{"id":"aae0a3cf-0f85-4609-9c3a-b98ff87de668","arxiv_id":"2502.10282","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Using the JADES Origins Field, the authors show that medium-band photometry modestly improves high-z galaxy selection completeness, while bluer broadband depth matters most for reducing contamination.","lead":"A JWST deep field with 14 photometric bands shows that medium-width filters add about 10 percent completeness when selecting high-redshift galaxies, but contamination only drops when several medium bands are combined. The same data extends measurements of the ultraviolet luminosity function about one magnitude fainter at redshifts 8.5 to 13.5.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 10σ contamination claim rests on uncalibrated JAGUAR contaminant SEDs; the only real-data check shows contamination higher, not lower, at 5–8.5σ.","rationale":"The reader's identified assumption — that full-depth photo-zs are correct when labelling degraded-data contaminants — is a genuine limitation, and Section 5.2 explicitly states it. The JADES-GS-z14-0 case shows the assumption can fail, but that failure inflates the measured contamination rate because a true high-z source is counted as a contaminant, so it does not threaten the 'contamination is low' direction of the central claim. The more load-bearing assumption for the central claim is the representativeness of JAGUAR's contaminant population: the 10σ low-contamination number comes almost entirely from the simulation, and the only real-data comparison shows contamination higher than simulated at lower S/N. The paper is honest about JAGUAR's limitations, and the factor-10 correction is conservative, but that correction addresses the known under-density of true high-z galaxies, not the possible absence of rare dusty/strong-line Balmer-break interlopers from the mock catalogues. A simulation-injection test using empirical contaminant SEDs can settle this directly without new observations. Because the concern does not overturn the qualitative medium-band utility or blue-depth recommendations, and the paper already frames its conclusions conditionally, the reader's CONDITIONAL verdict stands without adjustment.","tokens_in":39287,"tokens_out":12700,"duration_ms":123373,"concrete_test":"Add to the JAGUAR mock catalogues a population of empirical contaminant SEDs drawn from spectroscopically confirmed low-z interlopers (e.g. Arrabal Haro et al. 2023; Zavala et al. 2023; Gandolfi et al. 2025), with a number density normalized to the contamination fraction seen in recent NIRSpec follow-up of JWST high-z candidates; then rerun the Section 4.2 completeness/contamination pipeline. If the 10σ contamination fraction rises above a few percent, Conclusions (i) overstates the safety of broad-band-only selections; if it stays below, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The headline quantitative claim in Conclusions (i) — contamination 'low (<4%) at 10σ+ even without medium bands' — is taken from the JAGUAR mock-catalogue experiments, but JAGUAR is a pre-JWST semi-analytic model whose contaminant population has not been calibrated to the Balmer-break/strong-line interlopers actually found in JWST samples. The paper's only real-data contamination measurement (Section 5.2) gives 20–38% at 5–8.5σ and is described as 'slightly higher than initially anticipated from the JAGUAR simulation work'; that is the wrong direction for a simulation that is meant to validate the low-contamination conclusion. The factor-10 up-weighting of z>11.5 sources in Section 4.2.2 corrects only the known under-density of true high-z galaxies (JAGUAR is ~20× low there), not any missing contaminant SEDs, so it cannot make the simulated rate a reliable upper limit if the interloper population itself is incomplete. At 10σ, the regime of the strongest claim, there is essentially no empirical constraint: the single 10σ source that changes classification in degraded data, JADES-GS-z14-0, is a true high-z galaxy mislabelled low-z in the full-depth catalogue, not a contaminant. The central 'contamination low at 10σ+' statement therefore rests on an unvalidated model population at exactly the signal-to-noise where the claim is made.