FLAMINGO Hydrodynamic Simulations

• FLAMINGO Hydrodynamic Simulations is a suite of cosmological simulations that combine large volumes with calibrated baryonic physics, explicit neutrino treatments, and on-the-fly lightcone outputs.
• The simulations employ advanced SPH methods and machine-learning calibration to model galaxy formation, stellar and AGN feedback, and baryonic suppression in large-scale structures.
• FLAMINGO provides extensive public data products, enabling robust predictions for weak lensing, SZ, CMB lensing, and high-redshift galaxy and cluster studies.

FLAMINGO—“Full-hydro Large-scale structure simulations with All-sky Mapping for the Interpretation of Next-Generation Observations”—is a suite of cosmological hydrodynamical simulations developed by the Virgo Consortium to connect galaxy-formation modelling with large-scale-structure cosmology. The suite is designed primarily to support late-time Universe probes—weak lensing, SZ, CMB lensing—and studies of rare objects such as galaxy clusters, while also modelling the galaxy population on large scales. Its defining combination is large volume, calibrated baryonic physics, explicit neutrino treatments, and on-the-fly lightcone products, with subgrid prescriptions for stellar and AGN feedback calibrated to the observed low-redshift galaxy stellar mass function and cluster gas fractions (Schaye et al., 2023).

1. Project remit and run hierarchy

FLAMINGO was introduced to ensure that hydrodynamical simulations are sufficiently realistic for studies of large-scale structure while retaining enough volume to model survey observables and sufficiently flexible physics to explore uncertainties in galaxy formation. The scientific goals stated for the suite include robust predictions for large-scale-structure observables at $k\approx0.1$–$10\,h\,\mathrm{Mpc}^{-1}$ and $\ell\approx10^2$–$10^4$, provision of light-cone outputs for lensing, SZE, X-ray, dark matter, gas, stars, black holes, and neutrinos, and direct study of baryonic effects and neutrino free-streaming on the halo mass function and matter power spectrum (Schaye et al., 2023).

The hydrodynamical component consists of cubic boxes of side length $L=1\,\mathrm{Gpc}$ at three resolutions and a large $2.8\,\mathrm{Gpc}$ box at intermediate resolution. The fiducial hydrodynamical simulations span three numerical resolutions that have each been calibrated to reproduce the present-day galaxy stellar mass function and gas fractions in low-redshift clusters. The public release contains 22 hydrodynamical simulations and 16 gravity-only simulations, including the $10080^3$ particles FLAMINGO-10k run, with initial conditions that match those of the corresponding hydrodynamical runs (Helly et al., 27 Apr 2026).

Run	Volume and resolution	Role in the suite
L1_m8	$(1\,\mathrm{Gpc})^3$, $m_{\rm gas}=1.34\times10^8\,M_\odot$	High-resolution galaxy-formation and high-$z$ studies
L1_m9	$10\,h\,\mathrm{Mpc}^{-1}$0, $10\,h\,\mathrm{Mpc}^{-1}$1	Intermediate-volume fiducial and feedback/cosmology variants
L2p8_m9	$10\,h\,\mathrm{Mpc}^{-1}$2, $10\,h\,\mathrm{Mpc}^{-1}$3	Flagship large-volume hydrodynamical run

At fixed “D3A” cosmology, $10\,h\,\mathrm{Mpc}^{-1}$4 resolution, and $10\,h\,\mathrm{Mpc}^{-1}$5, the project includes gas-fraction variants $10\,h\,\mathrm{Mpc}^{-1}$6, an SMF variant $10\,h\,\mathrm{Mpc}^{-1}$7, a combined $10\,h\,\mathrm{Mpc}^{-1}$8 and $10\,h\,\mathrm{Mpc}^{-1}$9 model, and two Jet variants using kinetic AGN feedback. Four additional cosmology variants are provided at the same resolution and box size, including Planck, PlanckNu0p24Var, PlanckNu0p24Fix, and LS8 (Schaye et al., 2023).

The suite is organized around a DES Y3–based fiducial flat $\ell\approx10^2$0CDM cosmology. In the flagship suite description this is quoted as $\ell\approx10^2$1, $\ell\approx10^2$2, $\ell\approx10^2$3, $\ell\approx10^2$4, $\ell\approx10^2$5, $\ell\approx10^2$6, and $\ell\approx10^2$7, with additional Planck-based, low-$\ell\approx10^2$8, and varied-neutrino cosmologies implemented in matched runs (Schaye et al., 2023).

