We're gleaning much of the metadata using regular expression parsing of the plain text. This is bound to be incomplete and occasionally inaccurate. Create a spreadsheet containing all the structured metadata for each entry and a process by which the values in the spreadsheet can be treated as authoritative.