Stars, measured.
Beyond planets, SetuHQ detects stellar flares, extracts pulsation frequencies, reads public CHEOPS light curves, and finds eclipsing binaries, rotation periods and transients. Each capability was compared with published results under a protocol committed before its run, and each page and tool output says how far its validation went.
Flares
partly validated
5 of 8 counted checks pass; 1 tuning-set result not counted
Pulsations
partly validated
7 of 9 counted checks pass
CHEOPS light curves
not validated
2 of 5 counted checks pass; 2 comparisons inconclusive (gated)
Protocol v3 (docs/results/stellar_validation_protocol_v3.json, sha256 c978f04c24b8…), frozen 2026-10-22, committed in b2ac54c26149 before its run. Rule: validated when every counted check passes; partly validated when more than half do; otherwise not validated. Protocol v3 re-tests the targets earlier versions used, so it is a re-test after review fixes, not an independent validation.
What the detectors see
Every counted check
Each check compares SetuHQ with a published result; where they disagree, the check fails and stays on this list. A tuning-set result was used to set a threshold and is not counted; a gated comparison whose gate failed is inconclusive.
| Capability | Check (protocol v3) | Value | Result |
|---|---|---|---|
| flares | AU Mic S1 (tuning star): recall of Günther et al. (2020) flares, 95% interval lower bound >= 0.70 | 0.644 | fail |
| flares | AU Mic S1 flux-inverted null at most 1 event per 10 days: the shape threshold was set on this star and this null, so it is a tuning-set result, not counted | 0 | tuning set (computed: pass) |
| flares | Proxima S11-12: recall of Medina et al. (2020) flares with ED >= 3 s (pooled), 95% interval lower bound >= 0.5 (relaxed after v1) | 0.583 | pass |
| flares | Proxima S11-12: at least 85% of our flares are in Medina et al.'s list | 1 | pass |
| flares | Proxima S11-12: flux-inverted null (real flares masked) at most 1 event per 10 days | 0 | pass |
| flares | Proxima S11-12: TESS-band energy measured as Medina et al. do (data ED x magnitude-route luminosity), ours/theirs, 95% interval inside [0.67, 1.5] | [1.15, 1.51] | fail |
| flares | TIC 150359500: 9000 K bolometric energies computed as Günther et al. do (TIC Teff and radius), ours/theirs, 95% interval inside [0.67, 1.5] | [1.02, 1.29] | pass |
| flares | Quiet stars: fewer than 0.02 detections per day | 0.00679 | pass |
| flares | Injected single flares on Proxima and GJ 1243 (not the tuning star), pooled: the 95% upper bound of the rate that come back as two or more components is at most 15% (tightened after v2) | 0.194 | fail |
| cheops | Gate. CHEOPS DRP M dwarfs: flux-inverted null (real flares masked) at most 10% of the real flare count, with and without denoising | in the CHEOPS file | fail |
| cheops | Gate. CHEOPS DRP M dwarfs: injected flares with S/N >= 10 and FWHM >= 3 cadences recovered, 95% interval lower bound >= 0.8 | in the CHEOPS file | fail |
| cheops | CHEOPS DRP M dwarfs (denoised): complex fraction's 95% interval contains Poyatos et al.'s ~0.40 (only if both gates pass) | in the CHEOPS file | inconclusive (computed: pass) |
| cheops | CHEOPS DRP M dwarfs (denoised): power-law index within 2 sigma (combined) of Poyatos et al.'s 1.92 +/- 0.07 (only if both gates pass) | in the CHEOPS file | inconclusive (computed: pass) |
| cheops | Same input as pycheops: on every reference visit our decorrelation matches pycheops' to < 10 ppm rms, or reaches a lower rms on the same points and terms (relaxed after v1) | in the CHEOPS file | pass |
| cheops | End to end (DACE DRP arrays vs pycheops' DRP product): correlation >= 0.9 on every reference visit | in the CHEOPS file | fail |
