A direct black-hole mass measurement in a little red dot at high redshift Authors: Ignas Juodžbalis, Cosimo Marconcini, Francesco D’Eugenio, Roberto Maiolino, Alessandro Marconi, Hannah Übler, Jan Scholtz, Xihan Ji, Gareth C Jones, Michele Perna, Santiago Arribas, Jake S Bennett, Volker Bromm, Andrew J Bunker, Stefano Carniani, Stéphane Charlot, Giovanni Cresci, Pratika Dayal, Eiichi Egami, Andrew Fabian, Kohei Inayoshi, Yuki Isobe, Lucy R Ivey, Sophie Koudmani, Nicolas Laporte, Boyuan Liu, Jianwei Lyu, Giovanni Mazzolari, Stephanie Monty, Eleonora Parlanti, Pablo G Pérez-González, Brant Robertson, Raffaella Schneider, Debora Sijacki, Sandro Tacchella, Alessandro Trinca, Rosa Valiante, Marta Volonteri, Joris Witstok, Saiyang Zhang Published: 2026 ABSTRACT AbstractRecent discoveries of faint active galactic nuclei (AGN) at the redshift frontier have revealed a plethora of broad Hα emitters with optically red continua, named little red dots (LRDs)1, which comprise 15–30% of the high-redshift broad-line AGN population2. Owing to their peculiar properties3–6, modelling LRDs with standard AGN scenarios has proven challenging. In particular, the validity of single-epoch virial mass estimates in determining the black-hole masses of LRDs has been called into question, with some models claiming that masses might be overestimated by up to two orders of magnitude7–10. Here we report a direct, dynamical black-hole mass measurement in a strongly lensed LRD at a redshift of 7.04. The combination of lensing with deep spectroscopic data reveals a rotation curve that is inconsistent with a nuclear star cluster, yet can be well explained by Keplerian rotation around a point mass of 50 million solar masses, consistent with virial black-hole mass estimates. The Keplerian rotation leaves little room for any stellar component in a host galaxy, as we conservatively infer MBH/M⁎ > 2 (where MBH is the black-hole mass and M⁎ is the stellar mass). Such a ‘naked’ black hole, together with its near-pristine environment11, indicates that this LRD is a massive black-hole seed caught in its earliest accretion phase.Subject terms: Compact astrophysical objects, Galaxies and clusters Main Abell 2744−QSO1 (hereafter QSO1) is a strongly lensed, triply imaged system, first discovered in the James Webb Space Telescope (JWST) Near Infrared Camera imaging by ref. 12, whose redshift was confirmed to be z = 7.04 through Near Infrared Spectrograph (NIRSpec) prism spectroscopy13, which also revealed broad components in Hα and Hβ lines. The compactness and a red optical (rest frame) slope together with a blue ultraviolet (rest frame) slope classify it as a typical ‘little red dot’ (LRD)2,14. Further observations clearly spectrally resolved the broad- and narrow-line emission, and also detected line variability4,15,16, thereby robustly identifying QSO1 as hosting an accreting black hole (BH). On the basis of virial relations using broad-line widths and luminosities, a BH mass of about 4 × 107 M⊙ was estimated4,13,15. However, these results rest on the assumption that ‘virial relations’17 that are calibrated locally, are still applicable at z = 7. In this work, we provide a direct BH mass measurement in the high-redshift Universe, indeed illustrating that virial BH mass calibrations apply to this prototypical LRD. It is first interesting to note that, given the low narrow-line velocity dispersion in QSO115 (σN < 22 km s−1; Supplementary Information section 1.1), the BH’s sphere of influence, assuming the mass estimated from the virial relations, has a radius of ≳270 pc, which should be resolvable with JWST, owing to the gravitational lens shearing of about 3.5 (ref. 13). However, we will not make any a priori assumptions about the BH mass in the following analysis. We perform a detailed analysis of the spatial and kinematic information present in the current deep, high-resolution NIRSpec integral field spectroscopy data, by focusing on the narrow Hα emission. The line flux, line-of-sight velocity (first moment) and velocity dispersion (second moment) maps are shown in Fig. 1 (top). Clearly, the narrow Hα velocity field reveals a velocity gradient with an amplitude of approximately 10 km s−1. Hα narrow-line emission and moment maps of QSO1. The top row shows the observed integrated narrow Hα flux, and the first and second moments of the flux distribution, which correspond to line-of-sight velocity and velocity dispersion, respectively. The middle row shows the best-fit Keplerian MOKA3D model to the Hα maps. The bottom row shows the residuals of the best-fit model. The dashed green ellipse in the top-left panel shows the point spread function and the inner red