{"id":"b1477cc9-ce08-4e75-a575-ff80c95c4e43","arxiv_id":"2506.02775","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":8,"one_line_summary":"A power-law model with six per-pixel parameters describes image persistence in Euclid's NISP detectors with residuals of a few electrons, but the accuracy is mostly measured on the fitting data.","lead":"This paper presents an empirical model of image persistence in Euclid's infrared detectors, built from ground tests and checked against in-flight calibration data. It uses a power-law decay with fluence-dependent parameters to predict residual ghost images that can contaminate Euclid's cosmological measurements.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Flight validation is qualitative: §2.7 admits per-detector ground-to-flight differences 'can be significant,' and flight data cannot constrain short-term α/β, so the in-flight few-electron accuracy is unverified.","rationale":"The reader's weakest_assumption identifies the same load-bearing premise: the ground-to-flight transfer of the persistence model. The paper's own §2.7 is the weakest link because it asserts compatibility without a quantitative test and, at the same time, concedes that differences can be significant for some detectors. The reader's CONDITIONAL verdict is appropriate: the modeling approach is plausible and the ground-based characterization is fairly detailed (per-pixel parameters, bias/noise maps across fluences, model selection favoring power-law over exponentials), but the central practical claim—that the model can predict persistence in routine Euclid observations—requires a quantitative flight validation. I do not see an internal inconsistency that would justify rejection; the missing analysis is a well-defined comparison that the authors could add. The β ≤ 1 non-integrability noted in §2.8 is a secondary theoretical loose end, but the paper explicitly flags it as a finite-time-interval limitation and it does not by itself overturn the ground model's descriptive accuracy. The recommended verdict remains CONDITIONAL, so no change from the reader's assessment is needed.","tokens_in":13530,"tokens_out":6393,"duration_ms":72253,"concrete_test":"Perform a blind in-flight validation: for each of the 16 NISP detectors, take the ground-calibrated per-pixel model (Eqs. 2–4 with ground-fitted a1,b1,c1,a2,b2,c2), predict the persistence charge in each of the 15 dark exposures following each of the four PV flat-field levels described in §2.7, and compare with the measured integrated charges. Report per-detector bias and RMS in electrons, and per-detector ratios α_flight/α_gd and β_flight/β_gd with uncertainties. If any detector's bias exceeds ~10% of the persistence signal or the RMS exceeds the few-electron target, the transfer fails; if all 16 pass, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that the model is usable for Euclid flight observations depends on transferring ground-characterized parameters to flight. Section 2.7 states that flight-derived α and β are 'on average compatible' with ground values but that differences 'can be significant for some detectors'; no statistical test, threshold, or per-detector numbers are given. Moreover, flight data are available only as integrated fluxes over 87 s and 550 s exposures (MACC modes), so the short-time decay parameters α and β that dominate persistence in the immediately next NISP exposure are not independently constrained by flight data; the fitted flight parameters therefore carry substantial degeneracies. The summary (§5) asserts 'no significant change' in flight, but this sits in tension with the unquantified 'significant' differences acknowledged in §2.7. Consequently, the 'accuracy of a few electrons' claimed in §5 has only been demonstrated on ground and in-sample (§2.6), and the in-flight predictive accuracy is unverified.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents an empirical image-persistence model for the 16 H2RG detectors of the Euclid NISP instrument. Ground characterisation data at 85 K with flat-field stimuli between 5,000 and 95,000 e are used to fit a power-law decay current I(S,t)=alpha(S)(tau/(t-t0+tau))^beta(S), with fluence-dependent coefficients alpha(S)=(a1+b1*S)(1-exp(-S/c1)) and beta(S)=a2*(1+S/b2)^c2, giving six parameters per pixel. The model is used to predict persistence ramps in MACC mode; the bias and noise are estimated from residuals over 15 repeated sequences, and flight Performance Verification data are compared with ground results. The paper also reports persistence above saturation, a pattern inversion at saturation, and a long-term change of detector state after saturating stimuli.","tokens_in":13829,"tokens_out":6728,"duration_ms":60590,"significance":"The model is directly relevant for Euclid operations: persistence correction or masking is needed for weak-lensing and photometric measurements, and the paper documents per-detector and per-pixel variability for all 16 flight detectors. The main strengths are the large ground dataset, the per-pixel approach, and the explicit attempt to compare with in-flight PV data. However, as it stands the central accuracy claim is supported only by in-sample residuals, and the flight comparison is qualitative, so the paper's operational conclusion requires additional quantitative validation.","major_comments":[{"comment":"The statement that a power law fits better than an exponential or a sum of two or three exponentials is not supported by any reported fit statistics. Please provide a quantitative model comparison (e.g., chi-square per degree of freedom, AIC/BIC, or residual r.m.s. per candidate model) for a representative set of pixels and fluences; without this, the choice of Eq. (2) is not established.","section":"§2.5, Eq. (2)"},{"comment":"The quoted accuracy (\"bias ranges from about 5% to 10%\", \"median noise lower than 8e r.m.s.