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper uses the JADES Origins Field (JOF), a single NIRCam pointing with 14 broad- and medium-width bands, to assess how the inclusion or removal of medium-width photometric bands affects the selection completeness and contamination of high-redshift (z > 7.5) galaxy samples. The authors run the same band-removal experiments on real JOF data and on mock catalogues from the JAGUAR semi-analytic model, and they also degrade the real imaging by factors of 2, 4, and 8 to mimic shallower surveys. They report that at 10-sigma and above, completeness is high (80%+) and contamination low (<4%) even without medium bands, that medium bands increase completeness by about 10% but multiple medium bands are required to reduce contamination, that Balmer-Lyman degeneracies become common at 5-7 sigma, and that deeper blue broad bands (F090W/F115W) are important for reducing contamination. They then measure the UV luminosity function at 8.5 < z < 13.5 and find number densities consistent with other JWST studies. An appendix investigates the effect of changing the SED template set.","tokens_in":39590,"tokens_out":2433,"duration_ms":23896,"significance":"If the results hold, this is a useful empirical and simulation-based assessment of JWST filter-set choices for high-z surveys, with direct implications for survey design (e.g., the recommendation to invest in blue broad-band depth). The paper's strengths include the use of a uniquely deep 14-band field, a systematic comparison of band-removal and image-degradation experiments, an explicit external validation of photo-zs against 20 spectroscopic redshifts (19/20 consistent), a test of template-set dependence (Appendix A), and the public release of custom wisp templates and data products. The central qualitative result that medium bands help most in the 5-8 sigma regime and that blue broad-band depth is key is well supported by the real-data and simulation experiments. However, the headline quantitative claim of 'contamination low (<4%) at 10-sigma+ even without medium bands' relies entirely on the JAGUAR simulation, whose contaminant population has not been empirically calibrated, and the only real-data contamination measurement (20-38% at 5-8.5 sigma) is based on very small numbers.","major_comments":[{"comment":"The claim that contamination is 'low (<4%) at 10-sigma+ even without medium bands' is derived from the JAGUAR simulation, but the JAGUAR contaminant population (low-z Balmer break and strong-line galaxies) is pre-JWST and has not been calibrated against the interlopers actually found in JWST samples. The up-weighting factor of 10 for z>11.5 sources corrects only the known under-density of true high-z galaxies; it does not correct any missing contaminant SEDs. The real-data contamination measurement in §5.2 gives 20-38% at 5-8.5 sigma and is described as 'slightly higher than initially anticipated' — a direction that does not support treating the simulated 10-sigma contamination rate as a reliable upper limit. Please provide an explicit calibration check of the JAGUAR contaminant SEDs against JWST spectroscopic samples, or reframe the 10-sigma contamination claim as a model-dependent prediction with clearly stated caveats.","section":"§5.2"},{"comment":"The real-data contamination rates of 20-38% are based on only 12, 3, and 3 newly selected sources in the three degradation steps. The associated Poisson uncertainties are large (e.g., 6/12 has a 95% confidence interval of roughly 25-75%), yet the text presents these as a quantitative check on the simulation. Also, the calculation assumes that the full-depth photo-zs are correct for these sources; if any of the faint full-depth photo-zs are wrong, the measured contamination rate would be biased. Please report confidence intervals and explicitly discuss the robustness of the 20-38% estimate to the full-depth photo-z assumption.","section":"§5.2"},{"comment":"The completeness and contamination corrections applied to the real JOF UV luminosity function are taken from the JAGUAR simulation, so any systematic error in the simulated contamination rate propagates directly into the faint-end LF. Given the concern raised in §4.2.2 about uncalibrated contaminant SEDs, the agreement between LF measurements using different band combinations (C-ALL vs C-0/C-1/C-2) does not by itself validate the corrections, because all catalogues use the same JAGUAR-based corrections. Please test the sensitivity of the LF to an alternative contamination correction (e.g., one scaled to the real-data 20-38% rate), or explicitly state that the LF results inherit the simulation's contamination assumptions.","section":"§6"}],"minor_comments":[{"comment":"The