When columns from different data-frames that have a common lineage are used in inequality conditions in joins, they are not resolved correctly. In particular, both the column from the left DF and the one from the right DF are resolved to the same column, thus making the inequality condition either always satisfied or always not-satisfied.
Minimal example to reproduce follows.
The result is:
which violates the condition that the timestamp from the right DF should be between df_left["timestamp"] and df_left["timestamp"] + 2.
The plan shows the problem in the column resolution.
Note, the columns used in the equality condition of the join have been correctly resolved.