Phylogenetic trees play a key role in the reconstruction of evolutionary relationships. Typically, they are derived from aligned sequence data (like DNA, RNA, or proteins) by using optimization criteria like, e.g., maximum parsimony (MP). It is believed that the latter is able to reconstruct the \enquote{true} tree, i.e., the tree that generated the data, whenever the number of substitutions required to explain the data with that tree is relatively small compared to the size of the tree (measured in the number $n$ of leaves of the tree, which represent the species under investigation). However, reconstructing the correct tree from any alignment first and foremost requires the given alignment to perform differently on the \enquote{correct} tree than on others. A special type of alignments, namely so-called $A_k$-alignments, has gained considerable interest in recent literature. These alignments consist of all binary characters (\enquote{sites}) which require precisely $k$ substitutions on a given tree. It has been found that whenever $k$ is small enough (in comparison to $n$), $A_k$-alignments uniquely characterize the trees that generated them. However, recent literature has left a significant gap between $n\leq 2k+2$ -- namely the cases in which no such characterization is possible -- and $n\geq 4k$ -- namely the cases in which this characterization works. It is the main aim of the present manuscript to close this gap, i.e., to present a full characterization of all pairs of trees that share the same $A_k$-alignment. In particular, we show that indeed every binary phylogenetic tree with $n$ leaves is uniquely defined by its $A_k$-alignments if $n\geq 2k+3$. By closing said gap, we also ensure that our result is optimal.