When building the database, the taxid in the hash table was not checked. Only the existence of a minimizer key was queried. If the same minimizer exists in different taxid sequences, then the compare_and_set function should be executed.

[eric9n]

https://github.com/DerrickWood/kraken2/blob/acc22481bbeb32b870c12b403ce595fe7a6db770/src/build_db.cc#L382C7-L390C8

for (size_t i = 0; i < mm_ct; i++) { // once we've determined that a minimizer won't be an insertion, // don't bother calling FindIndex() again if (insertion_list[i]) { size_t idx; insertion_list[i] = ! hash.FindIndex(minimizer_list[i], &idx); index_list[i] = idx; } }

  // Determine safe prefix of sorted set to insert in parallel
  std::set<uint64_t> novel_insertion_points;
  size_t safe_ct = 0;
  for (safe_ct = 0; safe_ct < mm_ct; safe_ct++) {
    if (insertion_list[safe_ct]) {
      if (novel_insertion_points.count(index_list[safe_ct]) > 0)
        break;
      novel_insertion_points.insert(index_list[safe_ct]);
    }
  }

  // Adjust CHT values for all keys in the safe zone in parallel
  #pragma omp parallel for
  for (size_t i = 0; i < safe_ct; i++) {
    SetMinimizerLCA(hash, minimizer_list[i], taxid, tax);
  }