What must I proofread before e-filing my patent application?

The proprietary ABX format for US e-filings includes both XML and PDF versions of the to-be-filed patent application. Recall that according to the ABX 1.1 documentation:

The PDF files are the source documents for the official records in the Image File Wrapper system (IFW) of the specification, claims, abstract and drawings. It is important to review the PDF files to verify that they contain the same information as was entered into the Word source application file. The Word source document file is never transmitted to the USPTO, and cannot be relied upon as evidence of the contents of the application as filed.

The XML and TIFF files are the source documents for the bibliographic information used in the USPTO records, for the 18-Month Publication (PG-Pub) document and for other USPTO records respectively as submitted.

From this description in the ABX 1.1 documentation, it seems the filer must carefully proofread both the PDF and the XML versions. If either version is missing anything or is wrong, the filer risks losing substantive rights. For example, if there were something wrong with the XML file, then the 18-month publication would be defective. The publication would be less effective as a defensive publication, and the ability to collect pre-grant damages would be put into question. It is not clear what the "other USPTO records" are that somehow derive from the XML file, but whatever they are, they would presumably be harmed as well by any problem with the XML file.

We have found several test cases where the PDF and XML files are decidedly non-identical. In one case, Greek letters were present in the PDF and missing from the XML. In another case, inline images were present in the PDF and missing from the XML. The moral of the story is that one must proofread both the XML and PDF versions of the application character by character so that you can catch things like this.