sentence 'contamination is slightly higher than expected at 20-38%, through it is measured using a small number of sources' contains a typo: 'through' should be 'though'.","section":"§5.2"},{"comment":"The text says 'In Feburary of 2024' — 'Feburary' should be 'February'.","section":"§2"},{"comment":"The phrase 'more statistically probable' in Section 3.2(v) is informal; consider 'statistically preferred' or 'more probable under the model'.","section":"§3.2"},{"comment":"The completeness curves in Figure 8 show scatter of up to 4 percentage points across realisations; please state in the caption or text whether the quoted completeness values (e.g., 80% at 10-sigma) are the mean, median, or a single realisation.","section":"§4.2.2"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. First, this is the first real-data test of how medium-band photometry changes high-z galaxy selection, using the 14-band ultra-deep JADES Origins Field. Second, the paper's flagship number—contamination below 4% at 10σ even without medium bands—is a JAGUAR simulation result, not a real-data measurement. The real-data degraded-imaging experiment finds 20–38% contamination at 5–8.5σ, higher than the simulation predicted. That's the wrong direction for a simulation used to validate the headline claim, and at 10σ there is only a single source that changes classification, JADES-GS-z14-0, a true high-z galaxy mislabeled low-z in the full-depth catalog. So treat the abstract's contamination claim with caution.\n\nWhat's genuinely new: the systematic band-removal and image-degradation experiments on real JWST data, plus the first faint-end UV LF points at M_UV ~ -17.5 to -18 from this field. The paper is careful and honest—they flag the small numbers, the assumption that full-depth photo-zs are correct, and the ad hoc factor-10 up-weighting of z>11.5 sources. The completeness results are on firmer ground: 80%+ at 10σ, with medium bands adding ~10% completeness, and the point that blue broadband depth is key for rejecting Balmer-break interlopers comes through from both real data and simulations.\n\nSoft spots, in proportion. The contamination story is the weakest link. JAGUAR is a pre-JWST semi-analytic model; its contaminant population has not been calibrated to the Balmer-break/strong-line interlopers seen in real JWST samples. The factor-10 up-weighting fixes only the known underdensity of true z>11.5 galaxies, not any missing contaminant SEDs. The real-data contamination counts are tiny (12, 3, 3 new sources across degradation steps). The LF measurements are consistent with earlier work but subject to cosmic variance in a single pointing. None of this invalidates the qualitative conclusion that medium bands help and blue depth matters; it does mean the quantitative contamination rates should be presented as simulation-based expectations, not empirical measurements.\n\nWho should read this: anyone designing JWST survey filters or doing photo-z selection of z>8 galaxies, and people working on the faint-end UV LF. It deserves a serious referee. My recommendation: send it to review, but push the authors to separate JAGUAR-based contamination rates from the real-data measurements in the abstract and conclusions, and to either release the full catalogue or state clearly when it will be public.","headline":"A careful, survey-design-useful study whose headline contamination number is a simulation estimate, not an empirical measurement.","tokens_in":40189,"tokens_out":2582,"would_cite":true,"duration_ms":24464,"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":"This paper shows that at 10-sigma significance JWST broad bands alone select z>7.5 galaxies with 80%+ completeness and under 4% contamination, while medium bands add about 10% completeness and only deeper blue bands fully suppress low-z…","keywords":["high-redshift galaxies","photometric redshifts","medium-band photometry","JWST NIRCam","galaxy luminosity function","sample completeness","contamination","Lyman break galaxies"],"falsifier":"Spectroscopically measure the handful of sources that enter the high-z sample only when the JOF images are degraded: if most turn out to be genuinely at z>7.5, the claimed 20-38% contamination in the 5-8.5-sigma regime is an overestimate caused by incorrect full-depth photo-z labels, whereas confirmation of z<6 solutions would validate both the contamination rates and the Balmer-Lyman degeneracy