TRICERATOPS: Bayesian Exoplanet Vetting

• TRICERATOPS is a Bayesian framework that models transit signals on target and neighboring stars to assess false positive probabilities.
• It leverages TIC and Gaia data with Monte Carlo sampling and contrast curves to yield robust FPP and NFPP metrics.
• Its flexible application—from bulk candidate triage to detailed system-specific analyses—underscores its utility in crowded stellar fields.

TRICERATOPS is a Bayesian vetting and statistical validation framework for transiting exoplanet candidates, developed for TESS Objects of Interest and designed to estimate whether a transit-like signal is more likely to be produced by a bona fide planet or by an astrophysical false positive (Giacalone et al., 2020). Its defining feature is explicit multi-star modeling: it evaluates scenarios on the target star, on unresolved bound or background/foreground companions, and on resolved nearby stars identified in Gaia and the TIC, and it reports both a conventional false positive probability (FPP) and a nearby false positive probability (NFPP) (Giacalone et al., 2020). This architecture is motivated by the TESS observing regime, in which pixels are $21$ arcsec and contamination by nearby sources is common (Giacalone et al., 2020).

1. Origin, scope, and design context

TRICERATOPS was introduced as a fast Bayesian tool tailored to TESS, whose large pixels and crowded apertures make explicit contamination modeling a central part of transit vetting (Giacalone et al., 2020). In its original presentation, it was described as the only TESS vetting and validation tool that models transits from nearby contaminant stars in addition to the target star, rather than assuming that nearby-star pathways have already been excluded by separate pre-vetting (Giacalone et al., 2020). The framework works directly with TIC and Gaia stellar properties, aperture definitions, physical transit models, and optional follow-up constraints such as high-resolution imaging contrast curves (Giacalone et al., 2020).

The original benchmark applied the method to 68 previously classified TOIs and then to 384 unclassified TOIs. In that application, TRICERATOPS statistically validated 12 objects, classified 125 as likely planets, and classified 52 as likely nearby false positives (Giacalone et al., 2020). Later studies extended its use from bulk triage to system-specific statistical validation, including surveys of hot Neptunes, K-star hosts, and super-Earth candidates, as well as detailed analyses of individual TOIs with extensive ground-based follow-up (Magliano et al., 2022, Mistry et al., 2023, Mistry et al., 2023).

This broad usage reflects a specific niche. TRICERATOPS is not principally an instrumental false-alarm detector; rather, it is a probabilistic astrophysical scenario comparator. As a result, it is routinely embedded in larger vetting pipelines that also inspect centroid motion, odd-even depth differences, secondary eclipses, archival imaging, and follow-up photometry (Giacalone et al., 2020, Magliano et al., 2022).

2. Bayesian formulation and hypothesis space

The framework computes posterior probabilities for competing scenarios $S_j$ using Bayes’ theorem. In the original formulation, the posterior is written as $p(S_j\mid D)\propto p(S_j)\,p(D\mid S_j)$, with normalized scenario probabilities

$$\mathcal{P}_j=\frac{p(S_j\mid D)}{\sum_j p(S_j\mid D)}.$$

The marginal likelihood for each scenario is estimated by Monte Carlo sampling, typically with $N=10^6$, and the transit-data likelihood is evaluated from modeled fluxes relative to the observed phase-folded light curve (Giacalone et al., 2020). Later papers retain the same Bayesian structure, often expressing it as

$$P(H_i\mid D)=\frac{P(D\mid H_i)\,P(H_i)}{\sum_j P(D\mid H_j)\,P(H_j)}.$$

(Escolà-Rodrigo, 16 Feb 2026, Barrientos et al., 4 Aug 2025)

In the original paper, TRICERATOPS enumerates 18 astrophysical scenarios (Giacalone et al., 2020). These include target-star cases without unresolved companions, cases with unresolved bound companions, cases with unresolved background/foreground stars, and scenarios on resolved nearby stars. The scenario vocabulary used throughout the later literature centers on a stable subset of labels: TP for a transiting planet on the target, PTP for a planet on the primary when an unresolved bound companion is present, DTP for a diluted transiting planet on the target in the presence of a blended background/foreground star, NTP for a transiting planet on a nearby resolved star, and EB-, HEB-, and BEB-like families for eclipsing-binary false positives (Giacalone et al., 2020, Mann et al., 2022, Barrientos et al., 4 Aug 2025).

