We report 1.7GHz very long baseline interferometry (VLBI) observations of IRAS F10214+4724, a lensed z = 2.3 obscured quasar with prodigious star formation. We detect what we argue to be the obscured active nucleus with an effective angular resolution of <50pc at z = 2.3. The S1.7 =210µJy (9σ) detection of this unresolved source is located within the Hubble Space Telescope rest-frame ultraviolet/optical arc, however, 100 mas northwards of the arc centre of curvature. This leads to a source-plane inversion that places the European VLBI Network detection to within milliarcseconds of the modelled cusp caustic, resulting in a very large magniﬁcation (μ ∼70), over an order of magnitude larger than the CO (1→0) derived magniﬁcation of a spatially resolved Jansky Very Large Array (JVLA) map, using the same lens model. We estimate the quasar bolometric luminosity from a number of independent techniques and with our X-ray modelling ﬁnd evidence that the AGN may be close to Compton thick, with an intrinsic bolometric luminosity of log10( Lbol, QSO /L ) = 11.34 ± 0.27dex. We make the ﬁrst black hole mass estimate of IRAS F10214+4724 and ﬁnd log10(MBH/M ) = 8.36 ± 0.56 which suggests a low black hole accretion rate (λ = ˙M/ ˙ MEdd ∼3±7 2 percent). We ﬁnd evidence for an MBH/Mspheroid ratio that is one to two orders of magnitude larger than that of submillimetre galaxies (SMGs) at z ∼ 2. At face value, this suggests that IRAS F10214+4724 has undergone a different evolutionary path compared to SMGs at the same epoch. A primary result of this work is the demonstration that emission regions of different sizes and positions can undergo signiﬁcantly different magniﬁcation boosts (>1dex) and therefore distort our view of high-redshift, gravitationally lensed galaxies.