2. Numerical realization

FLAMINGO is run with the open-source SWIFT code and uses the SPHENIX flavour of modern SPH for gas dynamics. Across the project descriptions, the gravity solver is specified as a Tree–Particle–Mesh or 4th-order Fast Multipole Method plus PM implementation, while the hydrodynamics uses Wendland or quintic-kernel SPHENIX formulations with pressure-entropy coupling, conductivity, viscosity and time-step limiters, and adaptive individual timesteps (Kugel et al., 2023).

Massive neutrinos are followed explicitly with particles using the $\ell\approx10^2$9 method. In the public description, the number of neutrino particles is smaller than the CDM count by a factor of $10^4$0, and in the largest hydrodynamic box the particle load is $10^4$1 for dark matter plus initial gas together with $10^4$2 neutrino particles. The flagship $10^4$3 run follows $10^4$4 total particles to $10^4$5 (Schaye et al., 2023).

The initial conditions are generated with monofonIC using 3-fluid 3LPT in $10^4$6-body gauge with back-scaling for massive neutrinos and panphasia phase-fixed initial conditions, with a starting redshift of $10^4$7 in the flagship description. Specific clustering analyses of FLAMINGO also describe initial conditions set at $10^4$8 using second-order Lagrangian perturbation theory together with a pair-fixed Gaussian random field in which amplitudes of each $10^4$9-mode are held equal to their ensemble mean and phases are chosen in paired opposite sets, reducing sample variance by up to $L=1\,\mathrm{Gpc}$0 (Schaye et al., 2023, Contreras et al., 2024).

Halo and subhalo finding in the main analysis pipeline is based on VELOCIraptor and SOAP. VELOCIraptor first runs a standard 3D Friend-of-Friend with linking length $L=1\,\mathrm{Gpc}$1 mean interparticle spacing to find parent haloes, then an iterative 6D FoF in phase space to isolate self-bound substructures. SOAP computes spherical-overdensity masses and galaxy properties such as stellar masses, star-formation rates, and metallicities in fixed 3D apertures of $L=1\,\mathrm{Gpc}$2 proper, excluding unbound particles and particles belonging to other substructures. In the data release, merger trees and cross-matching catalogues are supplied through HBT-HERONS alongside SOAP-based object catalogues (Contreras et al., 2024, Helly et al., 27 Apr 2026).

3. Baryonic physics and machine-learning calibration

All FLAMINGO hydrodynamical runs include radiative cooling, star formation, stellar mass loss and the resulting chemical enrichment, supernova feedback, and two implementations of AGN feedback. Radiative cooling and photo-heating are computed element by element using CLOUDY-based or Ploeckinger & Schaye tables, with photo-ionisation, self-shielding, dust, cosmic rays, and UV/X-ray backgrounds included in the suite descriptions. Star formation follows the pressure-law prescription of Schaye & Dalla Vecchia, calibrated to reproduce the Kennicutt–Schmidt relation above a metallicity-dependent density threshold (Kugel et al., 2023).

In the calibration paper, the star-formation law is written

$L=1\,\mathrm{Gpc}$3

with $L=1\,\mathrm{Gpc}$4, $L=1\,\mathrm{Gpc}$5, $L=1\,\mathrm{Gpc}$6, and $L=1\,\mathrm{Gpc}$7. Stellar evolution and enrichment track multiple elements, with yields from SN II, SN Ia, and AGB channels, while stellar feedback is implemented through stochastic kinetic kicks or thermal injection depending on the run description, with core-collapse supernova energetics normalized to $L=1\,\mathrm{Gpc}$8 per event (Kugel et al., 2023).

Black holes are seeded in sufficiently massive haloes and grow through a boosted Bondi–Hoyle–Lyttleton prescription capped at the Eddington rate. In the calibration paper this is written

$L=1\,\mathrm{Gpc}$9

with $2.8\,\mathrm{Gpc}$0 and $2.8\,\mathrm{Gpc}$1 a calibrated parameter. The fiducial thermal AGN model stores a fraction $2.8\,\mathrm{Gpc}$2 of the rest-mass energy of accreted gas and injects it once enough energy is accumulated to heat one neighbour by $2.8\,\mathrm{Gpc}$3. FLAMINGO also includes a kinetic jet implementation in which pairs of neighbours are kicked along the BH spin axis with target velocity $2.8\,\mathrm{Gpc}$4 (Kugel et al., 2023).