The summary statistics are derived from these scenario posteriors. In the original paper,

$$\mathrm{FPP}=1-(\mathcal{P}_{\mathrm{TP}}+\mathcal{P}_{\mathrm{PTP}}+\mathcal{P}_{\mathrm{DTP}})$$

and

$$\mathrm{NFPP}=\sum(\mathcal{P}_{\mathrm{NTP}}+\mathcal{P}_{\mathrm{NEB}}+\mathcal{P}_{\mathrm{NEBx2P}}).$$

(Giacalone et al., 2020) Later applications often use the equivalent compact interpretation that FPP is the posterior mass assigned to all false-positive scenarios, while NFPP isolates the nearby-source subset (Mann et al., 2022, Escolà-Rodrigo, 16 Feb 2026).

A recurring feature of published analyses is that low TP probability does not imply high FPP if posterior mass is concentrated in other planet-containing scenarios. This is stated explicitly for TOI-7701.01, where TP accounts for $78.1\%$ of the posterior, but total planet-containing configurations account for $99.7\%$, leaving $S_j$0 and $S_j$1 (Escolà-Rodrigo, 13 Jun 2026).

3. Inputs, priors, and forward modeling

TRICERATOPS requires the transit light curve, ephemeris, an aperture definition, host-star properties, and a catalog of nearby stars. In the original implementation, users provide the TIC ID, phase-folded TESS light curve and uncertainty, the observed transit depth $S_j$2, the ephemeris, and the extraction aperture for each sector; the code then queries nearby stars within a 10-pixel radius and computes per-star flux contributions $S_j$3 using a Gaussian approximation to the PRF with $S_j$4 pixels (Giacalone et al., 2020). The corresponding blending relation is

$S_j$5

so stars with $S_j$6 are too faint to generate the observed signal (Giacalone et al., 2020).

The transit and eclipse models are batman-based physical light-curve models with quadratic limb darkening, and the original implementation assumes circular orbits and fixed $S_j$7 for speed and tractability (Giacalone et al., 2020). Likelihoods are combined with priors on orbital period, planet radius, mass ratio, background stellar population, and multiplicity. For unresolved background/foreground blends, simulated stars are drawn from TRILEGAL in a $S_j$8 cone centered on the target (Giacalone et al., 2020).

High-resolution imaging enters through a weighting term on scenario likelihoods, suppressing models that require undetected companions inconsistent with the supplied contrast curve (Giacalone et al., 2020). This mechanism became standard in later applications. Studies such as the validation of TOI-1221 b explicitly fed SOAR/HRCam and Gemini South/Zorro contrast curves into TRICERATOPS, while also using TIC and Gaia information, empirical background-star rates, and multiplicity priors (Mann et al., 2022). Other works supplied Gaia DR3 source environments, propagated source positions, PRF/aperture maps, centroid constraints, and follow-up multi-band photometry to refine dilution and off-target probabilities (Escolà-Rodrigo, 13 Jun 2026, Peláez-Torres et al., 17 Apr 2026).

A central practical distinction in the literature is between photometry prepared for geometric parameter inference and photometry prepared for validation. In the validation of TOI-7701.01, PDCSAP was used for geometric radius inference, while re-normalized un-detrended SAP was deliberately supplied to TRICERATOPS so that field dilution and crowding would remain encoded in the data stream used for false-positive modeling (Escolà-Rodrigo, 13 Jun 2026). This suggests that TRICERATOPS is often most informative when the photon environment is preserved rather than aggressively flattened away.

4. Decision metrics, thresholds, and workflow placement

The original FPP–NFPP decision plane defines three standard classes for 2-minute cadence TESS data: validated planets with $S_j$9 and $p(S_j\mid D)\propto p(S_j)\,p(D\mid S_j)$0, likely planets with $p(S_j\mid D)\propto p(S_j)\,p(D\mid S_j)$1 and $p(S_j\mid D)\propto p(S_j)\,p(D\mid S_j)$2, and likely nearby false positives with $p(S_j\mid D)\propto p(S_j)\,p(D\mid S_j)$3 (Giacalone et al., 2020). The same thresholds are widely reused in later target-by-target validation papers (Mann et al., 2022, Escolà-Rodrigo, 16 Feb 2026, Peláez-Torres et al., 17 Apr 2026). The original study also states that performance is best for $p(S_j\mid D)\propto p(S_j)\,p(D\mid S_j)$4, and gives a typical runtime of about 5 minutes per target on a standard 2-core laptop (Giacalone et al., 2020).

