This slightly complicates the code, but makes a **massive** difference in memory efficiency - preliminary results show a ~30% reduction in heap usage, with no measurable performance impact otherwise.