{"id":"32c2e326-ab59-492d-b3af-1bd857e934cc","arxiv_id":"2605.27364","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"DREAMS DR1 provides ~59 million bulge star light curves at minute cadence to z~22, with a pilot search yielding one microlensing event, two flares, and 24 new short variables.","lead":"A new survey of the Galactic bulge, DREAMS, has released minute-cadence light curves for about 59 million stars, reaching about two magnitudes deeper than earlier bulge time-domain surveys. A small pilot search already found a new microlensing event, two stellar flares, and 24 new short-period variables.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Extrapolated flare and short-variable yields rest on a pilot region observed at double the survey's nominal cadence, inflating the headline counts.","rationale":"The reader's verdict (CONDITIONAL) is appropriate. I identify a more specific, technical reason the pilot-search extrapolation is unreliable: the pilot region sits in the D01∩D02 overlap, which the paper itself (§2.1) describes as being observed at twice the nominal cadence. The reader flagged the non-random selection of the pilot region, but the cadence bias is a distinct and arguably more decisive flaw — it means the detection efficiency for the searched short-duration phenomena is systematically higher in the pilot than across most of the 5 deg² field. This is not a matter of disagreement with consensus; it is an internal inconsistency between the survey design described in §2.1 and the unqualified extrapolation in §5.2/Abstract. I also considered the catalog-duplication issue (§3.5.2), which affects the 59,372,789-star count, but the paper discloses it explicitly and it does not change the core unique-dataset claim. The 'twice as many stars' claim rests on a single-region comparison (§5.3), also a valid concern, but the star-count claim is less central to the paper's scientific promise than the yield extrapolation. My recommended verdict remains CONDITIONAL: the catalog itself appears genuine and useful, but the abstract's yield statements need revision or qualification before acceptance as stated.","tokens_in":22929,"tokens_out":6892,"duration_ms":69919,"concrete_test":"Run the identical box-fit pilot search on a 0.04 deg² region in the D01 or D02 field outside the 1 deg² overlap, where the cadence is the nominal Γ_z≈20 hr⁻¹, matched as closely as possible in extinction/crowding. Inject simulated flare and short-period-variable signals into the real light curves of both the pilot and control regions to measure relative recovery fractions. If the flare and variable surface densities in the control region, after correcting for the measured detection-efficiency ratio, are consistent within Poisson errors with the pilot counts, then the extrapolation is sound; if not, the 'hundreds/thousands' statements must be revised or removed.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The abstract and §5.2 claim that DR1 'may contain hundreds of stellar flares and thousands of previously unknown short variables,' extrapolating from a pilot search over 0.04 deg² that found 2 flares, 1 microlens, and 24 new variables. But the pilot region is the D01∩D02 overlap (D01 N22 / D02 S18 chips), which §2.1 states is observed at twice the nominal cadence: 'an additional ~1 deg² region is observed at twice these cadences,' i.e., Γ_z≈40 hr⁻¹ rather than 20 hr⁻¹. The box-fit search requires ≥5 consecutive in-window points with >3σ residuals and scans windows as short as ~58 min, so detection efficiency is strongly cadence-dependent: a 20–30 min flare that yields 3–4 points at Γ=20 produces 6–8 points at Γ=40, passing the threshold only in the pilot region. Linearly scaling 2 flares over 0.04 deg² to 5 deg² therefore overstates the full-survey yield; the correct extrapolation must divide by the relative sensitivity. The same bias affects the 24 short variables, whose periods/amplitudes are preferentially recovered at higher cadence. Small-number statistics (2 flares) add further uncertainty. The authors caution against extrapolating the microlensing rate because the region was not random, but do not apply the same caution to flares/variables, despite the cadence bias being quantitatively more important. This does not invalidate the catalog itself, but it directly undermines the headline yield claims in the abstract.