solid ellipse shows 3σ uncertainty of the centroids of the velocity channels used for spectroastrometry. The green segment shows the orientation of the integral-field-unit slices, approximatively parallel to the rotation field, indicating that the latter cannot be a calibration artefact. The spatial scale is given by the black line in the top-left panel. The narrow-line emission is spatially extended on scales of up to approximately 200 pc, as found by ref. 11 (whereas the continuum is unresolved13). We sample the rotation curve across 2 spatial bins corresponding to 100-pc and 150-pc distances from the rotation centre (Methods). We do not bin velocities at distances closer to the centre, as these are affected by beam smearing with the velocities on the opposite side, and require other techniques to be resolved (spectroastrometry and full three-dimensional (3D) analysis, discussed below). The binned velocities are shown as blue circles in Fig. 2. One-dimensional rotation curve of QSO1. Blue points are from the binned two-dimensional line-of-sight velocity field and the magenta crosses are the spectroastrometry measurements with lighter violet markers showing spectroastrometry of a finer split line profile, discussed in Methods. Line-of-sight velocity is given by v, while r represents the distance from the dynamical centre. The solid black line indicates the Keplerian best fit with a point mass, giving a BH mass of 5.6 × 106 M⊙ (which is a lower limit given that with this method the inclination is not constrained). The dot-dashed line is for an NSC (see text), which results in a worse fit. The small size of the object makes it difficult to trace the inner parts of the rotation curve. However, given the high signal-to-noise on Hα, it is possible to recover kinematic information below the spatial resolution beam through spectroastrometric techniques. In brief, spectroastrometry consists in identifying the centroid positions when scanning spectroscopic channels of a line18, producing a map of average gas positions across a velocity range (Extended Data Fig. 1 and Supplementary Information section 1.2). We find that the centroids of the Hα emission in the +50 km s−1 and −50 km s−1 velocity channels are separated by 24.9 ± 9.4 pc in the source plane (magenta crosses in Fig. 2). Taking half of this separation gives a radius scale rspec=12.5-4.7+4.7pc, which, coupled with the velocity, yields a spectroastrometric enclosed mass in solar masses (Mspec/M⊙) of log(Mspec/M⊙)=6.70-7.2 (Supplementary Information section 1.2). This is a lower limit, as the inclination is unconstrained with this method. We have also leveraged spectroastrometry by splitting the line in finer velocity bins (lighter violet markers in Fig. 2; Supplementary Information section 1.2)—in this case, the signal-to-noise is lower, hence these points are less constraining, but they clearly show evidence of a rapidly declining rotation curve. Spectroastrometry on the narrow line. Panel a - locations of the centroids of each velocity channel with the green line showing the PA of the IFU slicer. Panel b narrow Hα line profile extracted from a 0.1″ aperture (coinciding with the scales traced by our kinematics fits) with colored bars showing the velocity bins. While the line profile appears to show some evidence of a wing at ~ −150 km s−1, expanding the velocity channels to include it does not meaningfully alter the results. The spectroastrometric measurements above indicate a compact and dense system. However, they alone cannot exclude a significant contribution of stars, gas or dark matter to the mass budget. We therefore combine the spectroastrometric measurements with the large-scale rotation to construct a rotation curve for QSO1 (Fig. 2). We fit the data in Fig. 2 with (1) a point mass and (2) a compact, yet extended, mass distribution mimicking the (thoroughly studied) nuclear star cluster (NSC) of the Milky Way19. Details are given in Methods; however, Fig. 2 shows that an extended Milky-Way-like NSC mass distribution is disfavoured (reduced chi-squared χR2=3.2) when compared with a pure point mass (Keplerian) model (χR2=1.0), with BH mass, log(MBH/M⊙) = 6.75 ± 0.15. The evidence from kinematics alone corresponds to >5σ preference for a point mass. Additional kinematic evidence ruling out the Milky Way NSC (and more extensive NSC cases) are presented further below and in Methods, also through the full 3D kinematic fitting. Here we note that the implied NSC masses, logM⁎( 5 and in an LRD. An immediate implication is that alternative scenarios explaining this LRD without accreting BHs are essentially ruled out. In addition, we can investigate the reliability