{"id":"41927684-5c35-469a-9f3a-029810e719ef","arxiv_id":"2504.12907","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":7,"one_line_summary":"An image-shape-based detrending algorithm removes simulated spacecraft jitter from Ariel photometry and spectroscopy, reducing 1-hour noise from 214 ppm to 12 ppm against a 20 ppm requirement.","lead":"Small movements in a space telescope's pointing, called jitter, can corrupt exoplanet light curves. This paper tests an improved algorithm that removes jitter using the shape of the star's image, and shows the simulated Ariel mission can still meet its precision target.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Single-jitter-realization validation: all 128 noise draws reuse one ADS timeline, so the 12 ppm ADEV and the 65 ppm Rp bias are one draw from an unknown jitter/payload distribution, not a verified compliance margin.","rationale":"I agree with the reader's weakest assumption: the single-realization jitter model is the most load-bearing gap. The paper is honest about it—Sec. 2.4 says multiple timelines would increase robustness and Sec. 3.2 explicitly says the bias is one possible realization—but the central compliance claim is exactly what that gap is supposed to support. The algorithm is empirical: Eq. (2) asserts a linear, time-invariant mapping from PSF moments to photometric signal, and the IWCoG Gaussian widths are tuned to maximize correlation on this data. A different jitter realization with different harmonic content or a different non-stationary envelope could in principle violate that linearity or render the tuned widths suboptimal. The 128 noise realizations only quantify random photon/readout noise; they do not sample the systematic dimension. The additional internal tension is that the detrended FGS-1 ADEV is 15-30% above the jitter-free reference and the Rp bias is 65 ppm, so 'photon noise-limited' is an overstatement even within the simulation; however, the core de-jittering result (12 ppm vs 20 ppm requirement) is plausible and well-supported for this realization. For these reasons the reader's CONDITIONAL verdict is appropriate; no change is needed, but the conditionality should explicitly require multi-realization jitter and payload validation before the abstract's compliance claim is accepted.","tokens_in":17624,"tokens_out":6630,"duration_ms":70263,"concrete_test":"Run the full ExoSim2 pipeline on at least 10 independent jitter realizations—ideally new ADS timelines, otherwise synthetic timelines generated from the same PSD with randomized harmonic phases and varying non-stationary RMS envelopes—each with at least 32 independent noise realizations, and with random PRNU/IPRF maps. Then compute the across-jitter distribution of the FGS-1 detrended ADEV at 1 hr and the Rp bias. The concern is settled if the 95th percentile of the detrended ADEV remains below 20 ppm and the Rp bias distribution remains centered near 65 ppm with an across-realization scatter not much larger than the quoted 60 ppm; it lands if a substantial fraction of realizations exceed 20 ppm, or if the mean bias shifts by more than the 20 ppm transmission-spectrum budget, or if the 15-30% excess over the jitter-free reference grows.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central extrapolation is from one ADS jitter timeline to a mission-level compliance claim. Section 2.1 describes a single 10 hr, 1 kHz pitch/yaw time series, with the 2 mas/hr linear drift removed before injection; Section 2.4 states that the 128 noise realizations all use this same jitter timeline and that payload properties are not randomized. The paper's headline numbers are therefore conditional on one jitter/payload draw: FGS-1 detrended ADEV at 1 hr is 11.8 +/- 2.7 ppm versus 9.0 +/- 1.9 ppm for the jitter-free reference (15-30% excess), and the detrended Rp distribution is offset by 65 +/- 5 ppm. The authors themselves attribute this offset to 'residual systematic uncertainties, of which we are seeing only one possible realization' because of the single jitter timeline and non-randomized payload properties (Sec. 3.2). Since the 20 ppm requirement and the 16 ppm systematic contribution to the transmission spectrum are evaluated from this one realization, there is no evidence that the algorithm—including the empirically tuned IWCoG Gaussian widths (2*sqrt(2) for FGS-1, sqrt(2) for Ch0, Sec. 2.3) and the linear moment basis of Eq. (2)—remains compliant for other jitter phase, harmonic content, RMS envelope, or PRNU/IPRF realizations. The abstract's 'photon noise-limited' phrasing is also stronger than the data: a 15-30% excess over the jitter-free reference is not photon-noise-limited in the strict sense, although it is still below the 20 ppm requirement.