{"id":"726b14c8-304f-4ce7-b505-894f628c7807","arxiv_id":"2505.09933","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"POSEIDON now offers validated, open-source high-resolution emission and transmission retrievals, but results depend noticeably on detrending choices.","lead":"A team of exoplanet scientists has added a high-resolution spectroscopy analysis tool to the open-source POSEIDON retrieval code, letting researchers measure the chemistry and temperature of exoplanet atmospheres from ground-based data in under a day on a regular computer. The tool reproduces earlier measurements of two well-studied hot Jupiters, but shows that preprocessing choices can change the inferred chemical abundances.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Fast-filtering projection (Eq. 4) may distort line shapes wavelength-dependently; injection test's α≈0.5 is not shown to be benign for abundance recovery.","rationale":"The paper's central claim is that POSEIDON v1.3 offers a validated, open-source HRCCS retrieval framework. The authors support this with injection tests, reproduction of WASP-77Ab and WASP-121b retrievals, and a public code release. The most load-bearing assumption is the fast-filtering approximation in Eq. 4, which linearly projects the forward model onto the detrending basis. Because the basis U is estimated from data containing the planet signal, the projection can absorb part of that signal. The injection tests (Section 4.3) show α about half the injected value; the authors attribute this to T-P degeneracy, but the paper does not show quantitative abundance recovery (injected vs retrieved logX) to demonstrate that the residual distortion is only a uniform scaling. If the attenuation differs across wavelengths, the relative line depths—from which abundances are constrained—are distorted, biasing abundances even with α free. The spurious NH3 detection in WASP-77Ab (Section 5.2) and the PCA/SYSREM differences reinforce that filtering choices affect abundance constraints. The Data Availability statement ('will be available upon request') also limits reproducibility, though this is secondary to the correctness concern. Nevertheless, the paper is transparent about these caveats and provides a code release; the issue is not internal inconsistency but an unverified assumption that the α offset is wavelength-independent. The reader's conditional verdict is appropriate, with the added condition that the T-P-fixed injection test and quantitative recovery statistics be reported.","tokens_in":27822,"tokens_out":9298,"duration_ms":84790,"concrete_test":"Re-run the Section 4.3 emission injection test with the T-P parameters fixed to their injected values, removing the T-P degeneracy, and compare the posterior median α and the four logX posteriors to the injected values (α=1, logX=-4). If α remains near 0.5 while all logX are recovered within their 68% credible intervals, the filtering distortion is uniform and the framework is validated for abundance work. If any logX is biased by more than ~0.3 dex, or if α is not recovered even with a correct T-P profile, the fast-filtering projection introduces a wavelength-dependent signal loss that biases abundances, and the framework needs revision or explicit filtering-dependent calibration.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The fast-filtering preprocessing in Eq. 4 assumes the detrending distortion is well approximated by a linear projection M' = U(ΛU)†(ΛM). If the planet signal has nonzero overlap with the low-rank basis U (computed from data that include the planet), the projection removes part of that signal. Section 4.3 reports the retrieved scale factor α at roughly half the injected value. The authors attribute this to T-P profile degeneracy, but if the filter attenuation is wavelength-dependent, a single scalar α cannot correct it and retrieved abundances—derived from relative line depths—will be biased. The paper states abundances are 'accurately constrained' yet does not present quantitative injection-versus-retrieval statistics in the text, only referencing corner plots. The subsequent spurious NH3 detection in WASP-77Ab (Section 5.2) and demonstrated PCA/SYSREM differences show filtering choices materially change abundance constraints. The central claim of a validated, reliable framework thus depends on the unverified assumption that the α offset reflects only a uniform scaling, not a wavelength-dependent distortion of the model spectrum.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces a high-resolution cross-correlation spectroscopy (HRCCS) retrieval framework implemented in the open-source POSEIDON code, supporting both emission and transmission spectroscopy. It describes the forward-model modifications, detrending methods (PCA, SYSREM, NMF), a fast-filtering preprocessing step, and a Gaussian likelihood mapping that follows previous work by Brogi & Line (2019) and Gibson et al. (2020, 2022). The framework is validated with injection tests on real IGRINS and UVES data and then applied to WASP-77Ab emission and WASP-121b transmission observations, broadly reproducing previously published retrievals while documenting sensitivity to the choice of detrending filtering.","tokens_in":28004,"tokens_out":5171,"duration_ms":51465,"significance":"If the central claim holds, this paper makes a substantial contribution to the exoplanet high-resolution spectroscopy community by providing an open-source, CPU-only, documented retrieval framework within the widely used POSEIDON code, with runtimes an order of magnitude