of single-epoch BH mass virial estimates out to the epoch of reionization, and specifically for LRDs. Figure 3 shows a comparison between our BH mass direct measurement and literature estimates obtained using the virial scaling relations, showing full consistency. However, the scenario in which the broad lines are produced by electron scattering7, while providing a good spectral fit to the broad lines (Supplementary Fig. 8), underestimates the BH mass by nearly 2 dex (Fig. 3). Summary of BH mass estimates for QSO1. Comparison between our lower limit and MOKA3D direct measurements (green stars) with previous virial estimates (black circles)4,13,15, the scenarios assuming that the dominant broadening mechanism of the broad lines is due to electron scattering7 and the Lbol based estimates from ref. 36. The purple shaded region shows the 1σ uncertainty on the MOKA3D estimate. The resulting Eddington luminosity (LEdd) of the BH is 7.6 × 1045 erg s−1. By using standard scaling relations between broad Hα and bolometric luminosity (L)23, we infer that the BH is accreting well below its Eddington limit, with L/LEdd ≈ 0.02. If the broad Hα to bolometric luminosity relation is higher than estimated locally (for example, ref. 6), then this would imply L/LEdd ≈ 0.01 or lower, indicating that the BH might be in a near-dormant state. For this reason, the BH mass estimates using bolometric luminosity (Lbol) with assumed L/LEdd = 1, fail to match our dynamical measurement (Fig. 3). Yet, the BHs may still have experienced previous super-Eddington bursts, as inferred by ref. 24 for another overmassive, dormant BH at a similar redshift. Finally, we note that the Keplerian rotation curve leaves little room for any stellar component. Specifically, in Supplementary Information section 1.7, we conservatively derive a dynamical upper limit on the stellar mass in the host galaxy of M⁎ < 2 × 107 M⊙. To our knowledge, this upper limit makes QSO1 the most ‘naked’ massive BH ever found, with MBH/M⁎ > 2, and in line with the previous finding that this BH is hosted in an environment that is chemically nearly pristine11. This demonstrates the possibility of BH primacy, that is, BHs forming and growing earlier than their host galaxy. The lower limit on the MBH/M⁎ ratio is three orders of magnitude higher than observed locally. Figure 4 shows how extreme QSO1 is on the MBH versus M⁎ diagram relative the local relation—located at the extreme tail of the overmassive BHs identified by JWST in previous surveys25. Location of QSO1 on the MBH–M⁎ plane. QSO1 is indicated by the red star. The remaining points represent measurements from other JWST observations of low-mass active galactic nuclei (AGN)25,37–42 and quasars43–45. The solid green line shows the local scaling relation from ref. 17, with the scatter indicated by shading. The other green lines indicate constant MBH/M⁎ ratios. With MBH/M⁎ > 2, QSO1 lies orders of magnitude above the local scaling relations and is approximately 1 dex more overmassive than even the most extreme sources found by JWST so far. The only scenarios that can account for such a system are those invoking ‘heavy seeds’, such as direct-collapse BHs (resulting from the direct collapse of massive pristine clouds) or primordial BHs (formed in the first second after the Big Bang)26–28. Yet, most direct-collapse BH scenarios would require a strong source of ultraviolet radiation in the vicinity, which is not seen (not even a post-starburst galaxy that might have produced ultraviolet radiation in the past), although some scenarios may expect direct collapse in different environments29. However, direct-collapse BH models suggest that their early growth is limited by the baryon fraction in an atomically cooling halo30,31, placing an upper limit on the MBH/Mdyn ratio of approximately 0.1, that is, more than 1 dex lower than our inferred lower limit. However, some independent evidence for the primordial BH scenario comes from the very low metallicity of this system11. However, the observed mass of 5 × 107 M⊙ is significantly higher than the preferred 106 M⊙ primordial BH mass scale given by the electron–positron annihilation epoch in the ultra-early universe32 (but see also ref. 33); therefore, the observed mass would require either significant accretion or rapid merging of primordial BHs, which may be linked to their highly clustered nature34. Regardless of the specific model, the high mass in such a remote cosmic epoch, the extremely high MBH/M⁎, together with the near-pristine environment11, indicate that QSO1 is a massive BH seed caught in the earliest phases of accretion35. Methods Conventions Throughout this work we assume