Technique-Agnostic Exoplanet Demography (TAED)

• TAED is a framework that merges transit, microlensing, and other exoplanet samples by retaining native observables and modeling detection biases.
• It employs forward modelling using a Galactic stellar population synthesis (e.g., the Besançon Model) to simulate realistic planetary systems and survey conditions.
• TAED leverages the planet-host ratio relation and robust retrieval methods, such as differential evolution, for unified and efficient exoplanet demographic inference.

Technique-Agnostic Exoplanet Demography (TAED) denotes the combination of planet samples from different search methods without biasing by their underlying detection physics. In its Roman-era formulation, TAED is a forward-modelling and retrieval framework intended to combine large microlensing and transit samples for demographic studies, while retaining explicit treatment of selection effects, host-star properties, and Galactic environment (Priyadarshi et al., 29 Sep 2025). The immediate driver is the NASA Nancy Grace Roman Space Telescope, which is expected to discover large numbers of both cold and hot exoplanets across Galactic distances, creating a regime in which method-specific analyses are no longer sufficient for a unified census (Edmondson et al., 30 Jun 2025).

1. Emergence of the multi-technique demographic problem

TAED arose from a longstanding mismatch between the scientific ambition of exoplanet demography and the observational partitioning imposed by discovery technique. The National Academies Exoplanet Science Strategy identified as a primary goal the need “to understand the formation and evolution of planetary systems as products of the process of star formation, and characterize and explain the diversity of planetary system architectures, planetary compositions, and planetary environments produced by these processes,” while also finding that “Current knowledge of the demographics and characteristics of planets and their systems is substantially incomplete” (Group et al., 2023).

That incompleteness is not simply a matter of sample size. Transit photometry, radial velocity, direct imaging, microlensing, and astrometry each probe different regions of planet mass, size, separation, host type, and Galactic location, and they report different native observables and completeness information (Fischer et al., 2015). White-paper discussions of wide-orbit demographics made the same point in operational terms: full synthesis requires integration of multiple detection techniques, but joint analysis is obstructed by parameter mismatches, heterogeneous data products, and the absence of centralized detection-efficiency databases (Bennett et al., 2019).

Roman sharpens this problem. It is described as the first survey able to detect large numbers of both cold and hot exoplanets across Galactic distances, with approximately $1{,}400$ cold exoplanets via microlensing and $200{,}000$ hot, transiting planets (Edmondson et al., 30 Jun 2025). A simple merger of catalogs is therefore inadequate. TAED addresses the need for a single inferential framework in which demographic assumptions are shared, while observables and selection functions remain technique-specific.

2. Technique-agnostic observables and the planet-host ratio relation

A central TAED design choice is to prefer observables that are always measured by a given technique over derived quantities that exist only for restricted subsamples. In the Roman context, microlensing directly measures the planet-to-host mass ratio $q = M_p/M_\star$, whereas transit surveys directly measure the transit depth $\delta = (R_p/R_\star)^2$ (Edmondson et al., 30 Jun 2025). The conventional mass-radius relation (MRR) is less suitable for combining the two because it requires both mass and radius, a condition met by only $\sim 20\%$ of known exoplanets, and it exhibits discontinuities and scatter associated with rocky, icy, and gas-giant regimes (Edmondson et al., 30 Jun 2025).

The proposed alternative is the planet-host ratio relation (PHRR), which directly connects the always-measured quantities $q$ and $\delta$. Using 908 confirmed exoplanets from the NASA Exoplanet Archive, the Roman synthesis study found that transit depth and planet-host mass ratio obey a PHRR that is continuous over all planet scales, and that the relation is improved by including orbital period $P$ and host effective temperature $T_\star$ (Edmondson et al., 30 Jun 2025). Candidate relations of the form $\delta(q,T_\star,P)$ were compared with the Bayesian Information Criterion, which favored power-law dependence on $200{,}000$0 and $200{,}000$1, and broken power-law dependence on $200{,}000$2.

The favored form is

$200{,}000$3

with

$200{,}000$4

and

$200{,}000$5

with all logarithms base 10 (Edmondson et al., 30 Jun 2025).

In that calibration, the favored PHRR achieves a fairly uniform $200{,}000$6 relative precision in $200{,}000$7 for all $200{,}000$8, while approximately $200{,}000$9 of the sample has a transit depth that is strongly under-predicted; around half of these systems are associated with large stars $q = M_p/M_\star$0 potentially subject to Malmquist bias (Edmondson et al., 30 Jun 2025). Within TAED, the significance of the PHRR is methodological rather than merely empirical: it provides a bridge between microlensing and transit observables without requiring full mass-radius characterization.

3. Forward modelling architecture

The TAED retrieval framework is based on forward modelling rather than post hoc homogenization of detected planets. Its stated procedure is to use parameterised model exoplanet demographic distributions to embed planetary systems within a stellar population synthesis model of the Galaxy, enabling internally consistent forecasts to be made for all detection methods that are based on spatio-kinematic system properties (Priyadarshi et al., 29 Sep 2025).

In the first Roman-era implementation, the stellar backbone is the Besançon Galactic Model (BGM), which generates a synthetic catalog of stars with realistic distributions of distance, luminosity, temperature, metallicity, and related properties (Priyadarshi et al., 29 Sep 2025). Simulated BGM stars are matched to real survey targets with the similarity metric

$q = M_p/M_\star$1

so that host-star realism is inherited from an explicit Galactic population model rather than from an abstract target list alone (Priyadarshi et al., 29 Sep 2025).

