Upstream contribution · Thysrael/Horizon
Horizon · Keeping healthy GDELT items in Horizon source batches
Problem
One GDELT entry with an unexpected field type could raise before validation and discard the whole batch, including otherwise usable articles.
Engineering response
Horizon now validates the title, URL and date as strings before cleaning their values, so the reader can reject the malformed item and continue through its siblings.
Evidence and scope
The merge record reports that a mixed-batch regression failed on the earlier code and the focused GDELT suite passed 11 tests after the change, with HTTP fixtures.