Deliverability Monitoring and Mail Stream Separation

Google's published complaint threshold is 0.30 per cent and the ceiling I hold a promotional stream to is 0.10 per cent. That gap is the method. Postmaster Tools reports in daily buckets, one badly chosen segment can push a single day over the line, and recovering from that day costs weeks of sending. Everything below follows from wanting the margin rather than the threshold. What rotation means here, and what it does not: Streams separated onto their own subdomains with their own DKIM selectors: transactional on one, lifecycle on a second, promotional on a third, and above roughly a hundred thousand messages a month onto separate IPs. A complaint spike on a promotion should never be able to put a password reset or a policy document into a spam folder. That is the reason for the separation. Rotating sending identities to escape a reputation already earned is something I decline to build. Reputation attaches to the domain and to the content fingerprint, and the rotation buys days. Authentication, in the order it gets done: SPF and DKIM with 2048-bit keys on a rotation schedule, and alignment checked for every stream including the ones the client forgot about: the ticketing system, the survey tool, the payroll notifications. DMARC from p=none through a percentage ramp at quarantine to p=reject, with aggregate reports parsed weekly rather than filed. One-click unsubscribe per RFC 8058 with a List-Unsubscribe-Post endpoint that actually suppresses, required of bulk senders by Google and Yahoo since February 2024 and by Microsoft since May 2025. MTA-STS and TLS-RPT. BIMI only where a registered trademark already exists, because the certificate is not a project worth starting for its own sake. What gets monitored, and what I refuse to treat as truth: Postmaster Tools pulled daily by API into the client's own warehouse. Microsoft SNDS with the JMRP feedback loop registered. A seed panel read as direction only, since a seed panel measures placement for seed accounts and nothing else, and quoting it as an inbox rate is a vendor habit rather than a measurement. Ground truth is click rate by mailbox provider from the client's own logs, split out for Gmail, for Outlook, and for wp.pl, onet.pl, interia.pl and o2.pl, which decide whether a Polish campaign worked and which no international platform's dashboard separates for you. Suppression, written down and enforced: A sunset rule with a re-permission sequence ahead of it: typically three messages from around day 150, promotional sending stopped at 180 days without a click. For regulated senders the statutory stream is carved out and never suppressed. Tariff change notices, policy documents and safety communications reach everyone regardless of engagement, which is precisely why they must sit on a different subdomain from the promotional mail whose complaints would otherwise take them down with it. What I will not do: Point a list verification vendor at an engagement problem, since verification finds syntax errors and traps, not disinterest. Warm on engagement farms or seed networks. Buy or rent addresses to raise volume. Rewrite subject lines to slip past a filter, which fixes the day and confirms the reason you were filtered in the first place. Not included: Platform selection and migration management, template development, campaign and lifecycle strategy beyond suppression rules, SMS and push. Who this is not for: Senders under about fifty thousand messages a month, where the work costs more than the revenue it recovers. And anyone whose real problem is where the addresses came from; that is a different conversation and it starts with the file, not with DNS.