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a jitter-detrending algorithm for the Ariel mission, based on fitting low-order spatial moments of the PSF (centroid, width, skewness, kurtosis) to photometric and spectroscopic light curves. Using ExoSim2 simulations of the FGS-1 photometer and AIRS-Ch0 spectrometer, driven by a single 10-hour representative ADS jitter timeline and 128 independent noise realizations, the authors show that the FGS-1 Allan deviation at 1 hour improves from 213.9±2.7 ppm (raw) to 11.8±2.7 ppm (detrended), against a 20 ppm requirement and a 9.0±1.9 ppm jitter-free reference. The detrended Ch0 ADEV tracks the reference across the band. The retrieved planet radius from the detrended FGS-1 light curve is biased by 65±5 ppm, which the authors propagate to a 16 ppm systematic in the transmission spectrum. The paper claims that the algorithm leaves Ariel photon-noise-limited across the entire spectrum and fully compliant with mission requirements, while also acknowledging that the bias may reflect the single jitter realization and non-randomized payload properties.","tokens_in":18032,"tokens_out":8670,"duration_ms":83307,"significance":"If the headline numbers hold, the algorithm is a useful, physically motivated contribution to the Ariel data-reduction pipeline: it builds on established Spitzer-style pixel-level decorrelation, uses public simulation tools (ExoSim2, PAOS), and its comparison against a jitter-free reference with 128 noise realizations is a sound way to separate random from systematic residuals. The manuscript is transparent about many of its limitations, including the single jitter timeline and the exclusion of detector and astrophysical effects. However, the central mission-level compliance and photon-noise-limited claims go beyond what the evidence supports: the performance is evaluated on one jitter/payload draw, the detrending parameters are tuned on the same data, and the propagation of the Rp bias into the transmission spectrum is not fully reconciled with the 20 ppm noise floor. These issues are fixable by rewording the abstract and conclusions and by adding targeted robustness checks, so the paper is suitable for major revision.","major_comments":[{"comment":"The mission-level compliance claim rests on a single ADS jitter timeline. Sec. 2.4 states that all 128 noise realizations use the same jitter timeline and that payload properties are not randomized, and Sec. 2.1 removes the 2 mas/hr linear drift before injection; Sec. 3.2 attributes the 65±5 ppm Rp bias to 'only one possible realization' of residual systematics. The reported 11.8±2.7 ppm FGS-1 ADEV, the claimed 3-sigma margin against 20 ppm, and the 16 ppm systematic in the transmission spectrum are therefore one draw from an uncharacterized distribution over jitter phase, harmonic content, RMS envelope, and payload parameters. A concrete test would be to repeat the 128-realization analysis on at least one additional jitter realization and to randomize PRNU/IPRF parameters; until then, the abstract's 'fully compliant with mission requirements' should be qualified as holding for the representative realization studied here.","section":"Sec. 2.4, Sec. 2.1, Sec. 3.2"},{"comment":"The detrending parameters are tuned and evaluated on the same data. The IWCoG Gaussian width factor is 'empirically found to maximize the correlation between the moments and the signal' (Sec. 2.3), and the nuisance coefficients c_k in Eq. (2) are fitted to the same light curves whose residuals are scored in Sec. 3.1. The 128 independent noise realizations protect the random-noise component, but they do not protect against overfitting to the single jitter pattern, since all realizations share the same time-dependent moments. A cross-validation split (fitting nuisance parameters on one half of the timeline and evaluating ADEV on the other, or re-tuning the width factor on a separate jitter realization) would demonstrate that the de-jittering algorithm, rather than the tuning, is responsible for the reported performance.","section":"Sec. 2.3"},{"comment":"The statement 'The equivalent uncertainty on the transmission spectrum is 15 ppm (random) and 16 ppm (systematic). The combined effect is below the required noise floor of the payload (20 ppm)' is arithmetically inconsistent if 'combined' is a quadrature sum, since sqrt(15^2+16^2) ≈ 22 ppm. If the intended meaning is that each component is separately below 20 ppm, that should be stated explicitly and the 'combined effect below 20 ppm' wording removed. This matters because the compliance conclusion in Sec. 3.2 and the abstract depends on this number.","section":"Sec. 3.2"},{"comment":"The abstract's claim that performance 'remains photon noise-limited across the entire Ariel spectrum' is not supported by the paper's own numbers: Fig. 3 shows a 15–30% excess over the jitter-free reference at 1 hr (11.8±2.7 ppm versus 9.0±1.9 ppm), and Sec. 3.2 reports a 65±5 ppm bias in Rp. In addition, the analysis covers only FGS-1 and AIRS-Ch0 (Sec. 2), not VISPhot, FGS-2, NIRSpec, or AIRS-Ch1, so 'across the entire Ariel spectrum' is an extrapolation. I recommend rewording the abstract to state that the algorithm reduces jitter noise close to the