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper presents the first public data release of the DREAMS time-domain survey of the Galactic bulge. DR1 comprises 1,856 z-band and 325 r-band observations of 59,372,789 catalog sources in a 5 deg² field, with per-source light curves, finding charts, CMDs, and a web portal. The authors describe the survey design, DIA reduction using pySIS adapted to DECam, reference-image construction, catalog merging, and calibration against DECaPS2. They demonstrate the data quality with light curves of a BLAP and OGLE-TR-18, including a newly detected secondary eclipse, and a comparison of the measured per-exposure SNR with a pre-survey prediction. A pilot search in 0.04 deg² (245,490 sources) finds one microlensing event, two stellar flares, and 31 short-period variables (24 new), and the abstract extrapolates to 'hundreds of flares' and 'thousands of variables' in the full DR1. The central claim is that DR1 is a uniquely deep, minute-cadence, public bulge time-domain catalog.","tokens_in":23316,"tokens_out":9073,"duration_ms":81473,"significance":"If the data products are as described, DR1 is a valuable community resource: it is about two magnitudes deeper than prior bulge time-domain surveys at minute cadence, and the public portal is a significant contribution. The reduction appears careful, the SNR curve matches the earlier sensitivity estimate, and the example light curves demonstrate genuine capability. However, the headline yield extrapolations are not supported by the pilot search as currently presented, because the pilot region is observed at twice the survey cadence and no completeness simulation is provided. The catalog itself does not depend on these extrapolations, so the paper can be made sound with targeted revisions.","major_comments":[{"comment":"The pilot region (D01 N22 / D02 S18 overlap) lies inside the ~1 deg² portion of DREAMS that is observed at twice the nominal cadence (Γ_z≈40 hr−1 vs Γ_z≈20 hr−1). The box-fit search requires ≥5 consecutive in-window points with >3σ residuals, so detection efficiency is strongly cadence-dependent. A 20–30 min flare that produces 6–8 points at double cadence will typically produce only 3–4 points at the nominal cadence and fail the threshold. The paper cautions against extrapolating the microlensing rate because the region was not chosen randomly, but does not apply the same caution to the flare and variable counts. Linearly scaling 2 flares and 24 new variables from 0.04 deg² to 5 deg² therefore overstates the expected yields. The abstract's 'hundreds of flares / thousands of variables' should be removed or replaced with a properly cadence-corrected estimate.","section":"Section 5.2 and Section 2.1"},{"comment":"No completeness or false-positive simulation is presented for the pilot search. The thresholds Δχ²>300, ≥5 consecutive >3σ points, and the window grid are chosen by hand, and the search is run only in the double-cadence region. Without injection/recovery tests, the pilot counts of 2 flares and 24 variables do not provide a calibrated detection efficiency even at the pilot cadence, so any extrapolation is uncertain beyond the cadence bias. At minimum, a Poisson interval for 2 flares should be stated; the 1σ range is roughly 0.3–6 flares, which makes 'hundreds' extremely sensitive to the assumed sensitivity.","section":"Section 5.2"},{"comment":"The DR1 '59,372,789 stars' is not a de-duplicated unique-source count. The text states that the cross-matching removed only ~5.2 million of the ~13 million expected duplicates in the 1 deg² overlap and that 'a substantial number of duplicates likely remain,' mainly among fainter stars. The abstract's claim that the catalog 'contains about twice as many stars as previous catalog covering the same 5 deg² area' is based on Figure 11, a single 2.53′×2.40′ region, and the same figure shows a CFHT catalog with ~2.2× more stars in that region. Please either release a de-duplicated catalog, quantify the residual duplicate fraction, or rephrase to 'catalog entries' and soften the comparison to a regional demonstration.","section":"Section 3.5.2 and Abstract"}],"minor_comments":[{"comment":"Caption: 'Field stars are from the same 2.6′×2.6′ stamp on D01 field' but the BLAP target is in D02; this appears to be a typo.","section":"Figure 7"},{"comment":"Abstract says 'at least twice as many stars' while Section 5.3 says 'about twice as many stars'; please make the wording consistent.","section":"Abstract vs Section 5.3"},{"comment":"'which is likely to be much less affected by such selection bias' is not justified; the cadence bias discussed above applies directly to flares and variables.","section":"Section 5.2"},{"comment":"Grammar: 'their formation channels is still not yet completely understood' should be 'their formation channels are still not completely understood.'","section":"Section 5.1"},{"comment":"The description 'five z-band blocks and one r-band block' is slightly ambiguous next to Table 1's notation; clarify that each block contains multiple exposures.","section":"Section 2.2"}],"recommendation":"major_revision","confidential_remarks":"This is a solid data-release paper with a transparent pipeline; the main defect is the unsupported extrapolations in the abstract, which can be fixed by reanalysis or rewording. The duplicate-source issue also matters for how the catalog count is cited. I recommend major revision rather than rejection because the underlying catalog appears sound and useful."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The thing to know about this paper: the DREAMS DR1 catalog is real, public, and unique, and you should take it seriously. The abstract's extrapolated yields are the weak part, and they should be fixed before publication. The two issues are separate, and the first one wins.