Power and planned sample size

One may use $g=g_{\mathrm{robust}}$ only when that allowance is uniform for the actual attempted-run design. A positive ideal gap without calibrated actual-law control does not establish the power hypothesis.