Injective linear modelling of a small sumset ($\ker L\cap 12B=\{0\},|G^{\prime }|<4|12B|$)

For a finite subset $B\subseteq {\mathbb{F}}_2^N$ containing zero, choose $m=\lceil {\log }_2(2|12B|)\rceil$ and a uniformly random linear map $L:{\mathbb{F}}_2^N\to {\mathbb{F}}_2^m$. Each fixed nonzero vector has zero image with probability $2^{-m}$. The union bound gives a map whose kernel meets $12B$ only at zero. It is injective on $6B$, and $2^m<4|12B|$. The Plünnecke-Ruzsa inequality makes the image of $B$ have density at least $1/(4C^{12})$ if $|B+B|\le C|B|$.

Under injective linear modelling of a small sumset, every $w\in W$ has a unique lift $\sigma (w)\in 4B$, because differences of two lifts lie in $8B\subseteq 12B$. For $w_1,w_2\in W$, the discrepancy $\sigma (w_1)+\sigma (w_2)+\sigma (w_1+w_2)$ lies in $12B\cap \ker L$, so it is zero. Thus $\sigma$ is a linear map and its image is a vector subspace of the same size as $W$. This verifies the additive lifting step rather than assuming that a preimage chosen from a sumset is automatically a subspace.