In practice, TRICERATOPS is rarely used in isolation. A common workflow is front-end vetting followed by probabilistic validation. One systematic hot-Neptune study first used DAVE for visual and centroid-level vetting, then ran TRICERATOPS on the surviving candidates, and later re-ran it with follow-up imaging constraints; ARIADNE provided refined stellar parameters and juliet provided physically consistent transit fits (Magliano et al., 2022). VaTEST II combined LATTE, TESS-Plot, high-resolution imaging, and TRICERATOPS to validate 11 K-star planets, while VaTEST III pre-cleared nearby stars with ground-based photometry and then set $p(S_j\mid D)\propto p(S_j)\,p(D\mid S_j)$5 before running TRICERATOPS on eight super-Earth candidates (Mistry et al., 2023, Mistry et al., 2023).

Several studies report repeated runs to mitigate stochasticity. The hot-Neptune survey used 10 runs in its initial pass and 30 runs in its final validation stage (Magliano et al., 2022). VaTEST II used 15 iterations per target per contrast-curve file (Mistry et al., 2023). The upgraded multicolor framework reports mean values and 68% confidence intervals over 10 runs, noting the higher computational load of multiband likelihood evaluation (Barrientos et al., 4 Aug 2025). Target-specific papers also used 20 independent runs to test robustness against MCMC stochasticity (Escolà-Rodrigo, 16 Feb 2026, Greklek-McKeon et al., 10 Dec 2025).

The threshold values are not immutable. In the hot-Neptune desert, one study adopted a stricter validation criterion of $p(S_j\mid D)\propto p(S_j)\,p(D\mid S_j)$6 and $p(S_j\mid D)\propto p(S_j)\,p(D\mid S_j)$7, arguing that the default TRICERATOPS planet occurrence priors treat period and radius as independent and can therefore underestimate FPP in a sparse-occurrence region (Magliano et al., 2022). VaTEST III, by contrast, validated planets at $p(S_j\mid D)\propto p(S_j)\,p(D\mid S_j)$8 with $p(S_j\mid D)\propto p(S_j)\,p(D\mid S_j)$9 after pre-clearing nearby stars (Mistry et al., 2023).

5. TRICERATOPS+ and the multicolor extension

TRICERATOPS+ is an upgraded version that incorporates ground-based light curves in separate bandpasses together with TESS photometry and high-resolution imaging (Barrientos et al., 4 Aug 2025). Its core modification is the total likelihood,

$$\mathcal{P}_j=\frac{p(S_j\mid D)}{\sum_j p(S_j\mid D)}.$$0

which allows the framework to use chromatic information directly in model comparison (Barrientos et al., 4 Aug 2025). The approach exploits the approximate achromaticity of a planetary transit, aside from weak limb-darkening dependence, against the intrinsically chromatic depths expected from many eclipsing-binary blends (Barrientos et al., 4 Aug 2025).

The bandpass-specific dilution relation is written as

$$\mathcal{P}_j=\frac{p(S_j\mid D)}{\sum_j p(S_j\mid D)}.$$1

and TRICERATOPS+ supplements the original TESS-only flux-ratio machinery with bandpass-specific stellar limb darkening from ExoTiC-LD and flux-ratio splines in SDSS $$\mathcal{P}_j=\frac{p(S_j\mid D)}{\sum_j p(S_j\mid D)}.$$2 and 2MASS $$\mathcal{P}_j=\frac{p(S_j\mid D)}{\sum_j p(S_j\mid D)}.$$3, constructed using a TRILEGAL synthetic population at 100 pc (Barrientos et al., 4 Aug 2025). In this upgraded framework, users can also supply the fractional flux of resolved contaminants in ground-based apertures (Barrientos et al., 4 Aug 2025).

The 2025 multicolor validation study applied TRICERATOPS+ to 14 TESS candidates and statistically validated six new planets in five systems: TOI-1346 b, TOI-1346 c, TOI-2719 b, TOI-4155 b, TOI-6000 b, and TOI-6324 b (Barrientos et al., 4 Aug 2025). The same paper showed that multicolor photometry can either lower FPP and drive $$\mathcal{P}_j=\frac{p(S_j\mid D)}{\sum_j p(S_j\mid D)}.$$4, as for TOI-6000 b, or reveal chromaticity indicative of false positives, as for TOI-1254.01, TOI-1616.01, TOI-4051.01, and TOI-5706.01 (Barrientos et al., 4 Aug 2025).