Planet occurrence is then parameterized at the host level. Each host star is assigned a number of planets drawn from a Poisson distribution with mean $q = M_p/M_\star$2, while the core “universal” parameter is the planet-to-host mass ratio $q = M_p/M_\star$3, usually modelled as a broken power law,

$q = M_p/M_\star$4

The semi-major axis, or projected separation for microlensing, is likewise modelled with a broken power law in $q = M_p/M_\star$5 (Priyadarshi et al., 29 Sep 2025).

This architecture is explicitly multi-technique. It stores stellar population and spatial information, generates planet properties independently of detection method, and only afterwards computes the relevant observables—such as periods and radii for transits, or projected separation and mass ratio for microlensing (Priyadarshi et al., 29 Sep 2025). Sensitivity bias is therefore not imposed at the population-generation stage but emerges from the subsequent application of realistic detection-efficiency curves.

4. Retrieval methodology and validation

The TAED framework evaluates demographic models by comparing synthetic observed planet distributions against real or simulated data in observable space, such as histograms in period-radius planes (Priyadarshi et al., 29 Sep 2025). Several retrieval strategies were benchmarked in the first validation study: nested sampling, uniform random sampling, a two-stage machine-learning emulator, and differential evolution (DE).

The study reports strong performance differences. Nested sampling produced robust posteriors but required $q = M_p/M_\star$6–$q = M_p/M_\star$7 s for a 7-parameter problem; uniform random sampling required about $q = M_p/M_\star$8 s; the two-stage machine-learning method was very fast but carried a risk of sampling artifacts; and DE was selected as the preferred method because it provided an excellent balance between accuracy, speed, and scalability, with typical runtimes of $q = M_p/M_\star$9–$\delta = (R_p/R_\star)^2$0 s for problems requiring millions of models and with high parallelizability (Priyadarshi et al., 29 Sep 2025).

The DE implementation uses a parallel initial population with Sobol sampling, a robust strategy for mutation and selection, and a halting criterion based on stabilization of the population’s likelihood spread,

$\delta = (R_p/R_\star)^2$1

In the Kepler-like validation tests, the framework successfully recovered the injected demographic parameters for all tested ground truths, with the true values lying within the $\delta = (R_p/R_\star)^2$2 posterior of the recovered values (Priyadarshi et al., 29 Sep 2025).

A common misconception is that a technique-agnostic framework is equivalent to a detection-method-blind framework. The retrieval paper argues the opposite: host property correlations, location effects, and realistic sensitivity biases are built in, and demographic consistency is achieved by explicit modelling rather than by suppressing instrumental or survey-specific structure (Priyadarshi et al., 29 Sep 2025).

5. Metadata, detection efficiencies, and reproducibility

TAED depends on survey meta-data as much as on demographic parameterization. A major obstacle identified by the Exoplanet Program Analysis Group Science Interest Group 2 is “the lack of comprehensive meta-data accompanying published exoplanet surveys,” which was described as a “significant roadblock” for robust and reproducible demographics analyses (Group et al., 2023). The same document provides guidance on the most valuable data products for transit, radial velocity, direct imaging, microlensing, and astrometry, directed at survey architects, authors, referees, and funding agencies (Group et al., 2023).

The need for standardized survey products is reinforced by multi-technique comparisons in wide-orbit demographics. One difficulty is that microlensing often reports planet-star mass ratio $\delta = (R_p/R_\star)^2$3, whereas radial-velocity and imaging studies often report planet mass, and the conversion is not always possible at the level required for occurrence-rate synthesis because individual-target detection efficiencies are frequently unavailable (Bennett et al., 2019). The consequence is not merely inconvenience; it blocks re-expression of one survey in the native coordinates of another.

Microlensing demographic work provides a concrete example of why this matters. In the retrospective MOA analysis, weak planetary signals were argued to be essential for accurate occurrence rates, because statistical analyses must include such planets rather than discarding them for poor physical characterization (Ranc et al., 2021). Even when host mass and physical separation are weakly constrained, the accurately measured planet-to-host star mass ratio can still enter the statistical analysis of cold planet demography detected by microlensing (Ranc et al., 2021). TAED inherits this logic: completeness requires retention of the native observable space and the corresponding efficiency model, not only a curated subset of well-characterized planets.

6. Scientific reach, limitations, and adjacent developments

The principal scientific promise of TAED is a unified view of planetary architectures across Galactic distances. Roman’s combined hot and cold samples provide the first prospect of analysing transiting and microlensing planets over similar volumes and host populations, which in turn supports demographic mapping of exoplanet occurrence and architecture on a Galactic scale (Edmondson et al., 30 Jun 2025). The retrieval study states that such a framework enables joint demographic inferences from heterogeneous samples while modelling biases due to host properties, distance, method sensitivity, and Galactic environment self-consistently (Priyadarshi et al., 29 Sep 2025).

TAED is also meant to constrain formation and evolution models by making cross-technique trends statistically comparable. Related population-level work has argued for technique-agnostic study of radii, orbits, and atmospheric features as functions of precisely determined host-star ages, with the aim of testing theories of contraction, inflation, migration, and atmospheric evolution through observed time trends (Christiansen et al., 2019). This suggests that TAED is not restricted to Roman-specific retrieval machinery; it is part of a broader move toward host-aware, survey-aware, population-level inference.

Its present limitations are explicit. The first retrieval tests used generic, simple, often uncorrelated demographic models, and the PHRR calibration sample is affected by biases because it is RV- and transit-heavy (Priyadarshi et al., 29 Sep 2025). The PHRR study accordingly argues that Roman will need to self-calibrate a tailored relation using its own survey, detection-efficiency characterization, and host measurements for a subset of planets (Edmondson et al., 30 Jun 2025). A plausible implication is that TAED will remain an iterative programme in which the demographic parameterization, the Galactic forward model, and the observable-bridging relations are refined together as survey meta-data and cross-technique overlap improve.