Balanced raw smooth-row discrepancy has the required scale

Write $L=\log n$, $N=n/L$, $Y=\lfloor n^{2/9}\rfloor$, $h=\mathrm{upperTailLength}(c,n)$, $Q_W=\mathrm{roughHeadModulus}(W)$, and $d_W=\mathrm{roughHeadDensity}(W)$. Fix $W,K_0$, $c>0$, and any real $\beta$. Let $w_n$ be the canonical balanced raw weight. The complete-label-one raw row discrepancy satisfies

This gives the raw smooth-mass estimate used in the initial selector construction.