pipeline's robustness for typical citizen-science data is untested. These don't wreck the paper, but they should be addressed before the headline claim is accepted.\n\nBottom line: this is a serious software paper that deserves refereeing. I'd ask for a holdout validation or simulated signal injection, release of the reduced light curves, and at least a summary of AstroWISP's photometric equations. Then it would be a solid addition for the citizen-science/time-domain community. A desk reject would be wrong.","headline":"Useful open-source pipeline, but the sub-percent precision headline rests on in-sample detrending and needs a holdout check before it will convince.","tokens_in":22877,"tokens_out":1795,"would_cite":true,"duration_ms":19011,"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":"An automated pipeline converts ordinary DSLR color snapshots into sub-percent-precision three-color light curves of stars.","keywords":["DSLR photometry","citizen science","three-color photometry","light curves","aperture photometry","trend filtering","eclipsing binaries","automated pipeline"],"falsifier":"Take a set of raw DSLR frames from several amateur observers at different sites, run AutoWISP exactly as released with no manual frame rejection, and measure the combined-channel MAD for stars with Gaia G between 6 and 9; if the bright-star MAD exceeds roughly 10 millimagnitudes, or if the recovered eclipse phases of known binaries mismatch their published ephemerides, the claim of sub-percent precision for citizen-science data would be refuted.","tokens_in":21816,"feed_emoji":"🔭","tokens_out":5447,"duration_ms":52315,"temperature":0.7,"pith_summary":"AutoWISP is an automated software pipeline that takes raw color images from consumer DSLR cameras and turns them into high-precision light curves in three separate color channels. The paper's central claim is that this pipeline, built on the earlier AstroWISP photometry engine, achieves sub-percent photometric precision on wide-field citizen-science-style observations, demonstrated with a Sony α7R II camera attached to a HATNet telescope mount. The case study covers 19,210 object images across three fields, and after magnitude fitting, external-parameter decorrelation, and trend filtering, bright stars reach median absolute deviations near or below about 10 millimagnitudes in the combined channels. The same data recover the shapes, periods, and wavelength-dependent eclipse depths of five known eclipsing binaries, showing that simultaneous three-color photometry from a single consumer camera is scientifically usable.","feed_headline":"DSLR night-sky photos yield sub-percent stellar light curves","feed_subtitle":"AutoWISP turns ordinary camera color images into precise three-color light curves that recover known eclipsing binaries.","key_machinery":"The load-bearing mechanism is a photometry model that accounts for sub-pixel sensitivity—including the Bayer mask that makes each DSLR super-pixel a 2×2 arrangement of red, two green, and blue filters—and integrates it into both PSF/PRF fitting and aperture photometry. On top of that sits an iterative ensemble magnitude-fitting step that calibrates each frame against a master reference built from the previous iteration, and then EPD and TFA remove trends correlated with external parameters or common to many stars. The paper's own finding that PSF modeling was counterproductive on this dataset makes the sub-pixel-sensitivity-aware aperture photometry, rather than the PSF model, the component that actually carries the precision.","core_discovery":"The authors claim that a fully automated processing chain—calibration with per-pixel error tracking, Gaia-based astrometric registration, sub-pixel-sensitivity-aware aperture photometry, iterative ensemble magnitude fitting, external parameter decorrelation (EPD), and trend filtering (TFA)—can extract light curves from consumer color camera images that are precise enough for real science. On the Sony α7R II dataset, they report that the pipeline achieves sub-percent photometric precision in the combined color channels for bright stars, with the D26 dither pattern performing slightly better than D25. They also show that modeling the point-spread function did not help: the best photometry came from treating each pixel's illumination as uniform, which they attribute to the short 30-second exposures and a dithering pattern designed for three-minute exposures causing the PSF/PRF to vary from image to image. The method recovers the shapes, periods, and color-dependent eclipse depths of five known eclipsing binaries, thereby demonstrating that simultaneous three-color photometry from a single DSLR is a viable observational tool.","pith_inferences":["A concrete testable extension would be to run AutoWISP on a multi-site civilian dataset without manual frame selection and compare the bright-star MAD to the roughly 10 millimagnitude benchmark; if the automated quality-control steps cannot replace the hand curation, the pipeline may need a robust frame-selection module.","The absence of a detected transit