The spam that is getting through seems like really obvious spam. Not just for a human, but for a classifier, too.
I looked at an "out of the blue" example just now:
1. Sent from a Gmail address where the name doesn't come close to matching the name suggested by the sender address.
2. Email body is in Latvian, a language I have no association with.
3. About 20 other recipients, none associated with me.
4. Subject line is nonsensical (not even words) and an email address.
5. Email body is one line.
6. PDF attachment, with no mention of it in the email body.
7. Looks automated, but came from a Gmail address.
Of course, none of these in isolation is a definitive indicator of the email being spam, but given that there's at least 7 anomalies, considering the amount of data Google has & that they pride themselves on machine learning, shouldn't they be catching something like this?
Arriving in the inbox is one thing, but the thing really throwing me is they're getting the "important according to Google magic" flag applied. How can mail that fails all the tests you've mentioned (I just ran through them on some spam I've been getting) get such a high importance rating?
tbf, the Latvian bit probably counts in its favour because I doubt many spammers bother with Latvian (or that it's particularly unusual for Latvian speakers to have English Gmail accounts).
Still scratching my head at how their calibration can let something like that through and stick a random genuine circular from Twitter in my spambox though...
I looked at an "out of the blue" example just now:
1. Sent from a Gmail address where the name doesn't come close to matching the name suggested by the sender address.
2. Email body is in Latvian, a language I have no association with.
3. About 20 other recipients, none associated with me.
4. Subject line is nonsensical (not even words) and an email address.
5. Email body is one line.
6. PDF attachment, with no mention of it in the email body.
7. Looks automated, but came from a Gmail address.
Of course, none of these in isolation is a definitive indicator of the email being spam, but given that there's at least 7 anomalies, considering the amount of data Google has & that they pride themselves on machine learning, shouldn't they be catching something like this?