Here is an explicit construction, not just a picture. Number rows and columns from 1 to n, put n=k2, and leave black the cell Br=(r,k((r−1)modk)+k−⌊(r−1)/k⌋) for each 1≤r≤n. For every 0≤p,q≤k−2, place one k×k square tile Tp,q whose row interval is [2+kp+q,1+k+kp+q] and whose column interval is [k−p+kq,2k−1−p+kq]. The remaining boundary rectangles are these: for 1≤q≤k−1, rows [1,q] and columns [kq+1,k(q+1)]; for 0≤p≤k−2, rows [kp+1,k(p+1)] and columns [1,k−1−p]; for 1≤p≤k−1, rows [kp+1,k(p+1)] and columns [k2−p+1,k2]; for 0≤q≤k−3, rows [k2−k+2+q,k2−1] and columns [kq+1,k(q+1)]; and one last rectangle in row k2 and columns [1,k2−k]. Each displayed interval is nonempty in its stated range. The black-cell formula has one cell in every row and column. A direct interval check shows that the k×k tiles and these boundary rectangles are pairwise disjoint, avoid every Br, and cover every other cell: the four families fill respectively the top, left, right, and bottom gaps around the (k−1)×(k−1) array of square tiles. There are (k−1)2 square tiles and (k−1)+(k−1)+(k−1)+(k−2+1)=4(k−1) boundary tiles, hence (k−1)2+4(k−1)=k2+2k−3 tiles.