a flat Λ (dark energy) cold-dark-matter cosmology with matter density parameter Ωm = 0.315 and a Hubble constant H0 = 67.4 km s−1 Mpc−1 (ref. 46). All reported magnitudes are in the AB system. Following the lensing model of ref. 13, we adopt a flux magnification factor μ = 6.2 and a shear factor of 3.52 for our source (image A of QSO1). Hence, 1 arcsec in the image plane corresponds to 1.52 physical kiloparsecs. For robustness tests, we use the Bayesian information criterion (BIC), defined as BIC ≡ χ2 + klnn, where k is the total number of model parameters and n is the number of points fitted; a decrease in BIC, ΔBIC ≥ 5, between two models was required for robust preference of one over the other, although our main conclusions remain unchanged even if a stricter ΔBIC ≥ 10 threshold is adopted. Data reduction We use data from the BlackTHUNDER NIRSpec integral field unit survey, focusing on the 7.3-hour exposures with the G395H grating, giving a nominal spectral resolution R = 3,700 at the wavelength of Hα15. The NIRSpec integral field unit was centred on image A of QSO1 (right ascension 00:14:19.161; declination −30:24:05.664)12. A detailed description of the reduction procedures is available in refs. 4,15; however, a summary is provided here for context. The spectra were extracted following the procedures of ref. 47, but using version 1.17.1 of the JWST pipeline. At z = 7.04, the Hα line falls just outside the nominal wavelength coverage; however, the F290LP filter does not cut off longer wavelengths and the detector efficiency allows to recover Hα emission. We perform this recovery by extrapolating the wavelength solution, flat-field curves and the grating-equation-derived line spread function (LSF) out of the nominal range and towards the detector sensitivity limit of λ = 5.34 μm. The peak of the Hα line of QSO1 falls on λ = 5.278 μm; hence, our modification readily recovers the entirety of Hα emission. Although flux calibrations beyond the nominal range may suffer inaccuracies, the primary interest of this work is a kinematics study; hence, our key kinematics results are insensitive to flux calibrations. The BH mass measurements are more affected. However, the square-root dependence of the BH mass on luminosity means that flux calibrations have to be wrong by an order of magnitude to significantly impact the measurements. The nominal spaxel scale of the processed data was 0.05″; however, utilizing the large number of dithers, we are able to oversample the cube to a scale of 0.02″ per spaxel without incurring significant sampling artefacts. We choose the 0.02″ cube for the main kinematic and spectroastrometric analysis, with the 0.05″ cube used to perform consistency checks, ensuring that our results are not pixel sampling artefacts. Spectroastrometry of the rotation curve To constrain the density profile of QSO1, we combine spectroastrometric measurements with resolved kinematics. The technical details of spaxe-by-spaxel fitting and spectroastrometry are given in Supplementary Information sections 1.1 and 1.2; here we summarize that we subtract the broad Hα emission from the cube and create images of different velocity channels of narrow Hα for which centroids can be obtained at sub-point-spread-function scales (provided a sufficient signal to noise48) and used to map dynamics below the nominal instrumental resolution49,50. The fiducial spectroastrometric analysis utilizing two velocity channels for higher signal to noise is shown in Extended Data Fig. 1. However, as shown in Fig. 2, splitting the line into finer bins does not change our results. We infer the outer parts of the rotation curve by binning the line-of-sight velocity field (shown in Fig. 1) on scales >60 pc to avoid beam smearing. This procedure resulted in 4 bins covering the negative and positive sides of the rotation curve with ⟨vsini⟩ ≈ 10 km s−1 (where v is the line-of-sight velocity and i is the inclination angle) with nominal uncertainties of order 1 km s−1. However, these uncertainties, estimated through the standard root mean square (rms) weighting scheme, do not take into account the velocity field cross-correlation between spaxels of each bin owing to beam smearing and hence are probably significantly underestimated. An a priori derivation of the covariance matrix is intractable as it would require fitting individual dithers, which have far too low signal to noise. We thus use an empirical approach—scaling the naive rms-derived errors until the optimal model in the family of models fitted has χR2=1. This yields an upper limit on the possible errors as it assumes that the optimal model is the ground truth. Consequently, using this