These two paragraphs contradict each other.

When processing textual data, it is a schema definition error if the dfdl:alignment and dfdl:alignmentUnits properties are used to specify alignment that is not a multiple of the encoding-specified mandatory alignment.

If the data is not aligned to the proper boundary for the encoding when textual data is processed, then bits are skipped (parsing) or filled from dfdl:fillByte (unparsing) to achieve the mandatory alignment.

So is it to be SDE (conservative), or just move over to the mandatory alignment? E.g.,

<xs:sequence dfdl:encoding="utf-8" dfdl:alignment="7" dfdl:alignmentUnits="bits" dfdl:initiator="hello">....

The alignment property says 7 bits, but the initiator is utf-8 which requires 8 bits. So we either SDE, or we just add additional alignment bits to align the initiator text.