\") is computed from residuals against the same 15 sequences used to fit alpha_i and beta_i in §2.5. This is an in-sample validation. An out-of-sample test (e.g., leave-one-sequence-out, or fit on a subset of fluences and predict the remaining ones) is needed before the Summary's \"accuracy of a few electrons\" can be accepted.","section":"§2.6"},{"comment":"The flight verification is only qualitative. The text says flight-derived alpha and beta are \"on average compatible\" with ground values but that differences \"can be significant for some detectors\", and no statistical test, per-detector values, or uncertainties are given. Because flight data are integrated over 87 s and 550 s MACC modes, the short-term decay parameters are not independently constrained in flight. Please quantify the ground-to-flight comparison per detector and state explicitly what the flight data can and cannot constrain.","section":"§2.7 and §5"},{"comment":"The parameter tau is introduced to avoid divergence at t=t0 but its treatment is not specified: it is not stated whether tau is fixed a priori, fitted per pixel, per fluence, or per detector. If fitted, tau is degenerate with alpha and beta; if fixed, its value and justification should be stated. The same applies to the definition of t0.","section":"§2.5, Eq. (2)"}],"minor_comments":[{"comment":"The title contains a missing space: \"forEuclid\" should read \"for Euclid\".","section":"Title and header"},{"comment":"The notations UTR(286) and UTR(76) are used without definition; please define the readout mode before first use.","section":"§2.1"},{"comment":"The MACC(ng,nf,nd) notation is defined in the footnote but used in the main text; please ensure the definition is given at first use in the text.","section":"§2.6 footnote"},{"comment":"The sentence \"indicates the the time constants of capture\" contains a duplicated article; also, \"in agreement with suggestions of [19] while [11] proposes\" is a run-on sentence that should be rephrased.","section":"§2.8"},{"comment":"The sentence \"The signal-dependent persistence model derived in this work for stimuli below saturation that describes persistence signals with the accuracy of a few electrons\" lacks a main verb and should be rephrased.","section":"§5"},{"comment":"The caption contains a typo: \"bout 5000\" should read \"about 5000\".","section":"Figure 7 caption"}],"recommendation":"major_revision","confidential_remarks":"The paper is a useful instrument-characterisation contribution for Euclid, and the scope is appropriate for the journal. The main obstacles are quantitative rather than conceptual: the power-law choice needs a model-comparison table, the accuracy claim needs an out-of-sample validation, and the flight comparison needs per-detector statistics. Once these are supplied, I expect the paper to be acceptable. There is also a title typo that should be corrected."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here is my read.\n\nThe paper's real value is the dataset and a few specific observations, not the headline accuracy claim. The durable change in detector state after saturating stimuli (§4.3) is genuinely new: a detector does not return to baseline even after six hours in the dark, and the sign of the signal/persistence drift flips. That is a concrete, falsifiable finding for the H2RG community. The per-detector persistence maps and the fluence-dependent α(S) and β(S) parametrization applied to all 16 NISP flight detectors are useful calibration products. The paper does not ship code, so the claim burden rests entirely on the analysis.\n\nThree soft spots matter. First, the claim that a power law beats a sum of exponentials is asserted without fit statistics or a comparison table – surprising for a paper built on model choice. Second, the 'accuracy of a few electrons' in the Summary is computed in §2.6 as residuals against the same 15 sequences used to fit the per-fluence α_i and β_i. That is in-sample error, not predictive error, and the paper admits the bias is 5–10% of the persistence signal – which for high fluences is not 'a few electrons.' Third, the flight validation in §2.7 is qualitative: 'on average compatible,' differences 'can be significant for some detectors,' and no thresholds or per-detector numbers are given. The Summary's 'no significant change' sits in tension with that caveat. Because flight data are integrated over 87 s or 550 s, the short-term α and β that dominate the next exposure are essentially unconstrained in orbit. So the central claim that the model predicts persistence in routine Euclid observations is not established by the evidence shown.