- the result title is clearly not original, usually derived from content on the page
- the original title is known to be generated
- the original generated title is as close to harmless as any web content could be
- the result title is actively harmful and misleading
- the original title is demonstrably better
- this idiosyncrasy is applied to very high trust hosts (eg GitHub)
- it’s not applied to the same content from obvious scraped content/spam/scam sites with obvious tells