For an active call with 1<=u<=a and M>=u, set h=ceil(3M/u). For each nonzero t in S, put i=v_2(t), q_t=h max(0,u-i), and D_t=L/2^i. Only q_t>0 points are retained. A jet here is the first q_t ordinary coefficients of F(t+alpha z), not unscaled derivatives. Odd grid points alone supply Lhu>=3ML coefficients, exceeding degree M(2L-1). Also h<=4M/u, so q_t<=4M. The multiplicity decreases by at most h per valuation level; this taper is crucial for the later precision bound.