legacy VTK with insufficient precision?
Since PR !1694 (merged), legacy VTK data is written out with 11 digits of precision for double
data.
I don't quite understand where that number of digits comes from. Generally, a double float (https://en.wikipedia.org/wiki/Double-precision_floating-point_format) has 52 bits of mantissa, giving rise to log(2) / log(10) * 52
(~16) significant digits.