\n\nWhat's actually new: 59,372,789 stars in the Galactic bulge with minute-level cadence (20–40 hr^-1 in z) and single-exposure depth z_AB ~ 22, about two magnitudes deeper than prior bulge time-domain surveys. That combination does not exist in any other public dataset. The authors have shipped the light curves, finding charts, CMDs, and a working query portal — that is concrete, reproducible value, not a promise. The reduction pipeline (pySIS adapted to DECam) is standard DIA, but applied at scale, and the photometric SNR curve matches their earlier sensitivity estimate. They also show two known short-period variables recovered cleanly and disclose the known issue of duplicates in the overlap region. Those are marks in favor.\n\nThe soft spot is the pilot search in Section 5.2. The abstract says DR1 'may contain hundreds of stellar flares and thousands of previously unknown short variables.' That extrapolation is a straight linear scaling from 2 flares and 24 new variables found in a 0.04 deg^2 pilot region. But the pilot region is inside the D01/D02 overlap, which is observed at twice the nominal cadence (Gamma_z ~ 40 vs 20 hr^-1). The box-fit search requires at least five consecutive >3sigma in-window points, so a 20–30 min flare with 3–4 points at the nominal cadence gets 6–8 points in the overlap and passes the threshold only there. Linearly scaling 2 flares to 5 deg^2 therefore overstates the yield, and the same cadence bias affects the recovery of short-period variables. The authors note the pilot region was not random for the microlensing event but do not apply that caution to the flare/variable counts. Small-number statistics make it worse.\n\nThis does not invalidate the catalog. The depth, the cadence, and the public access stand on their own. But the abstract should either drop the numbers, give them with a cadence-sensitivity correction, or flag them as upper limits. Similarly, the 'twice as many stars' claim comes from a single 2.53' x 2.40' region; it is plausible, but it is a local comparison, not a survey-wide measurement.\n\nWho gets value: anyone doing bulge variability, stellar flares, short-period pulsators, or microlensing follow-up, and anyone who needs a deep, high-cadence public light-curve set. This deserves a serious referee. I would send it out, and ask for a revision that calibrates the yield extrapolation and softens the abstract.\n\nFor the reading group: worth a look when it lands in a journal, mainly to discuss how survey papers should present extrapolated event yields. I'd cite it if I worked in the area.","headline":"A genuinely useful public data release whose abstract oversells the pilot-search yields; the catalog itself deserves a careful referee.","tokens_in":797,"tokens_out":1075,"would_cite":true,"duration_ms":28823,"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":"DREAMS has released the deepest public time-series catalog of the Galactic bulge: minute-cadence light curves for 59 million stars.","keywords":["Galactic bulge","time-domain surveys","difference image analysis","photometric catalogs","microlensing","free-floating planets","stellar flares","short-period variables"],"falsifier":"Run the same sliding-box search (at least five consecutive >3σ points, Δχ²>300) on a randomly selected 0.04 deg² patch that had no prior visual inspection, and compare the yields of flares and short-period variables to the pilot's 2 flares and 31 variables per 245,490 stars. A significantly lower yield would falsify the extrapolated full-catalog counts; a comparable yield would support them.","tokens_in":22849,"feed_emoji":"🔭","tokens_out":7173,"duration_ms":67032,"temperature":0.7,"pith_summary":"The paper introduces the first public data release of DREAMS, a survey watching 5 square degrees of the Galactic bulge with a 4-meter-class camera at a cadence of roughly one z-band image every three minutes. DR1 contains 1,856 z-band and 325 r-band observations of 59,372,789 stars, about twice as many stars as any earlier catalog of the same area, reaching a depth of z_AB ≈ 22 — roughly two magnitudes fainter than previous bulge time-domain surveys. The survey's stated purpose is detecting low-mass free-floating planets through microlensing, but the same minute-level cadence captures stellar flares, short-period pulsators, and eclipsing systems that hours-long cadences miss. On 0.4% of the catalog the authors ran a pilot search and found one new short microlensing event, two stellar flares, and 24 previously unknown short-period variables, implying that the full release may hold hundreds of flares and thousands of short variables. If these numbers hold, the release gives the community a new tool for studying fast variability in a dense stellar field.","feed_headline":"59 million bulge stars get minute-cadence light curves","feed_subtitle":"DREAMS DR1 goes ~2 mag deeper than prior bulge surveys, opening minute-scale variability and planet searches.","key_machinery":"The