faster than earlier GPU-based implementations. The validation strategy—injecting known signals into real data and comparing retrievals against published WASP-77Ab and WASP-121b results—is appropriate and non-circular. The paper also explicitly demonstrates that detrending choices (PCA vs SYSREM) propagate into retrieved abundances, including a spurious NH3 detection in WASP-77Ab, which is a valuable public caveat for HRCCS abundance studies.","major_comments":[{"comment":"The injection tests recover the scale factor alpha at roughly half its injected value. The authors attribute this to degeneracy with the retrieved P-T profile and to the filtering process, but they do not demonstrate that the attenuation introduced by the fast-filtering projection M' = U(Lambda U)^dagger (Lambda M) is a uniform scalar. If the projection removes signal in a wavelength-dependent or line-strength-dependent manner, then a single scalar alpha cannot correct for it, and retrieved abundances derived from relative line depths will be biased. The paper states that abundances are 'accurately constrained' and points to corner plots, but it does not report quantitative injection-versus-retrieval statistics (e.g., bias and scatter for each species) in the text. To support the central validation claim, the authors should provide such statistics for both emission and transmission injection tests and show that the alpha deficit is not wavelength-dependent, for example by testing recovery on wavelength subsets or comparing line ratios in the filtered models.","section":"4.3, Eq. (4)"},{"comment":"Two supporting experiments are mentioned but not shown: the authors write that 'We confirm the scaling of the signal remains unchanged after the filtering through a noiseless realization of data' and that 'we have run another experiment demonstrating the scaling can be correctly retrieved when the true profile is isothermal.' These experiments are load-bearing because they are used to argue that the alpha offset is benign and attributable to T-P degeneracy rather than to filter-induced distortion. Without presenting these results (at least in an appendix), the claim that the alpha offset does not affect abundance accuracy is unsupported. The manuscript should include the relevant figures or tables, or remove these claims.","section":"4.3"},{"comment":"The application to WASP-77Ab shows that retrieved chemical abundances differ by about 0.2 dex between PCA and SYSREM (e.g., log H2O is -4.33 with fixed R_p and PCA versus -4.12 with fixed R_p and SYSREM), and that NH3 is spuriously detected with SYSREM but not with PCA. The text acknowledges these user-choice differences, but the paper still concludes that 'posteriors are in agreement' with Line et al. (2021). Given the central claim is a validated framework for reliable abundance constraints, the authors should quantify the systematic spread introduced by detrending choices and explicitly state the effective abundance precision of the framework (e.g., 0.2-0.5 dex) in the summary. Without this, the reliability claim is overstated.","section":"5.2, Figure 9"}],"minor_comments":[{"comment":"The acronym 'HRCSS' appears in the Introduction ('HRCSS retrieval techniques have seen limited application') and should be 'HRCCS'.","section":"1"},{"comment":"In the paragraph on shifting the model to the data wavelength grid, 'This is typically done by by interpolation' contains a duplicated 'by'.","section":"2.4"},{"comment":"Section 4.3 states that the planet radius was fixed in the transmission injection test, but Table 2 lists R_r,ref as a free parameter with a Gaussian prior. Please clarify whether the radius was fixed in that test and, if so, remove it from the table or mark it as fixed.","section":"4.3, Table 2"},{"comment":"The sentence 'The offset in absolute abundances can be attributed the strong correlation between each chemical abundance and cloud deck pressure' is missing a 'to' after 'attributed'.","section":"6.2"},{"comment":"In the first sentence of Section 7, 'transmission and emission spectrosa' should be 'spectra'.","section":"7"},{"comment":"The paper states 'We have no concrete evidence of SYSREM outperforming PCA in preserving planet signal' but later the same section says the stronger CO detection 'hints at SYSREM's improvement in preserving planet signals.' These statements are in tension; please rephrase to be consistent.","section":"5.2"}],"recommendation":"major_revision","confidential_remarks":"This is a valuable methods paper with a clear open-source deliverable. The main concern is whether the alpha offset in the injection tests is benign for abundance recovery; the authors need to provide the quantitative evidence they cite. If that evidence is provided, the paper is likely acceptable. The paper's findings on detrending-induced abundance shifts and the spurious NH3 detection are important for the HRCCS community and should be highlighted more prominently in the abstract or conclusions."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague—the short version: this is a solid code paper that belongs in the literature, but the validation section leaves one real question open. The authors bring HRCCS retrievals into POSEIDON, which is already the standard open-source retrieval package for low-resolution work, and they do it for both emission and transmission. That is the gap. The framework is fast, CPU-only, and documented with notebooks; the runtime claims look credible relative to past work. The injection tests recover known abundances for four-species emission and three-species transmission models, and the WASP-77Ab/WASP-121b reproductions match published results to the level one would expect given different detrending choices. The paper's demonstration that PCA vs SYSREM changes the WASP-77Ab results and produces a spurious NH3 detection is worth reading—it is the kind of honest negative result the field needs.