A useful way to separate model error from measurement error is to compare the same reference plane and loss budget in both worlds. In simulation, include the actual matching network, tuner R_on, connector/contact resistance, display stack, battery, metal-frame gaps, grounding springs, and any plastic or adhesive layers with measured dielectric data rather than generic materials. In measurement, de-embed the fixture and cable up to the antenna feed, and verify that the chamber calibration, phantom position, and cable choke arrangement are fixed. If the simulated and measured S11 resonance shifts but the shape remains similar, the main issue is usually geometry/material or grounding mismatch; if S11 matches but efficiency is still lower, look for underestimated conductor, dielectric, tuner, or feed loss.
For metal phones, I would also treat total radiated efficiency as the key metric, not just realized gain or return loss. A hand or head phantom can detune the antenna and increase absorption at the same time, so a tuner may restore match while still leaving substantial mismatch-plus-loss behind the matching network. Run a controlled sequence: free space, head-only, hand-only, and head-plus-hand, keeping the same tuner state and then separately allowing retuning. That tells you whether the gap comes from environmental detuning, phantom loss, chassis construction, or the chamber/cable setup. A remaining 1–3 dB free-space discrepancy is plausible in complex handsets, but a larger repeatable gap usually points to a specific missing loss mechanism or an assembly difference. AI-generated technical discussion; verify against authoritative documentation.