LEO-Vetter: Automated TESS Vetting

• LEO-Vetter is an automated system that classifies TESS transit-like signals into statistically robust exoplanet candidate catalogs.
• It employs a suite of quantitative flux-level tests, including SNR thresholds and model comparisons, to ensure reliable candidate selection.
• Its pixel-level diagnostics, such as centroid offset and difference image analysis, verify the on-target signal origin to reduce false positives.

LEO-Vetter is an automated exoplanet candidate vetting system tailored for the Transiting Exoplanet Survey Satellite (TESS) mission. Developed in the tradition of the Kepler Robovetter, its primary goal is to streamline the classification of transit-like signals in TESS light curves into statistically robust catalogs of planet candidates. LEO-Vetter implements a rapid suite of quantitative flux-level and pixel-level tests, achieving both high completeness (91%) and reliability (97%) for simulated datasets. Its public availability provides a framework for catalog creation and thorough occurrence rate studies with TESS data.

1. Methodological Foundations and Objectives

LEO-Vetter is explicitly designed to address two critical limitations of current TESS planet candidate searches:

• The labor-intensive nature of manual vetting, which restricts demographic studies due to time and subjective bias.
• The lack of statistically uniform catalogs, which hinders robust occurrence rate analyses.

The system comprises two principal modules:

• Flux-level tests: These evaluate signal strength, transit morphology, model comparisons (transit vs. null fits), uniqueness, and odd-even transit depth consistency.
• Pixel-level tests: These determine the on-target origin of signals through centroid offset measurements and difference image analysis.

LEO-Vetter supports all major TESS cadences (e.g., 30-min, 10-min, 200-sec) and provides a modular framework for completeness and reliability assessment via injection (simulated planets) and scrambling (false alarms) experiments.

2. Flux-Level Vetting Procedures

The flux-level tests collectively serve to distinguish genuine planetary transits from instrumental noise or astrophysical false positives:

• Signal-to-Noise Ratio (SNR): Candidates must satisfy a minimum threshold (e.g., SNR > 6.2), ensuring robust transit detection.
• Model Comparison: The relative fit quality of a transit model versus a straight-line null is quantified using the Akaike Information Criterion (AIC). Only candidates with a significant AIC improvement are retained.
• Uniqueness Assessment: Using template shift techniques, the system ensures that individual transit events are not replicated elsewhere in the phase-folded light curve, mitigating contamination from repeating systematics.
• Odd-Even Depth Test: Differences between depths of alternate transits (odd vs. even) are computed to identify eclipsing binaries.
• Asymmetry Metric: The significance of ingress/egress asymmetry is computed as:

$$\mathrm{ASYM} = \frac{|q_\mathrm{left} - q_\mathrm{right}|}{\sqrt{\sigma^2_{q_\mathrm{left}} + \sigma^2_{q_\mathrm{right}}}}$$

where $q_\mathrm{left}$ and $q_\mathrm{right}$ represent transit ingress and egress flux quantiles and $\sigma$ denotes their uncertainties.

• Chases Metric: Evaluates the consistency of transit events in the phase-folded light curve, distinguishing persistent signals from isolated outliers.

Thresholds for passing each flux-level test are optimized using differential evolution algorithms on injected transit and scrambled noise datasets.

3. Pixel-Level Diagnostics

LEO-Vetter employs pixel-level analysis to verify that the transit source coincides with the target star:

• Difference Image Analysis: Constructs average in-transit and out-of-transit images, fitting a synthetic Point Spread Function (PSF) model to determine the transit source location.
• Centroid Offset ($\Delta\theta$): The spatial displacement between transit and quiescent image centroids is calculated. Candidates with $\Delta\theta > 15$ arcsec are flagged as off-target, typically indicating contamination from background sources.
• PSF Fitting: Utilizes the TESS Pixel Response Function for sub-pixel localization, enabling accurate discrimination between target and contaminant signals.
• These tests ensure that vetted candidates are not misclassified due to blending effects or background eclipsing binaries.

4. Catalog Completeness and Reliability

Rigorous performance evaluation underpins LEO-Vetter’s utility:

• Completeness: Defined as the fraction of injected planets passing all vetting steps; for the adopted metric thresholds, the system achieves 91% completeness overall and up to 96% for higher SNR events ($\mathrm{SNR} > 12$ and $N_{\mathrm{tr}} \geq 5$).
• Reliability: The fraction of non-planet signals (from scrambled or systematically contaminated light curves) that are rejected; measured at 97% for noise/systematics and approaching 100% for high-SNR, multi-transit events.
• These metrics are essential for credible occurrence rate calculations, allowing users to explicitly account for detection efficiency and contamination when inferring planetary population statistics from TESS catalogs.

5. Large-Scale Application and Scientific Impact

Application to well-defined samples demonstrates the tool’s efficacy:

• In a search of ~200,000 M dwarf light curves, LEO-Vetter reduced ~20,000 initial transit-like events to a catalog of 172 planet candidates after flux-level and pixel-level vetting.
• This reduction facilitates robust demographic studies, particularly in stellar regimes (e.g., M dwarfs) where variability and faintness challenge manual or less sophisticated approaches.
• Thresholds optimized for M dwarfs also generalize to FGK targets, indicating versatility across spectral types.
• By streamlining catalog creation, the system enables uniform statistical analyses that were previously impractical for the full candidate set.

6. User Accessibility, Framework, and Extensibility

LEO-Vetter is implemented in Python, publicly available (e.g., via GitHub), and intended for broad community use:

• Users input their lists of transit-like detections; the framework applies its suite of flux and pixel-level tests and outputs uniformly vetted candidate classifications alongside explanatory quantitative metrics.
• Pass-fail thresholds are adjustable, permitting custom tuning for particular scientific priorities or stellar populations.
• The framework includes support for completeness and reliability assessment via injected transit and scrambled datasets, allowing calibration to new detection pipelines or survey characteristics.
• Transparent output and documentation ensure tractability and integration into existing TESS workflows.

7. Context, Advances, and Implications

Relative to predecessor systems, LEO-Vetter improves the objectivity, statistical rigor, and scalability of TESS planet candidate cataloging:

• The method directly addresses the bias and inefficiency of manual inspection, providing a path to uniform, reproducible candidate lists.
• By incorporating both flux and pixel-level tests, LEO-Vetter increases resilience to instrumental noise, systematics, and astrophysical contaminants.
• Its high completeness and reliability metrics, validated on simulated datasets, equip researchers with the tools necessary for meaningful occurrence rate calculations.

A plausible implication is that adoption of this fully automated system will accelerate demographic studies, facilitate comparative population analyses across surveys, and inform future mission designs where statistical uniformity and scalability are paramount.