| cheops | HR 10 CHEOPS, three public visits together: the two strongest TESS delta Scuti frequencies (f1, f2) found | in the CHEOPS file | pass |
| pulsations | beta Pic S5-7: the 10 strongest of Zieba et al.'s (2019) frequencies all recovered within 3 sigma | 1 | pass |
| pulsations | beta Pic S5-7: at least 80% of the 54 published frequencies within the Rayleigh resolution | 1 | pass |
| pulsations | HR 10 S2 and S29: the 10 strongest of Salmon et al.'s (2024) frequencies recovered within 1/T in both sectors | 1 | pass |
| pulsations | HR 10 S2 and S29: at least 80% of the published frequencies within 1/T | 0.742 | fail |
| pulsations | HR 10 S29 residual: AR(1) surrogates give fewer than 1 false frequency per light curve on average | 0.6 | pass |
| pulsations | HR 10 S2 and S29: a random frequency would be flagged as a combination at most 15% of the time | 0.028 | pass |
| pulsations | HR 10 S29: granulation characteristic frequency from Salmon et al.'s estimator (averaged one-day periodograms, one Harvey profile) within a factor 1.5 of their 150.87 microHz | 0.463 | fail |
| pulsations | TOI-197 S2: numax within 2 sigma of Huber et al. (2019) | 0.17 | pass |
| pulsations | TOI-197 S2: delta-nu within 1 microHz of Huber et al.'s 28.94 (the search window is 0.7-1.4 x the scaling value, about 20-40 microHz) | 0.144 | pass |
Not validated
- Flare energies: on TESS they use the TESS magnitude and distance (Sullivan et al. 2015 zero point) when the TIC has them, else TIC Teff and radius; on CHEOPS (no magnitude route) the TIC radius and, where the TIC has no Teff, the Teff from the CHEOPS archive, labelled. A star with neither route gets energies reported as unavailable.
- AU Mic sector 1 is the tuning star of the flare shape test: its numbers are reported, its null is not counted.
- GJ 1243: no machine-readable per-flare TESS catalogue was found; detections are reported, not compared.
- GJ 65: no per-flare table is public for Bruno et al. (2024); only an internal 120 s vs 20 s check.
- Poyatos et al. (2025) worked on 3-s CHEOPS imagettes reduced with PIPE; DACE serves DRP subarray light curves (about 42 s and longer), so the CHEOPS comparison is of aggregate numbers on different data products, not flare by flare, and it counts only when the CHEOPS null and injection gates pass. Their per-flare catalogue is not public.
- Bruno et al. (2024): no per-flare catalogue; their complex fraction (≳ 30%) is quoted for context only.
- Wavelet denoising: tuned and checked by injection on SetuHQ's own data; Poyatos et al.'s optimal parameters were found on imagettes and are not reproduced here.
- HR 10 CHEOPS frequencies: Salmon et al. (2024) publish no CHEOPS frequency list; only the TESS lists are compared.
Full results: stellar_validation.md. Run the tools yourself: Flares, pulsations and CHEOPS ›
Binaries, rotation and transients
Three more capabilities were tested under protocol v4, written and committed before any of their code existed. Their targets were drawn from published catalogues by seeded rules; part of each sample tuned the methods, and only the held-out rest, run once, is counted.
Eclipsing binaries
partly validated
5 of 8 counted checks pass
Rotation
not validated
3 of 6 counted checks pass
Transients
not validated
1 of 9 counted checks pass
The transient tools are a search aid whose candidates need checking by eye; their detections, class hints and false-alarm screens are not validated.
Protocol v4 (docs/results/stellar_validation_protocol_v4.json, sha256 968ffab34db2…), frozen 2026-10-24, committed on its own in b6f22ebeb1af before any v4 code or run; targets drawn by seeded rules from the catalogues, a tuning split used while the methods were written and a held-out split run once, which alone is counted.
A supernova shines in a distant galaxy, not in a catalogued star, so no star disc is drawn beside it.