argument is carried by difference image analysis (DIA): a deep, stacked master reference image is built from the best-seeing exposures, each science image is convolved with a spatially varying kernel to match the reference's PSF and flux scale, and the subtraction residuals are fitted at catalog positions. The source list starts with a PSF-fitting detection on the stacked z-band reference, then merges overlapping stamps and the two survey fields to a final non-redundant list of 59,372,789 stars. Forced photometry on the difference images gives the time series; quality metrics flag unreliable epochs. The cadence design — 42–60 second z-band exposures grouped into blocks that yield an effe","core_discovery":"DREAMS DR1 claims to be the deepest and densest public time-series catalog of the Galactic bulge: 59,372,789 stars with light curves from 1,856 z-band and 325 r-band exposures taken in 2025, reaching z_AB ≈ 22. The photometry comes from forced PSF fitting on difference images built against a 20-exposure stacked reference, with per-epoch quality flags. The paper demonstrates the data's value by showing that a known ~48-minute blue large-amplitude pulsator is fully phase-covered in a single night and that a known transiting system shows a newly revealed secondary eclipse and out-of-transit variations. A pilot sliding-box search over 0.04 square degrees (245,490 stars) yields one new microlensi","pith_inferences":["The pilot area was chosen because it visibly contained a new microlensing event, so the flare and variable counts it yields are likely inflated; an identical search over randomly selected 0.04 deg² patches would give a defensible rate for the full survey. (My inference.)","Because the r-band cadence in 2025 was only about 1 hr⁻¹, color information for minute-scale events will be sparse in DR1; the planned interleaved r-band strategy from 2026 onward is what will let flares and short events be separated from truly achromatic microlensing by color. (My inference.)","The reported duplicate fraction in the overlapping 1 deg² region (about 40% of expected duplicates merged) implies some faint stars are counted twice; population counts based on DR1 should treat the overlapping region carefully until a deeper reference catalog is incorporated. (My inference.)"],"forward_implications":["If the pilot event rate is representative, DR1 contains hundreds of stellar flares and thousands of previously unknown short-period variables; a full search of the released light curves would be the direct test.","The same data can recover the periods of short-period pulsators such as blue large-amplitude pulsators from a single night, so DR1 alone can measure period changes on day-to-month timescales.","The two-magnitude gain in depth over earlier bulge time-domain surveys brings intrinsically faint sources — including K-dwarf microlensing sources and faint short variables — into the monitored population.","The public portal, with finding charts and color-magnitude diagrams for each source, lets other surveys quickly characterize transient alerts against deep bulge photometry.","The revealed secondary eclipse in the known transiting system shows that minute-cadence data can update or correct the classification of previously cataloged systems."],"fun_headline_variants":["Deepest bulge time-series catalog: 59M stars","Minute-cadence DREAMS DR1 catalogs 59M bulge stars","2 mag deeper: DREAMS DR1 tracks 59M stars","Pilot search finds new microlensing and flares in bulge"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The pilot search region was deliberately chosen because a visual inspection had already revealed a microlensing event there, so the extrapolation that the full DR1 holds hundreds of flares and thousands of short variables assumes this small patch is representative of the whole 5 deg² survey — an assumption the paper itself cautions against.","fun_headline_variants_meta":{"raw":{"variants":["Deepest bulge time-series catalog: 59M stars","Minute-cadence DREAMS DR1 catalogs 59M bulge stars","2 mag deeper: DREAMS DR1 tracks 59M stars","Pilot search finds new microlensing and flares in bulge"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00069,"raw_usage":{"total_tokens":3052,"prompt_tokens":927,"completion_tokens":2125,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":671,"completion_tokens_details":{"reasoning_tokens":2048}},"tokens_in":671,"tokens_out":2125,"duration_ms":18525,"temperature":1.0,"reasoning_tokens":2048,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T12:57:23.168764+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same sliding-box search (at least five consecutive >3σ points, Δχ²>300) on a randomly selected 0.04 deg² patch that had no prior visual inspection, and compare the yields of flares and short-period variables to the pilot's 2 flares and 31 variables per 245,490 stars. A significantly lower yield would falsify the extrapolated full-catalog counts; a comparable yield would support them.","supporting_citations":[],"review_version":2}