A system-specific demonstration appears in the validation of TOI-2267 d in a $$\mathcal{P}_j=\frac{p(S_j\mid D)}{\sum_j p(S_j\mid D)}.$$5 M-dwarf binary. There, TRICERATOPS+ combined a stacked TESS profile, two diffuser-assisted Hale/Palomar WIRC $$\mathcal{P}_j=\frac{p(S_j\mid D)}{\sum_j p(S_j\mid D)}.$$6-band transits, and speckle-imaging contrast curves. The reported result was $$\mathcal{P}_j=\frac{p(S_j\mid D)}{\sum_j p(S_j\mid D)}.$$7 with a 68% range of $$\mathcal{P}_j=\frac{p(S_j\mid D)}{\sum_j p(S_j\mid D)}.$$8 and $$\mathcal{P}_j=\frac{p(S_j\mid D)}{\sum_j p(S_j\mid D)}.$$9, while the host-star assignment remained ambiguous despite transit-shape stellar-density analysis (Greklek-McKeon et al., 10 Dec 2025).

6. Empirical behavior, strengths, and limitations

Published applications illustrate how strongly TRICERATOPS responds to contamination information. In TOI-4616 b, an initial TESS-only run on detrended data gave $N=10^6$0 and $N=10^6$1, dominated by nearby eclipsing-binary and nearby transiting-planet scenarios associated with a neighbor in the TESS aperture. Repeating the analysis with uncontaminated ground-based photometry and a NESSI contrast curve removed those pathways, yielding a final $N=10^6$2 and $N=10^6$3, sufficient for validation under the standard criterion (Lang et al., 11 Mar 2026). In TOI-1221 b, the framework returned $N=10^6$4 and $N=10^6$5; the authors emphasized that these values were conservative because TRICERATOPS did not ingest reconnaissance RV constraints or nearby-eclipsing-binary clearances from ground-based photometry (Mann et al., 2022).

Other studies show the same pattern in more favorable environments. TOI-1752 b and c were both validated with low FPP and essentially zero NFPP after combining TESS photometry, Gaia DR3 information, AO contrast curves, SPOC centroid diagnostics, and multicolor follow-up (Peláez-Torres et al., 17 Apr 2026). For TOI 7475.01, a workflow based on natural-flux-preserving light-curve preparation, BLS detection, Gaia DR3 contamination checks, and centroid stability produced 20 TRICERATOPS runs with mean $N=10^6$6 and $N=10^6$7; the posterior mass was concentrated entirely in TP, PTP, and DTP, giving a cumulative planetary probability in the TIC 376866659 system of approximately $N=10^6$8 (Escolà-Rodrigo, 16 Feb 2026). For TOI-7701.01, ensemble TRICERATOPS analysis on re-normalized SAP yielded $N=10^6$9 and $$P(H_i\mid D)=\frac{P(D\mid H_i)\,P(H_i)}{\sum_j P(D\mid H_j)\,P(H_j)}.$$0, while the dominant TP scenario converged to $$P(H_i\mid D)=\frac{P(D\mid H_i)\,P(H_i)}{\sum_j P(D\mid H_j)\,P(H_j)}.$$1, close to the PDCSAP-derived geometric radius of $$P(H_i\mid D)=\frac{P(D\mid H_i)\,P(H_i)}{\sum_j P(D\mid H_j)\,P(H_j)}.$$2 (Escolà-Rodrigo, 13 Jun 2026).

The strengths of the framework are therefore clear in the literature: explicit nearby-source modeling, direct use of Gaia/TIC environments, natural integration of contrast curves, and a quantitative NFPP that is especially relevant for TESS’s large pixels (Giacalone et al., 2020). The multicolor extension adds a direct chromaticity test that is often decisive for crowded or blended systems (Barrientos et al., 4 Aug 2025).

The limitations are equally consistent across studies. TRICERATOPS focuses on astrophysical scenarios rather than instrumental/systematic false alarms (Giacalone et al., 2020, Barrientos et al., 4 Aug 2025). The original implementation assumes circular orbits and depends on accurate stellar parameters, catalog completeness, and the Gaussian PRF approximation (Giacalone et al., 2020). Several papers note that results depend on priors and on the completeness of the background-star model, and that some external constraints—especially RV limits and certain neighbor clearances—may not be ingested directly, which can make the reported FPP conservative or, in occurrence-sparse regimes, optimistic unless stricter thresholds are imposed (Mann et al., 2022, Magliano et al., 2022, Escolà-Rodrigo, 16 Feb 2026). This suggests that TRICERATOPS is most robust when used as the probabilistic core of a broader vetting architecture that includes centroid diagnostics, high-resolution imaging, and targeted ground-based photometry.