Every counted check, protocol v4
Misses stay on this list as misses. An eclipsing binary or a transient SetuHQ finds is a candidate, never new and never confirmed; class and morphology hints are heuristic.
| Capability | Check (protocol v4) | Value | Result |
|---|---|---|---|
| binaries | Held-out Prša et al. (2022) EBs: flagged as eclipsing-binary candidates, 95% interval lower bound >= 0.80 | 0.508 | fail |
| binaries | Held-out EBs: period within 0.5% of the catalogue's (an EB not flagged is a miss), 95% interval lower bound >= 0.70 | 0.508 | fail |
| binaries | Held-out EBs with a period match: primary depth ours/catalogue (2-Gaussian), bootstrap 95% interval of the median inside [0.67, 1.5] | [0.95, 1.10] | pass |
| binaries | Held-out EBs with a catalogue secondary and a period match: secondary phase within 0.01 of the catalogue's, 95% interval lower bound >= 0.70 (a secondary we do not find is a miss) | 0.702 | pass |
| binaries | Held-out EBs with a period match: Spearman rank correlation of our primary width with Wp-2g >= 0.6 | 0.958 | pass |
| binaries | Held-out EBs flagged as candidates: detached or not agrees with Morph < 0.5, 95% interval lower bound >= 0.60 | 0.657 | pass |
| binaries | Seven planet hosts and quiet stars: at most one flagged as an eclipsing-binary candidate | 1 | pass |
| binaries | Synthetic detached EBs injected into the seven contrast light curves: period within 0.5%, 95% interval lower bound >= 0.85 | 0.757 | fail |
| rotation | Held-out Canto Martins et al. (2020) stars: a rotation period detected, 95% interval lower bound >= 0.70 | 0.641 | fail |
| rotation | Held-out stars: period within 10% of the published one (an undetected star is a miss), 95% interval lower bound >= 0.60 | 0.388 | fail |
| rotation | Held-out stars: within 10% of the published period, half or double it, 95% interval lower bound >= 0.75 | 0.508 | fail |
| rotation | Held-out matches: |ours - published| within twice the combined uncertainty for at least 68% | 1 | pass |
| rotation | Three quiet stars (pi Men, zeta Tuc, beta Hyi, sector 1): no rotation period detected | 0 | pass |
| rotation | Synthetic spot modulation (1-12 d, 1-20 ppt) injected into the three quiet stars: period within 10%, 95% interval lower bound >= 0.85 | 0.886 | pass |
| transients | Held-out Fausnaugh et al. (2023) type Ia supernovae: a transient candidate within 10 d of their peak, 95% interval lower bound >= 0.60 | 0.702 | pass |
| transients | Held-out supernovae: peak time within 2 d of theirs (an undetected one is a miss), 95% interval lower bound >= 0.60 | 0.357 | fail |
| transients | Detected held-out supernovae: rise time ours/theirs, bootstrap 95% interval of the median inside [0.67, 1.5] | [0.55, 1.06] | fail |
| transients | Detected held-out supernovae: peak TESS magnitude ours minus theirs, bootstrap 95% interval of the median inside [-0.5, +0.5] mag | [-0.20, 0.85] | fail |
| transients | Detected held-out supernovae: class hint extragalactic-transient-like, 95% interval lower bound >= 0.60 | 0.268 | fail |
| transients | Proxima Cen S11-12: transient candidates matching a Medina et al. (2020) flare, class hint flare-like, 95% interval lower bound >= 0.80 | 0.597 | fail |
| transients | Three quiet stars, sector 1: fewer than 0.05 transient candidates per day not marked as likely false alarms | 0.55 | fail |
| transients | Control apertures beside the held-out supernovae: at most 10% carry an extragalactic-transient-like candidate | 0.233 | fail |
| transients | Synthetic momentum-dump spikes and asteroid-like bumps injected into the quiet stars: marked as likely false alarms, 95% interval lower bound >= 0.80 | 0.628 | fail |
Not validated
- Planet vetting: unchanged. The eclipsing-binary module's vetting hints are advisory, no vetting decision uses them, and they were not evaluated; using them in vetting needs its own pre-registered evaluation.
- Eclipse widths: Prša et al. (2022) give 2-Gaussian widths, SetuHQ a full width at half depth, so only their rank order is compared. The e·cosω values are first-order hints from the secondary's phase, not compared with radial-velocity orbits.
