Oh man, the "slight header variation" part hits home. We had a similar greylisting issue that turned out to be a trailing space in the `Reply-To:` header from a CRM integration. It only manifested on Outlook/Exchange recipients, and only for sends after 1PM their time. Took us a month of staring at graphs to spot the pattern.
Separate monitor is the only way. Campaign averages are a lie 😅
it worked on my machine
When even your hot segment is tanking, it's an infrastructure issue, not a list issue.
Check your DMARC aggregate reports for any sends not aligned with your domain. It's a long shot, but I've seen a bad actor in a shared IP pool spoof other senders and drag the whole pool down. If you find anything misaligned, that's your ammo for billing.
For transactional emails landing in spam, isolate one of those streams and run it through a secondary SMTP relay for a week. If performance jumps, you've got your proof the pool is poisoned.
Benchmarks or bust.
The DMARC aggregate report check is a crucial forensic step that's often overlooked. While it can reveal domain spoofing within the shared pool, a clean report doesn't exonerate the pool; it just means the damage is being done at the IP reputation layer, which is invisible to your domain's alignment data.
Your secondary SMTP relay test is the definitive control, but to make it statistically valid, you need to run it as an A/B split on a single, homogenous segment, not just isolate a stream for a week. Randomly assign half the segment's sends to the relay and half to Mailchimp's pool within the same time window to eliminate day-of-week or time-of-day biases that could skew the results.