The aggregation multiversion fuzzer currently treats different string renderings of NaN as result mismatches (e.g. "NAN"vs "-NAN(IND)") even though the underlying numeric value is NaN in both versions. This causes spurious BF failures like BF-42477 when $stdDevSamp produces NaN and $toUpper is applied to its string form. Update the fuzzer’s result-comparison logic (e.g. AggCompareResults / validateQueryResults) to canonicalize NaN string representations on both sides before deep comparison, so that platform or CRT-specific NaN spellings are considered equal while preserving sensitivity to real result divergences.