- Rotation: one sector per star, as the protocol fixes; periods longer than about 13 days are not measured, and stitching sectors (with care for sector-to-sector offsets) is not done and not validated. Canto Martins et al. (2020) used every sector available to them.
- Transients: one TESScut sector per supernova, as the protocol fixes. When first light falls before that sector begins, the reference image already holds supernova light, so the rise time and the peak brightness are measured from a non-zero floor; the supernovae compared are all type Ia, so the extragalactic class hint is not tested on tidal disruption events or other transients.
- Scattered light: no labelled real sample was found; the screen (quality flags and control apertures) is described and exercised on synthetic events and the supernova cutouts' control apertures, not validated against a catalogue of scattered-light events.
- Asteroid crossings: screened by shape and, in cutouts, by a moving centroid; no catalogue of TESS asteroid crossings was compared (Fausnaugh et al. 2023's asteroid flags excluded those supernovae from the sample).
Full results: stellar_validation_v4.md. Data: public TESS light curves and full-frame-image cutouts from MAST; published values from Prša et al. (2022), Canto Martins et al. (2020), Fausnaugh et al. (2023) and Medina et al. (2020) through VizieR.
CHEOPS: not validated (2 of 5 checks pass)
SetuHQ reads public CHEOPS light curves (the Data Reduction Pipeline's photometry) through DACE, the University of Geneva's public API in front of the CHEOPS Mission Archive; ESA does not designate DACE. Only visits the archive marks as public are read, with no credentials, and no raw CHEOPS file is stored in this repository or shipped in the package. The CHEOPS capability is not validated: its null and injection gates failed, so two comparisons are inconclusive, and the end-to-end match with pycheops' product failed. The CHEOPS-derived numbers are not on this page: they are in stellar_validation_cheops.md, under ESA's terms.
Licence. CHEOPS data, and outputs derived from them, stay under ESA's terms: data in the ESA Space Science Archives are distributed under the CC BY-NC 3.0 IGO licence, and any commercial use, including use that indirectly generates financial gain, needs ESA's prior authorisation ([email protected]). SetuHQ's commercial licence and SetuHQ Pro cover SetuHQ's code and services, not CHEOPS data or CHEOPS-derived outputs.
SetuHQ is independent and is not affiliated with, endorsed by or operated on behalf of ESA, the CHEOPS mission or the CHEOPS Consortium. No ESA logo, template or figure is used, and nothing here automates an ESA system.
ESA's acknowledgement, the programmes and their credit lines, and the papers
CHEOPS is an ESA mission in partnership with Switzerland with important contributions to the payload and the ground segment from Austria, Belgium, France, Germany, Hungary, Italy, Portugal, Spain, Sweden, and the United Kingdom. The CHEOPS Consortium would like to gratefully acknowledge the support received by all the agencies, offices, universities, and industries involved. Their flexibility and willingness to explore new approaches were essential to the success of this mission. CHEOPS data analysed in this article will be made available in the CHEOPS mission archive (https://cheops.unige.ch/archive_browser/).
Programmes whose public visits the validation used:
CH_PR100010: Credit: ESA, Gyula SZABOCH_PR100018: Credit: ESA, Isabella PAGANOCH_PR130057: Credit: ESA, Gyula SZABO
Cite: Benz, W., et al. 2021, Experimental Astronomy, 51, 109 (the CHEOPS mission); Hoyer, S., et al. 2020, A&A, 635, A24 (the CHEOPS Data Reduction Pipeline); Fortier, A., et al. 2024, A&A, 687, A302 (CHEOPS in-flight performance).
This publication makes use of The Data & Analysis Center for Exoplanets (DACE), which is a facility based at the University of Geneva (CH) dedicated to extrasolar planets data visualisation, exchange and analysis. DACE is a platform of the Swiss National Centre of Competence in Research (NCCR) PlanetS, federating the Swiss expertise in Exoplanet research. The DACE platform is available at https://dace.unige.ch.