photon-noise reference for the two tested channels, with a residual excess and a bias that are small relative to requirements.","section":"Abstract and Sec. 2"},{"comment":"The exclusion of bad pixels, glitches, detector non-linearity, stellar activity, realistic stellar spectra, and limb darkening is asserted to have 'negligible impact on the reduction quality' (Sec. 2.1), and the same effects are deferred to future work in Sec. 4. Since the paper makes a mission-compliance claim, this assumption is load-bearing. At minimum, the authors should either demonstrate insensitivity to a subset of these effects (for example, by adding limb darkening and a realistic stellar spectrum) or explicitly delimit the compliance claim to the idealized setting of this preliminary study.","section":"Sec. 2.1 and Sec. 4"}],"minor_comments":[{"comment":"The text refers to 'Figure 3, yellow curve' when discussing residual correlations, but Fig. 3 contains no yellow series; the detrended curve is green and the white-noise expectation is gray.","section":"Sec. 3.2"},{"comment":"The wavelength axis labels read 'Wavelength [ m]' and should be 'Wavelength [µm]'.","section":"Figs. 4–7"},{"comment":"The aperture is first described as sized to contain 95% of the signal and then as optimized to empirically maximize SNR; these two criteria may not coincide, so the effective criterion should be clarified.","section":"Sec. 2.2"},{"comment":"The paper does not include a data or code availability statement; given that ExoSim2 and PAOS are public, a brief statement on the availability of simulation configurations would aid reproducibility.","section":"Software and Data Availability"},{"comment":"The claim that the 20 ppm requirement is achieved with 3-sigma confidence relies on 11.8+3×2.7 ≈ 19.9 ppm, which is a marginal margin; the paper should state this explicitly rather than implying a comfortable margin.","section":"Sec. 3.1"}],"recommendation":"major_revision","confidential_remarks":"The paper is within scope for a mission-focused experimental-astronomy journal. The main issues are overstatement of the compliance claim and the single-realization evidence base; both are addressable through rewording and added robustness checks. I do not see grounds for rejection, but the abstract and conclusions should not be accepted in their current form."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth reading if you care about Ariel or about detrending exoplanet time series more broadly. The core technique is not new: spatial decorrelation against PSF position and shape goes back to the Spitzer BLISS/PLD/ICA work, and the paper cites that literature properly. What is new is the specific implementation for Ariel—low-order moments from image cross-sections with IWCoG centroids, a no-flat-field variant, and a clean simulation study with 128 noise realizations plus a jitter-free reference. The headline numbers are good: raw ADEV at 1 hr drops from about 214 ppm to 11.8±2.7 ppm, below the 20 ppm requirement with about 3-sigma margin, while the jitter-free reference is 9.0±1.9 ppm. That is a real result for the Ariel pipeline.\n\nThe soft spots are real, but most are disclosed in the paper itself. The validation uses a single ADS jitter timeline; the 128 realizations randomize only photon and readout noise, not jitter phase, harmonic content, RMS envelope, or payload parameters such as PRNU and IPRF. The 65±5 ppm Rp bias is, as the authors say, one possible realization of a systematic. The Gaussian widening factors for the IWCoG weights (2*sqrt(2) for FGS-1, sqrt(2) for Ch0) were empirically tuned to maximize moment-signal correlation on this dataset, so the reported residual is partly in-sample. The abstract's \"photon noise-limited\" claim is stronger than the data: the detrended curve sits 15–30% above the jitter-free reference, and the Rp distribution is broader and offset. None of this sinks the central engineering claim for this simulation—12 ppm clears 20 ppm with margin—but it turns \"verified compliance\" into \"promising, needs more jitter and payload realizations.\"\n\nFitting nuisance coefficients to the same light curves that are then evaluated is standard practice in PLD and BLISS, so I would not call that circularity a flaw; the open question is out-of-sample performance, which a single jitter timeline cannot answer. No code or data are released, though ExoSim2 and PAOS are public, so reproduction is not hopeless. The citation pattern looks appropriate.