The first year of JWST has revealed a surprisingly large number of luminous galaxy candidates beyond z  > 10. While some galaxies have already been spectroscopically confirmed, there is mounting evidence that a subsample of the candidates with particularly red inferred UV colours are, in fact, lower redshift contaminants. These interlopers are often found to be ‘HST-dark’ or ‘optically faint’ galaxies at z  ∼ 2 − 6, a population that is key to improving our understanding of dust-obscured star formation throughout cosmic time. This paper demonstrates the complementarity of ground-based mm-interferometry and JWST infrared imaging to unveil the true nature of red 1.5–2.0 μm dropouts that have been selected as ultra-high-redshift galaxy candidates. We present NOEMA Polyfix follow-up observations of four JWST red 1.5–2.0 μm dropouts selected by Yan et al. (ApJ, 942, L8) as ultra-high-redshift candidates in the PEARLS-IDF field. The new NOEMA observations constrain the rest-frame far-infrared continuum emission and efficiently discriminate between intermediate- and high-redshift solutions. We report > 10 σ NOEMA continuum detections of all our target galaxies at observed frequencies of ν  = 236 and 252 GHz, with FIR slopes indicating a redshift of z  < 5. We modelled their optical-to-FIR spectral energy distribution (SED) with multiple SED codes, finding that they are not z  > 10 galaxies but dust-obscured, massive star-forming galaxies at z  ∼ 2 − 4 instead. The contribution to the cosmic star formation rate density (CSFRD) of such sources is not negligible at z  ≃ 3.5 ( ϕ ≳ (1.9 − 4.4) × 10 ⁻³ cMpc ⁻³ ; or > 3 − 6% of the total CSFRD), in line with previous studies of optically faint and sub-millimeter galaxies. This work showcases a new way to select intermediate- to high-redshift dust-obscured galaxies in JWST fields with minimal wavelength coverage. This approach opens up a new window onto obscured star formation at intermediate redshifts, whilst removing contaminants with red colours from searches at ultra-high redshifts.