It looks like only about half the data is parsed, and the rest is skipped due to dfdl:length on a complex element. It unparses as expected, but the data is now much shorter and so it does not round trip correctly. We need to determine if our schemas for the message type are wrong, if the documentation from which the schemas were derived is wrong, or if the tool creating the example file is broken for this message type.

I appears that the schema does correctly mtach the documentation, so it's likely either a documentation problem or a tooling problem.