A distinctive feature of FLAMINGO is the use of Gaussian-process emulators trained on Latin hypercubes of 32 smaller-volume simulations to calibrate the stellar and AGN feedback models. The training boxes are $2.8\,\mathrm{Gpc}$5 for high resolution, $2.8\,\mathrm{Gpc}$6 for intermediate resolution, and $2.8\,\mathrm{Gpc}$7 for low resolution, each with $2.8\,\mathrm{Gpc}$8 particles. The emulators model how the galaxy stellar mass function and cluster gas fractions change as a function of the subgrid parameters and are then fit to observational data with MCMC, while allowing for observational random errors and biases in stellar masses, cosmic variance, and hydrostatic masses (Kugel et al., 2023).

The calibration data comprise the $2.8\,\mathrm{Gpc}$9 GAMA DR4 stellar mass function and cluster gas fractions from X-ray and weak-lensing samples at low redshift. The fitting machinery explicitly includes a systematic stellar-mass bias $10080^3$0, a cosmic-variance shift $10080^3$1 for the stellar mass function, and a hydrostatic-mass bias $10080^3$2 for X-ray gas fractions. This enables model variations to be defined in terms of shifts in the calibration observables rather than in terms of the values of specific subgrid parameters. For the intermediate-resolution thermal model, the maximum-likelihood parameters are quoted as $10080^3$3, $10080^3$4, $10080^3$5, and $10080^3$6 (Kugel et al., 2023).

4. Catalogues, lightcones, and public data products

The public FLAMINGO release describes more than $10080^3$7 petabytes of data. Snapshot outputs are provided at 13 redshifts, namely $10080^3$8, together with reduced snapshots containing all particles within $10080^3$9 of selected massive haloes and downsampled snapshots containing $(1\,\mathrm{Gpc})^3$0 of particles plus all black holes (Helly et al., 27 Apr 2026).

Halo and galaxy catalogues include spherical-overdensity masses such as $(1\,\mathrm{Gpc})^3$1 and $(1\,\mathrm{Gpc})^3$2, fixed physical apertures, inclusive and exclusive measurements, and projected apertures along $(1\,\mathrm{Gpc})^3$3. The release also includes complete merger trees, cross-matching catalogues to link objects across hydro and gravity-only runs and across model variants, and initial conditions for selected runs (Helly et al., 27 Apr 2026).

Lightcone output is a central architectural feature of FLAMINGO. The original suite description states that lightcone output is produced on-the-fly for up to 8 different observers, with full-sky HEALPix maps and particle lists for lensing, SZE, X-ray, dark matter, gas, stars, black holes, and neutrinos. In the public release, particle lightcones record gas, dark matter, stars, black holes, and neutrinos crossing the past lightcone to $(1\,\mathrm{Gpc})^3$4, halo lightcones extend to $(1\,\mathrm{Gpc})^3$5, and HEALPix shell maps provide mass by species, SFR, dispersion measure, kSZ Doppler $(1\,\mathrm{Gpc})^3$6, thermal SZ $(1\,\mathrm{Gpc})^3$7, X-ray emission in multiple bands, CMB lensing convergence, cosmic infrared background, and radio AGN emission (Schaye et al., 2023, Helly et al., 27 Apr 2026).

A practical innovation of the release is an HTTP-based HDF5 streaming service for partial remote access. Clients request filesystem metadata or dataset slices through URL endpoints, data are serialized using MessagePack and streamed in chunks, and a Python client named hdfstream mimics an h5py interface with lazy loading and slicing. The service is integrated with swiftsimio for high-level access to snapshots and SOAP catalogues, and a web-browser interface allows inspection of file and dataset metadata and generation of direct download links (Helly et al., 27 Apr 2026).

