Basemap and data are each correct but misalign when overlaid

• Data on an empty canvas sits exactly over the correct roads, but adding the Amap tile layer moves every feature a few hundred meters in one direction.
• The shift is constant within a city and roughly equals one WGS84 to GCJ-02 offset.
• In Cesium the vector layer floats above or sinks into the terrain because Cartesian3.fromDegrees expects WGS84 while the tiles are GCJ-02.

1. Hide the basemap and check the data alone.
2. Hide the data and check the basemap alone.
3. If both look right, compare the declared CRS of the two sources; a uniform few-hundred-metre shift means a datum mismatch.
4. If the offset is about twice the normal value, look for a second wgs84ToGcj02 call in the pipeline.
5. If the map is torn into mismatched tiles rather than uniformly shifted, suspect Baidu tile numbering rather than CRS.

Convert both sides to one CRS — normally GCJ-02 for Amap/Tencent/Baidu tiles, no conversion at all for Tianditu (CGCS2000) or OSM. For Baidu also switch to its own tile scheme instead of the standard {z}/{x}/{y} grid.

var gcj = wgs84ToGcj02(lng, lat);
var gcj2 = wgs84ToGcj02(lng, lat);   // 偏移变成两倍
var wrong = gcj02ToWgs84(lng, lat);  // 方向相反，越修越偏

• Converting WGS84 data to GCJ-02 while the basemap is already Tianditu CGCS2000, which introduces an offset that was not there.
• Feeding Baidu tiles into a standard Web Mercator tile layer and expecting a CRS fix alone to align them.