01 / THE OPEN-WEB CLASSICPreserve GUIDs as source identifiers
Preserve publisher-supplied item identifiers rather than deriving identity from a headline. Keep the display link separate, and include the source identity to avoid cross-source collisions. Decide on a documented fallback only when an identifier is absent or unusable. An edited title is a useful first fixture for verifying your update behavior.
02 / THE OPEN-WEB CLASSICTreat missing dates as missing information
Do not make an undated item look newly published by assigning it the current time. Keep the source value and the observation time separate. Your interface can use a clearly explained fallback ordering for undated entries. Test malformed dates and unexpected future values before using chronology to drive alerts or prominent placement.
03 / THE OPEN-WEB CLASSICSeparate parsing from safe HTML rendering
A parsed document can still contain content that needs a safe rendering policy. Keep extraction separate from display and prefer plain text when rich markup is unnecessary. Treat attachments as optional metadata rather than assuming they are images. A missing image should not prevent a reader from using the headline and source link.
Test a reader against imperfect items
Create a small fixture collection with a normal item, a missing date, a changed headline, embedded markup, and a non-image attachment. Decide the normalized output and the reader presentation before running the parser. Keep failure handling separate from transport errors. You will have a more useful compatibility suite than a collection of successful downloads from sources that happen to be easy.