5. Cluster and large-scale-structure predictions

A primary outcome of FLAMINGO is a calibrated description of baryonic suppression in large-scale-structure statistics. By comparing hydrodynamical and dark-matter-only simulations, the project introduction reports that baryonic effects can suppress the halo mass function and the matter power spectrum by up to $(1\,\mathrm{Gpc})^3$8 per cent. The halo-mass-function suppression dips by up to $(1\,\mathrm{Gpc})^3$9 per cent at $m_{\rm gas}=1.34\times10^8\,M_\odot$0, while the matter power spectrum is approximately tracer-like at $m_{\rm gas}=1.34\times10^8\,M_\odot$1, is suppressed by up to $m_{\rm gas}=1.34\times10^8\,M_\odot$2–$m_{\rm gas}=1.34\times10^8\,M_\odot$3 per cent at $m_{\rm gas}=1.34\times10^8\,M_\odot$4 due to gas ejection, and is boosted at $m_{\rm gas}=1.34\times10^8\,M_\odot$5 by stellar condensation (Schaye et al., 2023).

A more focused analysis of the gas field shows that while gas velocities do not differ from those of dark matter on large scales, the gas mass power spectrum is suppressed by up to $m_{\rm gas}=1.34\times10^8\,M_\odot$6 relative to matter even on gigaparsec scales, with $m_{\rm gas}=1.34\times10^8\,M_\odot$7–$m_{\rm gas}=1.34\times10^8\,M_\odot$8 at $m_{\rm gas}=1.34\times10^8\,M_\odot$9 and stronger suppression on $z$0. The same study interprets this large-scale effect as a “density bias” caused by star formation removing gas from the densest, most clustered regions and leaving gas preferentially in lower-bias regions. Around haloes of mass $z$1, AGN-driven outflows produce bubbles extending to $z$2–$z$3 with peak velocities $z$4–$z$5 (Ondaro-Mallea et al., 2024).

Galaxy-cluster thermodynamics are another core validation domain. FLAMINGO predicts $z$6–$z$7, $z$8–$z$9, $10\,h\,\mathrm{Mpc}^{-1}$00–$10\,h\,\mathrm{Mpc}^{-1}$01, $10\,h\,\mathrm{Mpc}^{-1}$02–$10\,h\,\mathrm{Mpc}^{-1}$03, and $10\,h\,\mathrm{Mpc}^{-1}$04–$10\,h\,\mathrm{Mpc}^{-1}$05 relations that match observed samples across more than two dex in mass, and the evolution of the scaling relations is close to self-similar to $10\,h\,\mathrm{Mpc}^{-1}$06 after the standard $10\,h\,\mathrm{Mpc}^{-1}$07 rescalings. Radial temperature, density, pressure, and entropy profiles of the fiducial run agree with de-projected XMM and Chandra measurements to within observational scatter for $10\,h\,\mathrm{Mpc}^{-1}$08, whereas core metallicities exceed observed values by $10\,h\,\mathrm{Mpc}^{-1}$09 dex and pressure profiles are slightly too flat (Braspenning et al., 2023).

The suite also resolves dynamical systematics in cluster mass calibration. For hydrostatic masses derived from X-ray–weighted profiles, the median hydrostatic mass bias increases from $10\,h\,\mathrm{Mpc}^{-1}$10 on group scales to $10\,h\,\mathrm{Mpc}^{-1}$11 on massive cluster scales, with minimum scatter $10\,h\,\mathrm{Mpc}^{-1}$12 around $10\,h\,\mathrm{Mpc}^{-1}$13. Analyses of the virial and Euler equations in the same work suggest that non-thermal motions, including rotation, account for most of the hydrostatic mass bias, and that X-ray luminosity weighted profiles strongly overestimate the deviations from hydrostatic equilibrium (Braspenning et al., 2024).

At the level of halo boundaries, FLAMINGO predicts a clear dark-matter splashback radius whose radius $10\,h\,\mathrm{Mpc}^{-1}$14 anti-correlates with the specific mass-accretion rate $10\,h\,\mathrm{Mpc}^{-1}$15, with $10\,h\,\mathrm{Mpc}^{-1}$16 falling from $10\,h\,\mathrm{Mpc}^{-1}$17 at $10\,h\,\mathrm{Mpc}^{-1}$18 to $10\,h\,\mathrm{Mpc}^{-1}$19 at $10\,h\,\mathrm{Mpc}^{-1}$20. The minimum in the stacked 3D gas-density and pressure slopes and the maximum in the entropy slope broadly align with the dark-matter splashback feature, while projected weak-lensing surface-density minima trace $10\,h\,\mathrm{Mpc}^{-1}$21 to within $10\,h\,\mathrm{Mpc}^{-1}$22 and X-ray/SZ minima occur at systematically larger radii (Towler et al., 2023).

