Write one test question

Use a question that can be answered by observable evidence: “Does a receipt sent through provider A authenticate with the visible From domain?” is testable. “Why does the internet dislike our email?” is not. Choose a message type, From domain, provider configuration, and authorized recipient account that represent the problem.

Create a run sheet before sending. Record the UTC start time, sender workflow, envelope identity if available, DKIM selector, recipient provider, subject marker, and the exact variable under test. Use a marker such as [DL-BASELINE-20261020-01]; do not put credentials or customer data in it.

Capture the baseline

Send through the normal application path rather than a provider’s unrelated console test. Preserve the sending event and SMTP response. In the receiving account, open the original message and record the trusted receiver’s Authentication-Results, relevant DKIM-Signature fields, received time, and observed folder.

Sanitize before sharing. Keep domains needed for diagnosis, but remove recipient local parts, reset links, message bodies, and unrelated headers. A result from one mailbox is evidence about that test, not a promise about all recipients.

Change only one thing

If you are testing a new DKIM selector, keep the From address, message, sending provider, and recipient constant. If you are comparing content, keep authentication and infrastructure constant. Wait for the intended configuration to be active, then send a new message; an old received copy will not be re-signed after DNS changes.

Compare in a table:

Field Baseline Changed run
Receiver/time Provider and UTC Provider and UTC
SPF identity/result Recorded value Recorded value
DKIM domain/selector/result Recorded value Recorded value
DMARC result Recorded value Recorded value
Placement observation Folder or unknown Folder or unknown

Decide whether to keep the change

State what improved, what stayed the same, and what remains unknown. Roll back if the documented success condition was not met or a legitimate stream regressed. Repeat a successful run before broad rollout, then monitor normal traffic. A useful test produces a reproducible observation and a defensible next action—not a universal inbox claim.