\n\nWho should read it: people building NISP persistence masking or correction, and anyone modeling H2RG traps. It is a mission-critical calibration paper and deserves serious review, but it needs revision before the accuracy claim can stand. A referee should ask for model-comparison statistics, an out-of-sample test (leave-one-sequence-out would do), quantitative flight agreement bounds, and reconciliation of §2.7 with §5. The above-saturation result alone justifies careful reading.","headline":"Solid H2RG persistence data and a genuinely new above-saturation effect, but the few-electron accuracy claim is in-sample and flight agreement stays qualitative.","tokens_in":14901,"tokens_out":4937,"would_cite":true,"duration_ms":49450,"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 fluence-dependent power-law model predicts image persistence in Euclid's NISP infrared detectors to within a few electrons, and in-flight data are on average compatible with ground calibration.","keywords":["image persistence","HgCdTe","H2RG","Euclid NISP","infrared detectors","power-law decay","charge trapping","detector characterization"],"falsifier":"Take an in-flight sequence in which a bright star has crossed a NISP detector, run the six-parameter model forward from the actual fluence history, and compare the predicted persistence with the measured residual image in the next pointing; if the residual exceeds a few electrons r.m.s. for one of the detectors flagged as discrepant, the ground-to-flight transfer claim fails.","tokens_in":13375,"feed_emoji":"🛰️","tokens_out":10790,"duration_ms":92862,"temperature":0.7,"pith_summary":"This paper tries to establish that the afterglow left by bright sources in Euclid's 16 near-infrared detectors, an effect called image persistence, can be predicted by a power-law decay whose two parameters depend on how much light the detector just saw. For stimuli below saturation, the model claims agreement with measured persistence at the level of a few electrons, using six fitted parameters per pixel. Such a predictive model matters because persistence contaminates later exposures and, if uncorrected, degrades the photometry and redshifts that Euclid's cosmology program depends on. The paper also claims that persistence parameters measured during in-orbit calibration are on average compatible with the ground-derived values, though the match is not exact for every detector.","feed_headline":"Power law predicts Euclid IR detector persistence to few electrons","feed_subtitle":"Fits ground and in-flight data, making persistence masking practical during Euclid's survey.","key_machinery":"The load-bearing object is the power-law persistence law with fluence-dependent parameters. The form $I(S,t)=\\alpha(S)(\\tau/(t-t_0+\\tau))^{\\beta(S)}$ encodes a broad distribution of trapping time constants, and the two empirical functions $\\alpha(S)$ and $\\beta(S)$ capture how the initial current and the decay slope respond to stimulus strength. The fitting procedure is two-stage: for each fluence level, $\\alpha$ and $\\beta$ are estimated from the integrated persistence current in dark ramps, and then the six constants $a_1,b_1,c_1,a_2,b_2,c_2$ are fitted pixel by pixel to describe the fluence dependence. The parameter $\\tau$ prevents divergence at the end of exposure, and using per-pixel parameters lets the model absorb the nonuniform trap distributions seen in the persistence maps.","core_discovery":"The central claim is that persistence current in NISP's HgCdTe H2RG detectors below saturation follows the power-law decay $$I(S,t)=\\$\\alpha$(S)\\left(\\frac{\\tau}{t-t_0+\\tau}\\right)^{\\$\\beta$(S)},$$ with the fluence-dependent amplitude and index $$\\$\\alpha$(S)=(a_1+b_1S)\\left(1-\\exp(-S/c_1)\\right),\\qquad \\$\\beta$(S)=a_2\\left(1+S/b_2\\right)^{c_2}.$$ Fitting these six parameters per pixel in two stages --- first per-fluence $\\alpha,\\beta$ from the integrated current in up-the-ramp dark exposures, then their stimulus dependence --- yields a model whose predicted persistence ramps have a bias of roughly 5--10% of the persistence signal (a few electrons) and per-pixel noise below 8 e r.m.s. for all stimuli tested. The paper further reports that persistence amplitudes across the focal plane are typically about 1% of the stimulus below saturation, that a power law fits the decay better than one, two, or three exponentials, and that in-flight persistence parameters are on average compatible with ground values with the same qualitative trends.","pith_inferences":["The same two-stage fitting scheme could be applied to any H2RG-based instrument with a flat-field stimulus sequence; the functional forms are generic and not tied to Euclid's specific filters or readout scheme.","The paper's 'on average compatible' statement is not backed by a statistical test; a concrete per-detector acceptance threshold, defined before comparing ground and flight parameters, would turn this into a falsifiable transfer check.","The measured dip in detected stimulus after a long dark period is not captured by the model, suggesting a natural extension in which an additional state variable tracks how many traps are filled at the start of each exposure.","A direct end-to-end test would be to feed the model the actual fluence history of a real Euclid pointing sequence containing bright stars and compare predicted persistence maps against the next dark or science exposure, quantifying the useful correction power in routine operations."],"forward_implications":["If the model is