Port extension is often the first fix tried after a coaxial calibration, but it only works when the added structure behaves like a uniform, single-mode transmission line with a stable reference-plane shift. A reproducible debug should start by connecting a short or open at the intended DUT plane and checking whether the reflected phase is linear over frequency after the extension. If the phase slope changes across band, or S11 of a matched termination still shows frequency-dependent ripple, a simple electrical delay is not sufficient.

First, keep the calibration plane fixed and characterize the cable or fixture in three separate states: bare calibrated port, port with adapter only, and adapter plus fixture. Save S2P traces for each step rather than overwriting them. Apply zero extension, then measure a short, open, and known load at the fixture end. The short/open expose phase delay and dispersion; the load exposes residual directivity, fixture mismatch, and radiation. If the load return loss changes strongly when the cable is repositioned, the problem is mechanical stability before it is a de-embedding problem.

Next, decide whether the fixture can be represented by delay alone. A homogeneous coax-like section may be corrected with port extension, including modest loss if the VNA supports loss compensation, but a PCB launch, via transition, probe pad, or millimeter-wave interconnect generally requires a fixture model or S-parameter de-embedding. The boundary condition is simple: if the two fixture halves are reciprocal, reasonably symmetric, and support only the intended propagating mode, 2x de-embedding from thru-reflect-line or equivalent fixture S-parameters is defensible. If common-mode conversion, cavity coupling, or higher-order modes appear before the upper measurement frequency, no scalar port extension can remove those effects because they are not just a phase rotation.

A practical pass/fail sequence is to calibrate at the coax plane, measure the verification kit or a precision load at that plane, then move the same load to the fixture end and compare before and after correction. The corrected result should improve both return-loss ripple and insertion-loss flatness without creating nonphysical passivity or causality violations. If de-embedding makes the trace smoother but pushes fixture S-parameters near singular values, inspect the thru definition and fixture symmetry instead of increasing the extension number. Document reference planes, torque, cable position, sweep points, IF bandwidth, and temperature so another operator can reproduce the boundary between a valid delay correction and a case requiring full fixture removal. AI-generated technical discussion; verify against authoritative documentation.