\n\nThe soft spots are real but not disqualifying. The fast-filtering approximation in Eq. 4 is the load-bearing assumption: if the planet signal has nonzero overlap with the background basis, the filter removes part of it, possibly wavelength-dependently. The paper reports the retrieved scale factor α at about half the injected value. They attribute this to T-P degeneracy and present supporting experiments (a noiseless realization and an isothermal retrieval) but they don't show a quantitative injection-recovery table—bias and scatter on log X species across the test. The statement that abundances are 'accurately constrained' is plausible from the corner plots, but it would be stronger with numbers. This is a fixable weakness, not a fatal one.\n\nThe other clear issue: the paper says all scripts are 'available upon request.' For an open-source methods paper, that is not good enough—a commit hash and archived notebooks are needed. Minor.\n\nThis deserves a serious referee. The code is a community resource, the method is a genuine extension of POSEIDON, and the weaknesses are addressable with quantitative validation and archived scripts. I'd send it out, with a referee asked to check the α behavior and ask for injection-recovery statistics.","headline":"Open-source unification of HRCCS retrievals in POSEIDON is real and useful; the α≈0.5 offset deserves fuller quantitative treatment, but the paper should be peer reviewed.","tokens_in":28574,"tokens_out":2214,"would_cite":true,"duration_ms":23649,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"One open-source framework unifies high-resolution emission and transmission retrievals and shows detrending choices change abundances.","keywords":["high-resolution spectroscopy","atmospheric retrieval","exoplanet atmospheres","HRCCS","transmission spectroscopy","emission spectroscopy","Bayesian inference","detrending"],"falsifier":"Take a real or synthetic dataset, inject a planet signal at several levels of overlap with the telluric/systematic basis (for example, shift one strong line onto a deep telluric line, or add a stationary component to the injected signal), and run the retrieval. If the recovered mixing ratios drift away from the injected values as the overlap increases, the linear-projection assumption has failed; a clean version is to compare the recovered scale factor $\\alpha$ to its injected value across the overlap grid, since the paper already finds $\\alpha$ at roughly half the injected value in its fiducial tests.","tokens_in":27603,"feed_emoji":"🔭","tokens_out":8767,"duration_ms":85970,"temperature":0.7,"pith_summary":"The paper sets out to make high-resolution cross-correlation spectroscopy (HRCCS) retrievals a routine, open-source activity rather than a custom-analysis specialty. It claims that one framework, added to the POSEIDON retrieval code, can handle both emission and transmission spectra, complete a typical retrieval in under 12 hours on CPUs alone, and reproduce previously published abundance results for WASP-77Ab and WASP-121b when the same detrending assumptions are used. The reason a reader should care is that this removes the GPU and specialist-code barriers that have kept HRCCS retrievals out of reach for many groups, and it exposes a systematic effect: different, equally standard detrending choices change the retrieved chemical abundances, enough that a spurious ammonia detection appears or disappears in WASP-77Ab.","feed_headline":"High-resolution exoplanet retrievals now run on CPUs in under 12 hours","feed_subtitle":"Open-source POSEIDON reproduces published results for two hot Jupiters and shows filter choice shifts abundances.","key_machinery":"The load-bearing device is fast filtering: after detrending, a low-rank basis $U$ describes the telluric, stellar, and instrumental background in each spectral order; every trial model $M$ is then filtered by projecting it out of that basis, $M' = U(\\Lambda U)^\\dagger (\\Lambda M)$, where $\\Lambda$ weights pixels by their time- and wavelength-dependent uncertainties. Because $U(\\Lambda U)^\\dagger \\Lambda$ is independent of the model, it is computed once per order and reused across the hundreds of thousands of models in a retrieval. This operation, rather than the forward model itself, is what makes the runtime drop to hours on a CPU.","core_discovery":"On its own terms, the paper's claim is that the distortion a detrending step (PCA, SYSREM, or NMF) imposes on a planet signal can be mimicked by a single linear projection applied to every trial forward model, and that with this operation in place a Bayesian retrieval over lines, temperatures, Doppler shifts, and scale factors is computationally cheap. Fast filtering replaces the old scheme of injecting each model into the background and refiltering it, cutting the per-model cost to a precomputable projection. The paper validates the framework with injection-recovery tests on realistic data, then re-derives the atmospheric properties of the hot Jupiter WASP-77Ab in emission and the ultra-hot Jupiter WASP-121b in transmission, finding results broadly consistent with previous published retrievals. It also reports that the retrieved signal scale factor