correct, Euclid's processing pipeline can predict persistence from the recent exposure history and either mask or subtract afterimages in ordinary survey frames.","A few-electron predictive accuracy would remove a known systematic in photometry and redshift measurements, directly protecting the weak-lensing and galaxy-clustering analyses.","Because $\\beta$ rises with fluence, persistence decays faster after brighter stimuli, so any correction must weight past exposures by their own signal levels rather than by a single fixed decay curve.","The stabilization measurements imply that calibration sequences taken after the detector has been dark for a long time reach steady state only after about 30 minutes, so persistence parameters should be derived from steady-state exposures.","Saturating stimuli change the persistence pattern and leave a long-lived altered detector state, so the below-saturation model does not apply after saturation and saturating events need separate handling."],"supporting_citations":[{"why":"Supplies the charge-trapping and release theory that motivates modeling persistence as a decaying current.","marker":"[17]"},{"why":"Reports persistence decay shape invariance and dependence on stimulus duration and intensity, used to interpret $\\beta(S)$.","marker":"[18]"},{"why":"Provides the finite-trapping-time interpretation used to explain steeper power-law indices at low fluence.","marker":"[11]"},{"why":"Earlier power-law persistence model with exposure-time dependence that this work adapts to Euclid's observation sequence.","marker":"[12]"},{"why":"Earlier power-law characterization of persistence in another H2RG-based instrument that supports the power-law choice.","marker":"[10]"},{"why":"Shows trap density is nonuniform in H2RG pixels, supporting per-pixel parameter maps and the observed persistence structures.","marker":"[1]"},{"why":"Alternative predictive persistence model whose assumptions about trapping time constants are compared with the stabilization measurements.","marker":"[19]"},{"why":"Defines the up-the-ramp signal estimator used to fit persistence ramps from nondestructive reads.","marker":"[9]"}],"fun_headline_variants":["Power-law model predicts Euclid IR persistence to few electrons","Euclid H2RG persistence fits power law to few-electron accuracy","Image persistence in Euclid NISP: power-law decay model validated","Persistence in Euclid IR detectors reduced to power-law prediction"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The few-electron accuracy in flight depends on the assumption that persistence parameters measured on the ground at 85 K with flat-field stimuli transfer to the real Euclid observation sequence; the paper reports flight parameters are only on average compatible with ground values and can differ significantly for some detectors, without a quantitative threshold.","fun_headline_variants_meta":{"raw":{"variants":["Power-law model predicts Euclid IR persistence to few electrons","Euclid H2RG persistence fits power law to few-electron accuracy","Image persistence in Euclid NISP: power-law decay model validated","Persistence in Euclid IR detectors reduced to power-law prediction"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000211,"raw_usage":{"total_tokens":1413,"prompt_tokens":943,"completion_tokens":470,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":559,"completion_tokens_details":{"reasoning_tokens":397}},"tokens_in":559,"tokens_out":470,"duration_ms":5351,"temperature":1.0,"reasoning_tokens":397,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T11:16:11.976064+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take an in-flight sequence in which a bright star has crossed a NISP detector, run the six-parameter model forward from the actual fluence history, and compare the predicted persistence with the measured residual image in the next pointing; if the residual exceeds a few electrons r.m.s. for one of the detectors flagged as discrepant, the ground-to-flight transfer claim fails.","supporting_citations":[{"cited_title":"Smith, Maximilian Zavodny, Gustavo Rahmer, and Marco Bonati","cited_arxiv_id":null,"evidence_quote":"Reports persistence decay shape invariance and dependence on stimulus duration and intensity, used to interpret $\\beta(S)$."},{"cited_title":"Long, Sylvia M","cited_arxiv_id":null,"evidence_quote":"Provides the finite-trapping-time interpretation used to explain steeper power-law indices at low fluence."},{"cited_title":"Long, Sylvia M","cited_arxiv_id":null,"evidence_quote":"Earlier power-law persistence model with exposure-time dependence that this work adapts to Euclid's observation sequence."},{"cited_title":"Long, Sylvia M","cited_arxiv_id":null,"evidence_quote":"Earlier power-law characterization of persistence in another H2RG-based instrument that supports the power-law choice."},{"cited_title":"Understanding Persistence: A 3D Trap Map of an H2RG Imaging Sensor","cited_arxiv_id":"1402.4181","evidence_quote":"Shows trap density is nonuniform in H2RG pixels, supporting per-pixel parameter maps and the observed persistence structures."},{"cited_title":"A New Signal Estimator from the NIR Detectors of the Euclid Mission","cited_arxiv_id":null,"evidence_quote":"Defines the up-the-ramp signal estimator used to fit persistence ramps from nondestructive reads."}],"review_version":1}