Grid nesting is the standard route to convection-resolving resolution in tropical-cyclone (TC) modelling, but in the widely used idealized workflow of the Weather Research and Forecasting model (WRF-ARW) the fine domain is <em>initialized</em> by interpolating the coarse parent state onto the fine mesh rather than by sampling the prescribed analytic vortex at the native fine resolution. Because the coarse parent under-resolves the inner core, this interpolated initial vortex carries a broadened radius of maximum wind, weakened radial gradients, and grid-scale truncation imprints, and the first hours of the run are contaminated by a geostrophic-adjustment shock—an inertia–gravity-wave burst and grid-imprinted asymmetries—whose downstream effect on structure and intensity is difficult to control. We present and document a modification to the WRF-ARW (v4.5) idealized nesting initialization that decouples the inner domain from coarse-to-fine interpolation: the source code is altered so that the child domain lays down the prescribed vortex and base-state environment directly at its own 2 km resolution, sharing the parent’s base state and so remaining physically self-consistent. We compare this refined initialization (RE) against the standard official two-domain (6→2 km; OR) and three-domain (18→6→2 km; OR<sub>3</sub>) configurations, and against a uniform 2 km single-domain run (UNI) whose inner-core initialization is identical to RE and which therefore serves as a resolution-matched reference solution. With a diagnostic suite—perturbation-vorticity texture and its azimuthal-wavenumber spectrum, time–radius diagrams of diabatic heating, radius–height tangential wind and secondary circulation, low- and upper-level radial profiles, and mean absolute deviation from UNI—we show that OR injects grid-scale pixelated noise and OR<sub>3</sub> a systematic wavenumber-4 truncation artefact, both of which trigger a spurious early adjustment (in OR<sub>3</sub>, a transient radius-of-maximumwind excursion to ∼120 km near <em>t</em> ≈ 8–12 h). The refined initialization removes these artefacts, produces a smooth quasi-balanced spin-up with the earliest and deepest boundary-layer inflow and the most compact, upright eyewall, and tracks UNI most closely in every intensity and structure metric (mean absolute deviation of minimum sea-level pressure 0.72 hPa versus 0.96–1.08 hPa; near-surface tangential wind 0.35 m s<sup>−1</sup> versus ∼1.7m s<sup>−1</sup>). Crucially, the compactness of the initial core paces the eyewall life cycle: the boundary-layer inflow, the radius–height eyewall structure, the secondary circulation, and the radius of maximum wind all show RE organizing earliest and the interpolated runs lagging by a consistent offset, so a comparison made at a single lead time can mistake this timing lag for a structural disagreement. The residual RE–UNI difference over the first 36 h arises not from initialization but from the nesting mechanism itself, and surfaces first—as expected—in the phase of the most sensitive convective (reflectivity) field. The modification is inexpensive, reproducible, and intended as a reusable improvement and test case for idealized nested TC modelling.