comes out about half the injected value, that abundance posteriors shift with the choice of detrending method, and that a low-significance NH3 detection in WASP-77Ab is an artifact of the filter choice.","pith_inferences":["A natural next step would be to treat the filter choice as a nuisance systematic, running every retrieval with two independent detrenders and quoting the spread as part of the error budget; the paper's results imply the spread can be comparable to the statistical uncertainty.","The roughly factor-of-two deficit in the retrieved scale factor suggests a calibration diagnostic: if injection tests covering realistic signal-to-noise and telluric overlap can map how much signal each filter removes, $\\alpha$ could become a correction factor for abundance estimates rather than just a nuisance parameter.","Because the framework runs on CPUs in hours, the large archive of existing high-resolution observations could be re-analyzed at the population level, turning individual-target HRCCS retrievals into survey-scale measurements.","Combining the fast-filtering projection with POSEIDON's existing multidimensional forward models would allow a direct test of whether 3D morning-evening or day-night structures can survive the detrending distortion; this is an implication the paper notes as future work but does not demonstrate."],"forward_implications":["Typical emission and transmission retrievals finish in about seven hours on twenty-four CPU cores, so HRCCS abundance analyses no longer require GPUs or proprietary code.","Reproducing the WASP-77Ab and WASP-121b results validates the method against independent analyses, which should increase confidence in, and scrutiny of, published high-resolution abundance constraints.","Because detrending choices shift abundance posteriors, published single-method HRCCS constraints carry a systematic uncertainty that the paper's experiments bound but do not remove.","The framework is positioned to combine high- and low-resolution data, since it reuses POSEIDON's forward models; this is the paper's stated goal for future work.","A spurious NH3 detection in WASP-77Ab that depends on filter choice implies that chemical detections made with a single filtering method should be re-checked with at least one other method."],"supporting_citations":[{"why":"Supplies the likelihood mapping and model-injection-into-PCA scheme for emission retrievals that this framework generalizes.","marker":"Brogi & Line (2019)"},{"why":"Introduces SYSREM-based transmission retrievals and the heteroscedastic noise model used in the likelihood.","marker":"Gibson et al. (2020)"},{"why":"Provides the fast-filtering linear projection (Equation 4) and the WASP-121b UVES analysis used as a validation target.","marker":"Gibson et al. (2022)"},{"why":"Sets the emission retrieval benchmark for WASP-77Ab whose results must be reproduced, plus the runtime comparison.","marker":"Line et al. (2021)"},{"why":"Independent open-source retrieval code used to cross-check detection significance on WASP-121b data.","marker":"Pelletier et al. (2023)"},{"why":"Open-source SPIRou HRCCS retrieval code cited as the runtime and methodology baseline for comparison.","marker":"Klein et al. (2024)"},{"why":"The POSEIDON package itself, into which the new high-resolution framework is built.","marker":"MacDonald (2023)"},{"why":"Defines the SYSREM algorithm that provides one of the low-rank bases for the fast-filtering step.","marker":"Tamuz et al. (2005)"}],"fun_headline_variants":["Open-source POSEIDON cuts retrieval time to <12 hours on CPUs","Exoplanet atmosphere retrievals go open-source and CPU-friendly","Fast, free exoplanet retrievals: POSEIDON reproduces hot Jupiter results","Filter choice shifts exoplanet abundance estimates, new open-source tool"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The analysis assumes that applying the same linear filtering used on the data to the forward model reproduces how the real planet signal was distorted; a signal that overlaps the removed background loses part of itself to the filter, which would skew the retrieved abundances.","fun_headline_variants_meta":{"raw":{"variants":["Open-source POSEIDON cuts retrieval time to <12 hours on CPUs","Exoplanet atmosphere retrievals go open-source and CPU-friendly","Fast, free exoplanet retrievals: POSEIDON reproduces hot Jupiter results","Filter choice shifts exoplanet abundance estimates, new open-source tool"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000994,"raw_usage":{"total_tokens":4209,"prompt_tokens":945,"completion_tokens":3264,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":561,"completion_tokens_details":{"reasoning_tokens":3183}},"tokens_in":561,"tokens_out":3264,"duration_ms":26515,"temperature":1.0,"reasoning_tokens":3183,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T21:20:43.238334+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a real or synthetic dataset, inject a planet signal at several levels of overlap with the telluric/systematic basis (for example, shift one strong line onto a deep telluric line, or add a stationary component to the injected signal), and run the retrieval. If the recovered mixing ratios drift away from the injected values as the overlap increases, the linear-projection assumption has failed; a clean version is to compare the recovered scale factor $\\alpha$ to its injected value across the overlap grid, since the paper already finds $\\alpha$ at roughly half the injected value in its fiducial tests.","supporting_citations":[],"review_version":1}