method, we can establish a lower limit on the significance of the optimal model (which turns out to be a point mass) over other models considered. For the spectroastrometric data points, we use a flux weighted average of the velocity channels, giving ⟨vsini⟩ = 51 ± 4 km s−1. The factor of sini is written to explicitly state that these are the projected values, uncorrected for inclination. As the resultant rotation curve, shown in Fig. 2, is sparsely sampled, we consider only simple one- or two-parameter models for fitting. Model curves were constructed following a Keplerian prescription: where G is the gravitational constant, R is the distance from the centre and M( 100 pc) kinematics, would predict a smoothly declining velocity towards the centre, in stark contrast to the steep inner velocity slope of the one-dimensional curve, as illustrated in Extended Data Fig. 6. Likewise, MOKA3D gives substantial residuals for such an extended density profile (Extended Data Fig. 7). Spectroastrometric rotation curve with an exponential disc model. Same as Fig. 2 in the main text, except the dotted line illustrates the velocity field expected from an exponential disc with ~107 M⊙ and Rs = 150 pc. As shown here, such a model provides a considerably worse fit than the others considered. MOKA3D residuals for an exponential disc. The disk was taken to have ~107 M⊙ and Rs = 150 pc. It can be seen that such a model does not reproduce the data as well as the fiducial point mass model. We finally emphasize that the MOKA3D analysis is fully independent of the spectroastrometry approach. Although both analyses use the same data cube as input, they are otherwise entirely separate approaches with spectroastrometry hinging on the simple approach of centroiding different velocity channels, whereas MOKA3D self-consistently models the entire data cube. Online content Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41586-026-10579-4. Supplementary information This file contains Supplementary Methods—additional technical verification carried out to ensure the robustness of the results and showcases fits to individual spaxels and null tests carried out on observations of a calibration star and the continuum of the source. This file also contains additional investigation into the presence of outflows in the object and technical details of the stellar-mass estimates along with electron-scattering fits to the broad line, which are secondary to the main analysis of the paper. Peer Review File Acknowledgements We thank P. Natarajan for comments. I.J. acknowledges support by the Huo Family Foundation through a P.C. Ho PhD Studentship. This work is based on observations made with the National Aeronautics and Space Administration (NASA)/European Space Agency (ESA)/Canadian Space Agency (CSA) JWST. The data were obtained from the Mikulski Archive for Space Telescopes at the STScI, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS 5-03127 for JWST. These observations are associated with programme PID 5015. R.M., F.D’E., J.S., I.J. and G.C.J. acknowledge support from the Science and Technology Facilities Council (STFC), by the European Research Council (ERC) through Advanced Grant 695671 ‘QUENCH’, by the UK Research and Innovation (UKRI) Frontier Research grant RISEandFALL. R.M. also acknowledges support from a Royal Society Research Professorship grant. S.A. acknowledges grant PID2021-127718NB-I00 funded by the Spanish Ministry of Science and Innovation/State Agency of Research (MICIN/AEI/ 10.13039/501100011033). H.Ü. acknowledges funding by the European Union (ERC APEX, 101164796). Views and opinions expressed are however those of the authors only and do not necessarily reflect those of the European Union or the European Research Council Executive Agency. Neither the European Union nor the granting authority can be held responsible for them. A.J.B. acknowledges funding from the ‘FirstGalaxies’ Advanced Grant from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement number 789056). S. Carniani acknowledges support by European Union’s HE ERC Starting Grant No. 101040227 - WINGS. S.K. has been supported by a Junior Research Fellowship from St Catharine’s College, Cambridge and a Research Fellowship from the Royal Commission for the Exhibition of 1851. R.S. acknowledges support from the PRIN2022 MUR project 2022CB3PJ3 - First Light And Galaxy aSsembly (FLAGS) funded by the European Union - Next Generation EU and from the EU-Recovery Fund PNRR - National Centre for HPC, Big Data and Quantum Computing. V.B., B.L. and S.Z. acknowledge the Texas Advanced Computing Center (TACC) for providing HPC resources under allocation AST23026. D.S. acknowledges support from the STFC, grant code ST/W000997/1. A.T. acknowledges support from the PRIN MUR project 2022935STW, funded by European Union - Next Generation EU, and from the INAF Fundamental Research 2023 Mini-grant project ’Cosmic Archaeology with the first black hole seeds’. G.C. acknowledges support from the INAF GO grant 2024 ‘A JWST/MIRI MIRACLE: MidIR Activity of Circumnuclear Line Emission’. K.I. acknowledges support from the National Natural Science Foundation of China (12573015, 1251101148, 12233001), the Beijing Natural Science Foundation (IS25003), and the China Manned Space Program (CMS-CSST-2025-A09). P.G.P.