For CMB lensing and neutrino-mass studies, FLAMINGO is used to measure a baryonic suppression function $10\,h\,\mathrm{Mpc}^{-1}$23 and to validate fast predictions of the non-linear lensing power spectrum. In that application, clustering suppression due to small-scale baryonic phenomena such as feedback from active galactic nuclei can reduce the lensing power by of order $10\,h\,\mathrm{Mpc}^{-1}$24, while the scale-dependent suppressions due to neutrinos and baryons are found to approximately factorize, $10\,h\,\mathrm{Mpc}^{-1}$25, implying that careful baryonic modelling can limit bias in neutrino-mass constraints (Upadhye et al., 2023).

6. High-redshift galaxies and SMBH populations

Although FLAMINGO was designed primarily for late-time large-scale structure and cluster surveys, the highest-resolution $10\,h\,\mathrm{Mpc}^{-1}$26 run is also used for early massive galaxies in the JWST era. In that configuration, the box contains $10\,h\,\mathrm{Mpc}^{-1}$27 independent “JWST-sized” sub-boxes of $10\,h\,\mathrm{Mpc}^{-1}$28. At $10\,h\,\mathrm{Mpc}^{-1}$29, the total variance in the number of haloes with $10\,h\,\mathrm{Mpc}^{-1}$30 or galaxies with $10\,h\,\mathrm{Mpc}^{-1}$31 is $10\,h\,\mathrm{Mpc}^{-1}$32–$10\,h\,\mathrm{Mpc}^{-1}$33 times the Poisson expectation, and for the most massive halo per $10\,h\,\mathrm{Mpc}^{-1}$34 field the $10\,h\,\mathrm{Mpc}^{-1}$35 scatter in $10\,h\,\mathrm{Mpc}^{-1}$36 exceeds the Poisson-only prediction by a factor $10\,h\,\mathrm{Mpc}^{-1}$37 at $10\,h\,\mathrm{Mpc}^{-1}$38 (Lim et al., 12 Nov 2025).

The same analysis finds a pronounced large-scale conformity. When sub-boxes are ranked by the stellar mass of their most massive galaxy, $10\,h\,\mathrm{Mpc}^{-1}$39, the stellar-to-halo mass relation and star-formation efficiency are coherently elevated or suppressed throughout the full $10\,h\,\mathrm{Mpc}^{-1}$40 volume. For $10\,h\,\mathrm{Mpc}^{-1}$41, the offset is $10\,h\,\mathrm{Mpc}^{-1}$42–$10\,h\,\mathrm{Mpc}^{-1}$43 dex, and the signal survives only to radii $10\,h\,\mathrm{Mpc}^{-1}$44 once galaxies outside the original volume are included. In this framework, $10\,h\,\mathrm{Mpc}^{-1}$45 is a better predictor of the volume-wide efficiency of massive galaxies than total number counts, and the stellar fraction of the most massive galaxies,

$10\,h\,\mathrm{Mpc}^{-1}$46

rises from $10\,h\,\mathrm{Mpc}^{-1}$47 at $10\,h\,\mathrm{Mpc}^{-1}$48 to a peak $10\,h\,\mathrm{Mpc}^{-1}$49 at $10\,h\,\mathrm{Mpc}^{-1}$50, with a log-normal dispersion $10\,h\,\mathrm{Mpc}^{-1}$51–$10\,h\,\mathrm{Mpc}^{-1}$52 dex (Lim et al., 12 Nov 2025).

A different high-redshift application uses FLAMINGO to compare against JADES spectroscopy of massive quiescent galaxies at $10\,h\,\mathrm{Mpc}^{-1}$53. In that study, spectroscopic number densities are found to be 10 times higher than predicted by galaxy formation models, and cosmic variance is ruled out at the $10\,h\,\mathrm{Mpc}^{-1}$54 level. Within FLAMINGO, the in-situ fraction of stellar mass in simulated high-$10\,h\,\mathrm{Mpc}^{-1}$55 quiescent galaxies exceeds $10\,h\,\mathrm{Mpc}^{-1}$56 per cent, supporting the conclusion that these systems generally did not undergo multiple major dry mergers. This suggests, in the language of the study, that major mergers are not a viable channel for quenching most massive galaxies (Baker et al., 2024).

