Freeze the baseline definition

For each stream, record From/return-path/DKIM identities, selector, outbound IP/pool, provider account/region, volume/rate, templates/MIME, links, bounce/complaint definitions, suppression, and representative SMTP/seed outcomes. Save UTC times and monitor health. Export only aggregated/sanitized evidence where possible.

Define old and new event mappings. “Delivered,” “deferred,” and “bounce” can differ between providers; translate to the underlying SMTP/evidence states before comparing charts.

Stage the new path

Verify domains/accounts, publish a new DKIM selector, configure return path/SPF as required, and test DMARC alignment. Import suppression with reconciliation and idempotency. Confirm unsubscribe, complaint, bounce, and webhook authentication before sending production traffic.

Generate the same approved message through old and new paths to controlled recipients. Inspect received originals for rendering, authentication, links, size, headers, and outbound route. Note unavoidable differences.

Build a comparable test series

Run paired seed tests in adjacent windows with stable account panel and health controls. Separate transactional workflows from campaigns. Record inbox/spam/other/missing/unknown and SMTP outcomes by provider. Repeat to estimate normal variance.

Do not “warm” the new system by sending to unengaged or unpermitted recipients. Use authorized engaged cohorts and provider-recommended rate patterns.

Cut over in bounded cohorts

Define cohort, volume, timing, success metrics, stop conditions, and rollback. Monitor authentication, deferrals, permanent failures, complaints, unsubscribe propagation, provider dashboards, seed observations, queue latency, and support reports. Keep old and new sending mutually exclusive for each recipient/event to prevent duplicates.

Retire after evidence

When new traffic is stable, stop old jobs and credentials, retain old DKIM public keys through the verification overlap, preserve required suppression/audit data, and remove obsolete SPF authorization later. Keep rollback available for the agreed window.

The placement comparison supports the migration when test conditions and metric definitions are documented. It cannot prove every placement change came from the platform, but it can reveal where evidence diverged and whether the new path meets the accepted release gates.