\n\nWho this is for: Ariel pipeline engineers, mission simulation people, and anyone working on high-precision time-series detrending. It deserves a serious referee. If I were handling it, I would ask for a softer abstract, at least one or two independent jitter realizations, and a quantitative statement about how the 65 ppm bias scales with jitter amplitude.","headline":"A solid, transparent mission-pipeline study whose headline numbers clear Ariel's 20 ppm requirement—but the single-jitter-realization validation and an overclaiming abstract mean the compliance claim is provisional, not proven.","tokens_in":18614,"tokens_out":3167,"would_cite":false,"duration_ms":33984,"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":"Using only the changing shape of the point-spread function, the new algorithm cuts Ariel's one-hour jitter noise from about 214 ppm to 12 ppm, meeting the 20 ppm requirement and approaching the 9 ppm jitter-free floor.","keywords":["Ariel mission","pointing jitter","detrending","Allan deviation","transmission spectroscopy","PSF moments","photon noise","space telescope"],"falsifier":"Take a second, independent jitter timeline with the same RMS level but different harmonic phases and non-stationary envelope, run the same 128-realization pipeline, and compare the detrended FGS-1 Allan deviation at one hour and the transmission-spectrum bias. If the one-hour Allan deviation exceeds 20 ppm, or the planet-radius bias propagates to more than 20 ppm in the transmission spectrum, the central compliance claim is falsified.","tokens_in":2001,"feed_emoji":"🛰️","tokens_out":5921,"duration_ms":123458,"temperature":0.7,"pith_summary":"The paper claims that the photometric and spectroscopic damage from Ariel's pointing jitter can be removed almost completely using nothing but the shape of the stellar image as seen in each exposure. If this holds, the mission's 10-100 ppm stability requirement for exoplanet transit spectroscopy is met without any flat-field calibration, because the algorithm auto-calibrates from the science frames themselves. In simulated 10-hour observations of a bright target, the one-hour Allan deviation of the FGS-1 photometer drops from about 214 ppm raw to about 12 ppm after detrending, against a 20 ppm requirement and a 9 ppm jitter-free reference. The residual planet-radius bias propagates to roughly 16 ppm in the transmission spectrum, within the required noise floor. The same detrending keeps the AIRS-Ch0 spectrometer photon-noise limited across its full wavelength range.","feed_headline":"Jitter noise on Ariel drops from 214 ppm to 12 ppm","feed_subtitle":"PSF-based detrending meets the 20 ppm requirement and removes flat-field calibration.","key_machinery":"The load-bearing object is the set of low-order moments of the PSF light distribution in each science frame: centroid ($x,y$), widths ($w_x,w_y$), skewnesses ($s_x,s_y$), and kurtoses ($k_x,k_y$). The model is $S_\\lambda(t) = f_\\lambda(\\vec X)\\,\\phi_\\lambda\\,\\Lambda_\\lambda(\\vec P,t) + n_\\lambda(t)$, where $f_\\lambda$ is a linear function of the moments with wavelength-dependent but time-independent coefficients, $\\phi_\\lambda$ is the incoming signal, and $\\Lambda_\\lambda$ is the transit light curve. The nuisance coefficients and the transit parameters are fit simultaneously, so the jitter is removed without any flat-field product. Centroids come from an iteratively weighted center-of-gravity estimate, and the spectrometer also registers frames by Fourier-domain cross-correlation before photometry. The moments act as the observable proxy for the time-varying jitter convolution kernel that the non-stationary pointing produces.","core_discovery":"Jitter moves and deforms the point-spread function on a focal plane whose pixels have non-uniform response, which makes the measured flux depend on where the star image sits at each moment. The paper's central claim is that this dependence is captured by low-order moments of the image (centroid position, width, skewness, and kurtosis), so the flux can be detrended by regressing it on those moments while simultaneously fitting the transit. Using a representative jitter timeline and 128 independent noise realizations, the method brings the FGS-1 photometer's one-hour Allan deviation from $213.9 \\pm 2.7$ ppm raw to $11.8 \\pm 2.7$ ppm detrended, matching the $9.0 \\pm 1.9$ ppm jitter-free reference closely enough to meet the 20 ppm requirement at 3-$\\sigma$ confidence. For the spectrometer, the detrended noise tracks the reference at every wavelength, and the raw noise is already at the photon floor at red wavelengths. The only detectable flaw is a small systematic shift of $65 \\pm 5$ ppm in the retrieved planet radius, which the authors propagate to about 16 ppm in the transmission spectrum and judge compliant with requirements.","pith_inferences":["Because the method uses only in-frame spatial information, a testable extension is to apply it to other Nyquist-sampled exoplanet time-series instruments; the same moment regression should remove pointing-induced systematics wherever the PSF shape changes measurably.","The single jitter realization is the main uncontrolled variable: the 65 ppm bias could change sign or amplitude under a different non-stationary envelope or different harmonic phases, so re-running the pipeline on a second prime-contractor realization or on commissioning data would bound this uncertainty.","The slight change in the detrended Allan-deviation slope near 