mechanism.","tokens_in":39097,"feed_emoji":"🔭","tokens_out":7326,"duration_ms":66667,"temperature":0.7,"pith_summary":"This paper asks whether JWST's ultra-deep medium-width photometric bands justify their cost when the goal is a clean sample of galaxies at z>7.5. Using the 14-band JADES Origins Field, it shows that at 10-sigma and brighter, selections are already 80%+ complete and less than 4% contaminated using broad bands alone. Medium bands add roughly 10% completeness, mostly by pinning down emission-line positions for 8<z<10 sources, but a single band barely changes contamination, which only drops when several medium bands jointly probe the Balmer and oxygen lines of low-redshift interlopers. The most consequential finding is that the relative shallowness of the blue bands (F090W and F115W) is what lets Balmer-break galaxies at z about 2-4.5 masquerade as high-z dropouts in the 5-8-sigma regime, so deeper blue imaging matters as much as or more than extra medium bands.","feed_headline":"High-z galaxy samples run 80%+ complete without medium bands","feed_subtitle":"JWST medium bands add ~10% completeness; only deeper blue filters suppress the low-z interlopers.","key_machinery":"The controlling instrument is the JADES Origins Field itself: a single NIRCam pointing with 14 co-equal photometric bands (broad F090W through F444W plus medium F162M through F410M) reaching AB magnitudes of roughly 29.8-30.35, which lets the authors rerun one selection pipeline under different band combinations rather than comparing across surveys. The pipeline combines SExtractor photometry with EAZY photometric redshifts run twice, once allowing z up to 25 and once capped at z=6, so every candidate is scored by the chi-squared difference between its high-z solution and its best low-z Balmer-break alternative. Completeness and contamination are quantified by repeating the same experiments on JAGUAR semi-analytic mock catalogues (with z>11.5 sources up-weighted by a factor of 10 to match observed number densities) and by degrading the real images to probe the low signal-to-noise regime. The failure mode that carries the argument is the Balmer-Lyman break degeneracy: at z about 2-4.5, H-beta and O[III] emission falling inside broad filters can replicate the blue slope of a Lyman-break galaxy, a confusion that medium bands resolve only when they happen to straddle those lines.","core_discovery":"The paper's central claim, stated in its conclusions, is that broad-band-only JWST imaging is already sufficient to select z>7.5 galaxies with high completeness (80%+ at 10-sigma) and low contamination (under 4%), and that the role of medium-width bands is real but secondary: the first medium band recovers about 10% of 8<z<10 sources whose photo-z's are systematically underestimated by strong emission lines boosting F444W, while contamination from z about 1.5-4.5 Balmer-break interlopers falls only when multiple medium bands cover the expected line positions. A parallel finding is that the measured faint end of the UV luminosity function at 8.5<z<13.5 is consistent whether or not medium bands are used, and agrees with earlier JWST results showing high number densities of moderately faint galaxies around M_UV about -18. The paper also demonstrates, by artificially degrading the JOF imaging by factors of 2-8, that the 5-8.5-sigma regime is where Balmer-Lyman degeneracy sets in: real degraded data yield 20-38% contamination there, slightly above the JAGUAR simulation-based expectations.","pith_inferences":["If the blue-depth conclusion generalizes, the optimal survey design for high-z science is not simply more medium bands but a rebalancing of exposure time toward F090W and F115W; this is a testable prescription that future JWST observing programmes could adopt and evaluate.","The Appendix A template-swap results imply that the measured contamination floor is also a statement about the current SED template set; as templates improve for young, blue, strong-line emitters, the same 5-8.5-sigma sources may re-classify without any new observations.","The JAGUAR-based completeness curves suggest that faint-end UV luminosity function slopes measured from a single deep pointing carry a completeness-correction uncertainty of order 10-15% per bin, comparable to Poisson error in the faintest bins; propagating band-set choice into the Schechter alpha error budget across several independent fields would quantify this directly."],"forward_implications":["Broad-band-only surveys at