The positive normalized Fejér kernelhas a 1/(nt2) tail. Split the error integral at δ=n−α and use the chaining inequality ω(f,t)≤(1+t/δ)ω(f,δ). Its tail contributes at most a constant times nα−1(1+αlogn)ω(f,δ). The factor is uniformly bounded for each fixed α<1, proving the displayed estimate. Its constant need not remain bounded as α tends to one.