The large $10\,h\,\mathrm{Mpc}^{-1}$57 FLAMINGO volume also enables direct studies of bright quasars. FLAMINGO reproduces the observed quasar luminosity function at low redshift and for faint quasars with $10\,h\,\mathrm{Mpc}^{-1}$58, but significantly underpredicts the abundance of bright quasars at $10\,h\,\mathrm{Mpc}^{-1}$59–$10\,h\,\mathrm{Mpc}^{-1}$60. A post-processing log-normal luminosity scatter of $10\,h\,\mathrm{Mpc}^{-1}$61 dex boosts the bright end by upscattering lower-luminosity systems, mostly low-mass black holes radiating above the Eddington limit. The same simulation reproduces observed quasar clustering across $10\,h\,\mathrm{Mpc}^{-1}$62, but underpredicts the clustering strength at $10\,h\,\mathrm{Mpc}^{-1}$63 (Ding et al., 28 Oct 2025).

7. Survey systematics, empirical models, and mock skies

Because FLAMINGO combines very large volume with full hydrodynamics, it is increasingly used as an end-to-end validation platform for survey systematics. One example is the validation of empirical mock-galaxy models. In a comparison against FLAMINGO galaxy samples constructed to mimic DESI-BGS and BOSS, GalaxyEmu-Planck precisely reproduces the two-point correlation function, galaxy-galaxy lensing restricted to scales greater than $10\,h\,\mathrm{Mpc}^{-1}$64, and higher-order statistics, while SHAMe performs better than a 13-parameter HOD model for higher-order statistics and galaxy assembly bias (Contreras et al., 2024).

FLAMINGO lightcones have also been used to assess claims of large-scale anisotropy from cluster scaling relations. Using 1,728 simulated lightcones constructed from the isotropic $10\,h\,\mathrm{Mpc}^{-1}$65 run, the probability of obtaining the observed joint X+SZ dipole amplitude after matching the intrinsic scatter to observations is $10\,h\,\mathrm{Mpc}^{-1}$66 $10\,h\,\mathrm{Mpc}^{-1}$67, and a bulk-flow interpretation at $10\,h\,\mathrm{Mpc}^{-1}$68 yields $10\,h\,\mathrm{Mpc}^{-1}$69 $10\,h\,\mathrm{Mpc}^{-1}$70. In that analysis, statistical noise accounts for over $10\,h\,\mathrm{Mpc}^{-1}$71 of the anisotropy amplitude in each lightcone, with large peculiar velocities contributing less than $10\,h\,\mathrm{Mpc}^{-1}$72 (He et al., 2 Apr 2025).

A further survey-facing application is intrinsic alignment modelling. For more than 4.9 million LRG-like galaxies at $10\,h\,\mathrm{Mpc}^{-1}$73, two-point position-position and position-shape correlations from FLAMINGO are well fit by both the Non-Linear Alignment and Tidal Alignment Tidal Torquing models, with a mass-dependent TATT model, TATT-M, very strongly preferred over NLA in Bayesian model comparison. The same analysis finds that variations in AGN and supernova feedback do not significantly change the alignment amplitude beyond the change associated with the dependence of galaxy stellar mass on the strength of feedback. A three-point follow-up based on the $10\,h\,\mathrm{Mpc}^{-1}$74 volume detects third-order intrinsic-alignment statistics and finds that tree-level EFT and reduced EFT models yield unbiased alignment amplitudes, whereas NLA and EFT without the velocity-shear term are biased (Herle et al., 22 Jan 2026, Vedder et al., 25 Jan 2026).

FLAMINGO now also supplies self-consistent mock maps of secondary CMB anisotropies and extragalactic foregrounds constructed from lightcone-based HEALPix maps and catalogues. These include CMB lensing, thermal and kinetic Sunyaev-Zeldovich effects, cosmic infrared background, radio point sources, and anisotropic screening maps. The mock skies reproduce a wide range of observational constraints and preserve the physical cross-correlations between different components that are absent from many dark-matter-only painting schemes. This suggests that FLAMINGO is not only a calibrated simulation suite but also a survey-inference infrastructure for joint analyses of cosmology, baryonic feedback, and multi-wavelength foregrounds (Yang et al., 10 Dec 2025).