-G. acknowledges support from grant PID2022-139567NB-I00 funded by Spanish Ministerio de Ciencia e Innovación MCIN/AEI/10.13039/501100011033, FEDER Una manera de hacer Europa. M.P. acknowledges support through the grants PID2021-127718NB-I00, PID2024-159902NA-I00 and RYC2023-044853-I, funded by the Spain Ministry of Science and Innovation/State Agency of Research MCIN/AEI/10.13039/501100011033 and El Fondo Social Europeo Plus FSE+. R.V. acknowledges support from PRIN MUR ‘2022935STW’ funded by European Union - Next Generation EU, Missione 4 Componente 2 CUP C53D23000950006 and from the Bando Ricerca Fondamentale INAF 2023, Theory Grant ‘Theoretical models for Black Holes Archaeology. J.W. acknowledges support from the Cosmic Dawn Center through the DAWN Fellowship. The Cosmic Dawn Center (DAWN) is funded by the Danish National Research Foundation under grant number 140. I.J. discloses support for the research of this work from the Huo Family Foundation. The other authors declare no relevant funding. Extended data figures and tables Author contributions I.J. led the data analysis and writing of the paper with C.M. performing independent verification of the results. Key contributions to the data analysis were provided by F.D’E., R.M., A.M., H.Ü, J.S. and X.J. Observations were prepared and data reduced by F.D’E., H.Ü. and M.P. Theoretical interpretation of the data and discussion of the seeding models was contributed by J.S.B., V.B., P.D., A.F., K.I., S.K., B.L., J.L., R.S., D.S., A.T., R.V., M.V. and S.Z. G.C.J. provided the standard star data for the null test. S.M. verified the star cluster mass profile models. S.A., A.J.B., S. Carniani, S. Charlot, G.C., E.E., Y.I., L.R.I., N.L., G.M., E.P., P.G.P.-G., B.R., S.T. and J.W. provided comments on the paper and assisted in checking the validity of the results. Peer review Peer review information Nature thanks the anonymous reviewers for their contribution to the peer review of this work. Peer reviewer reports are available. Data availability The data used in this study were obtained as part of JWST programme ID 5015, and are available from the Mikulski Archive for Space Telescopes (https://mast.stsci.edu/portal/Mashup/Clients/Mast/Portal.html) at the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS 5-03127 for JWST. The exact reduced data products used are available on Zenodo (https://zenodo.org/records/19402518). Code availability The MOKA3D code underpinning this study is available at https://github.com/cosimomarconcini/moka_3d. A simplified demo version capable of reproducing the main results is made available with the paper. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Extended data is available for this paper at 10.1038/s41586-026-10579-4. Supplementary information The online version contains supplementary material available at 10.1038/s41586-026-10579-4. References Associated Data Supplementary Materials This file contains Supplementary Methods—additional technical verification carried out to ensure the robustness of the results and showcases fits to individual spaxels and null tests carried out on observations of a calibration star and the continuum of the source. This file also contains additional investigation into the presence of outflows in the object and technical details of the stellar-mass estimates along with electron-scattering fits to the broad line, which are secondary to the main analysis of the paper. Peer Review File Data Availability Statement The data used in this study were obtained as part of JWST programme ID 5015, and are available from the Mikulski Archive for Space Telescopes (https://mast.stsci.edu/portal/Mashup/Clients/Mast/Portal.html) at the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS 5-03127 for JWST. The exact reduced data products used are available on Zenodo (https://zenodo.org/records/19402518). The MOKA3D code underpinning this study is available at https://github.com/cosimomarconcini/moka_3d. A simplified demo version capable of reproducing the main results is made available with the paper.