JWST depth can build high-z galaxy samples that are more than 80% complete and less than 4% contaminated at 10-sigma and brighter, so large-area statistical studies do not need medium bands to be reliable.","Adding the first medium-width band (for example F410M) recovers about 10% of 8<z<10 galaxies by correcting systematically underestimated photometric redshifts, but contamination only drops when several medium bands jointly probe the emission lines of low-z interlopers.","In the 5-8.5-sigma regime, Balmer-Lyman degeneracy produces 20-38% contamination in degraded real data, so ultra-faint high-z candidates near the detection limit must be treated as provisional unless spectroscopy or line-constraining bands are available.","Deepening the blue bands F090W and F115W by about 0.4 magnitudes is at least as effective at suppressing contamination as adding more medium bands, and NIRCam can acquire that blue depth in parallel with redder medium bands.","UV luminosity function measurements at 8.5<z<13.5 are insensitive to the choice of band set once completeness and contamination are corrected, supporting the high number densities of M_UV about -18 galaxies reported by other JWST deep fields."],"supporting_citations":[{"why":"Supplies the JAGUAR semi-analytic mock catalogues that are the simulation backbone for every completeness and contamination measurement in the paper.","marker":"Williams et al. (2018)"},{"why":"Presents the JADES Origins Field survey design, the 14-band dataset and depths that the entire experiment is built on.","marker":"Eisenstein et al. (2023b)"},{"why":"Provides the official JOF team high-z sample that the authors cross-match and compare their selections against.","marker":"Robertson et al. (2024)"},{"why":"Supplies EAZY, the SED fitting code that carries the photometric redshift and low-z-versus-high-z discrimination procedure.","marker":"Brammer et al. (2008)"},{"why":"Provides the young, blue, strong-emission-line template set used in the fiducial fits and tested against an alternative template set.","marker":"Larson et al. (2023)"},{"why":"Defines the EPOCHS selection criteria and UV luminosity function methodology that this work adapts and extends.","marker":"Adams et al. (2024)"},{"why":"Provides NIRSpec spectroscopy of JOF sources, anchoring the highest-redshift candidates and motivating the damped Lyman-alpha test.","marker":"Carniani et al. (2024)"},{"why":"Serves as the multi-survey comparison at z about 12.5 against which the JOF number densities are judged roughly 0.15 dex higher.","marker":"Donnan et al. (2024)"}],"fun_headline_variants":["Broad bands alone yield 80%+ complete high-z samples","Medium bands add just 10% to high-z completeness","Blue depth, not medium bands, cuts high-z contamination","High-z UV luminosity function robust to filter choice","JWST broad bands suffice for z>7.5 galaxy selection"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the photometric redshifts computed from the full 14-band, full-depth images are correct when used to label sources newly selected in the degraded images as contaminants; if those faint full-depth photo-z's are wrong, the 20-38% contamination rates measured in the 5-8.5-sigma regime would be biased.","fun_headline_variants_meta":{"raw":{"variants":["Broad bands alone yield 80%+ complete high-z samples","Medium bands add just 10% to high-z completeness","Blue depth, not medium bands, cuts high-z contamination","High-z UV luminosity function robust to filter choice","JWST broad bands suffice for z>7.5 galaxy selection"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000225,"raw_usage":{"total_tokens":1548,"prompt_tokens":1110,"completion_tokens":438,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":726,"completion_tokens_details":{"reasoning_tokens":356}},"tokens_in":726,"tokens_out":438,"duration_ms":4467,"temperature":1.0,"reasoning_tokens":356,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T18:40:38.195374+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Spectroscopically measure the handful of sources that enter the high-z sample only when the JOF images are degraded: if most turn out to be genuinely at z>7.5, the claimed 20-38% contamination in the 5-8.5-sigma regime is an overestimate caused by incorrect full-depth photo-z labels, whereas confirmation of z<6 solutions would validate both the contamination rates and the Balmer-Lyman degeneracy mechanism.","supporting_citations":[],"review_version":1}