2000 s hints at residual correlated noise that the current statistics cannot firmly confirm; if real, it sets a floor on how much longer integrations can improve stability and might motivate adding higher-order moments or flat-field information."],"forward_implications":["The FGS-1 photometer meets the 20 ppm one-hour stability requirement with 3-sigma confidence, at $11.8 \\pm 2.7$ ppm versus a $9.0 \\pm 1.9$ ppm jitter-free reference.","The AIRS-Ch0 spectrometer stays photon-noise limited across its wavelength range after detrending, and it is already photon-noise limited at red wavelengths even without detrending.","The residual systematic bias in the retrieved planet radius is $65 \\pm 5$ ppm, propagating to about 16 ppm in the transmission spectrum, below the 20 ppm payload noise floor.","De-jittering works without flat-field calibration, so the pipeline auto-calibrates from the science frames and avoids calibration-product biases.","For targets fainter than about K = 6.3, photon noise is expected to dominate and jitter detrending becomes less critical, though uncorrected jitter can still bias retrieved radii."],"supporting_citations":[{"why":"Supplies the end-to-end time-domain simulator that generates the mock Ariel observations used throughout the study.","marker":"[6]"},{"why":"Provides the precedent for using in-frame spatial information to detrend jitter-affected infrared photometry.","marker":"[20]"},{"why":"Provides the subpixel response parameters used to model the focal-plane intra-pixel response.","marker":"[21]"},{"why":"Supplies the intra-pixel response function model whose equation is used in the simulations.","marker":"[22]"},{"why":"Computes the optical PSFs used for the FGS-1 and AIRS-Ch0 focal planes.","marker":"[23]"},{"why":"Generates the transit light curve injected into the simulated observation.","marker":"[24]"},{"why":"Provides the Fourier-domain subpixel registration algorithm used to align the spectral frames.","marker":"[25]"},{"why":"Gives the analytic transit light-curve model that is fit together with the nuisance parameters.","marker":"[26]"},{"why":"Supplies the iterative weighted center-of-gravity algorithm used for high-precision centroid estimation.","marker":"[27]"},{"why":"Defines the Allan deviation metric used to judge noise at the one-hour timescale.","marker":"[29]"}],"fun_headline_variants":["Ariel jitter cut 18x to hit 20 ppm goal","Jitter-busting algorithm gets Ariel to 12 ppm","Ariel noise tamed: 214 to 12 ppm via PSF","Ariel de-jittered to photon-noise floor","Ariel jitter fixed with PSF detrending"],"cache_read_input_tokens":20480,"weakest_assumption_plain":"That the single representative jitter timeline used in all simulations behaves like the real spacecraft's pointing jitter over a 10-hour observation, including its slow rise-and-fall envelope and harmonic content; all noise and bias numbers are computed for that one realization, and the paper does not randomize payload properties.","fun_headline_variants_meta":{"raw":{"variants":["Ariel jitter cut 18x to hit 20 ppm goal","Jitter-busting algorithm gets Ariel to 12 ppm","Ariel noise tamed: 214 to 12 ppm via PSF","Ariel de-jittered to photon-noise floor","Ariel jitter fixed with PSF detrending"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000397,"raw_usage":{"total_tokens":2131,"prompt_tokens":1050,"completion_tokens":1081,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":666,"completion_tokens_details":{"reasoning_tokens":993}},"tokens_in":666,"tokens_out":1081,"duration_ms":11750,"temperature":1.0,"reasoning_tokens":993,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T12:19:44.629634+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a second, independent jitter timeline with the same RMS level but different harmonic phases and non-stationary envelope, run the same 128-realization pipeline, and compare the detrended FGS-1 Allan deviation at one hour and the transmission-spectrum bias. If the one-hour Allan deviation exceeds 20 ppm, or the planet-radius bias propagates to more than 20 ppm in the transmission spectrum, the central compliance claim is falsified.","supporting_citations":[{"cited_title":"Repeatability and Accuracy of Exoplanet Eclipse Depths Measured with Post-Cryogenic Spitzer","cited_arxiv_id":"1601.05101","evidence_quote":"Provides the precedent for using in-frame spatial information to detrend jitter-affected infrared photometry."},{"cited_title":"Optics Letters 33(2), 156 (2008) https://doi.org/10.1364/ OL.33.000156","cited_arxiv_id":null,"evidence_quote":"Provides the Fourier-domain subpixel registration algorithm used to align the spectral frames."},{"cited_title":"In: Korotkova, O","cited_arxiv_id":null,"evidence_quote":"Supplies the iterative weighted center-of-gravity algorithm used for high-precision centroid estimation."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines the Allan deviation metric used to judge noise at the one-hour timescale."}],"review_version":1}