in this dataset is not evidence against the pipeline, since the authors note signals were either too weak or transits did not occur during the observations; a dedicated test on a known transiting exoplanet with a predicted ephemeris would directly probe transit-depth sensitivity.","If sub-percent precision holds broadly, then color light curves from DSLRs could serve as a low-cost complement to space-based photometry, especially for temperature and atmospheric studies of eclipsing binaries where the wavelength dependence is the signal.","The PSF modeling failure hints that pixel-level sensitivity correction may matter more than the point-spread model in short-exposure untracked data; an interesting follow-up is to test whether improved guiding restores a benefit to PSF/PRF fitting."],"forward_implications":["The same pipeline can be adopted by Project PANOPTES, which the paper states will use AutoWISP to produce fully processed light curves across its network of low-cost robotic telescopes.","Because the light curves are produced separately in red, green, and blue channels, any observed variability carries color information, which can help distinguish astrophysical effects like eclipse-depth wavelength dependence from instrumental systematics.","The pipeline's per-pixel error tracking and self-documenting HDF5 files mean results are reproducible without re-running the processing.","If the precision generalizes, citizen scientists could contribute photometric follow-up of exoplanet transit candidates and variable stars, extending time baselines beyond what single surveys like TESS can cover.","The authors find that aperture photometry with a flat illumination assumption beat PSF/PRF fitting, implying that for short-exposure DSLR data, avoiding PSF modeling may be the more robust choice."],"supporting_citations":[{"why":"Supplies the PSF/PRF fitting and aperture photometry engine that AutoWISP wraps and automates.","marker":"(Penev et al. 2025)"},{"why":"Prior DSLR photometry with the same HATNet unit and lens; establishes the observational setup and the feasibility context this work builds on.","marker":"(Zhang et al. 2016)"},{"why":"Defines the Trend Filtering Algorithm used as the final post-processing step to remove common-mode trends from light curves.","marker":"(Kovács et al. 2005)"},{"why":"Introduces External Parameter Decorrelation, the method used to remove correlations with atmospheric and instrumental parameters.","marker":"(Bakos et al. 2010)"},{"why":"Supplies the calibration and master-flat field procedures inherited from the HATSouth project.","marker":"(Bakos et al. 2013)"},{"why":"Provides the fistar source-extraction tool used to find point sources before astrometric matching.","marker":"(Pál 2012)"},{"why":"Provides the Astrometry.net initial plate solution that the Gaia-based iterative astrometric refinement starts from.","marker":"(Lang et al. 2010)"},{"why":"Supplies the catalogue positions, magnitudes, and source properties used for astrometric registration and light-curve association.","marker":"(Gaia Collaboration et al. 2016, 2023)"}],"fun_headline_variants":["AutoWISP turns DSLR snaps into sub-percent light curves","Sub-percent photometry from ordinary color camera images","Three-color light curves from a single DSLR camera","AutoWISP pipeline: sub-percent photometry for citizen science"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the sub-percent precision measured on this one hand-curated Sony α7R II dataset, where the authors manually checked every night's images to remove cloudy or smeared frames, will also appear in typical citizen-science observations that arrive without such careful screening.","fun_headline_variants_meta":{"raw":{"variants":["AutoWISP turns DSLR snaps into sub-percent light curves","Sub-percent photometry from ordinary color camera images","Three-color light curves from a single DSLR camera","AutoWISP pipeline: sub-percent photometry for citizen science"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000801,"raw_usage":{"total_tokens":3500,"prompt_tokens":901,"completion_tokens":2599,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":517,"completion_tokens_details":{"reasoning_tokens":2530}},"tokens_in":517,"tokens_out":2599,"duration_ms":18717,"temperature":1.0,"reasoning_tokens":2530,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T15:23:04.043449+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a set of raw DSLR frames from several amateur observers at different sites, run AutoWISP exactly as released with no manual frame rejection, and measure the combined-channel MAD for stars with Gaia G between 6 and 9; if the bright-star MAD exceeds roughly 10 millimagnitudes, or if the recovered eclipse phases of known binaries mismatch their published ephemerides, the claim of sub-percent precision for citizen-science data would be refuted.","supporting_citations":[{"cited_title":"J., et al","cited_arxiv_id":null,"evidence_quote":"Supplies the PSF/PRF fitting and aperture photometry engine that AutoWISP wraps and automates."}],"review_version":1}