Define states before the test

Use a small explicit state machine:

  • Inbox: the monitored message was found in the account’s defined inbox/category scope.
  • Spam: it was found in spam/junk.
  • Other: it was found in another visible folder or category your test distinguishes.
  • Missing: sender evidence says the receiver accepted it, the account monitor is healthy, and the message was not found by the observation deadline.
  • Rejected/deferred: SMTP evidence shows a receiver response before acceptance.
  • Unknown: account access, search, correlation, or monitoring failed; evidence is insufficient.

Do not force unknown into missing. Do not call an SMTP rejection “spam placement,” because the message was not accepted into that mailbox.

Correlate one message exactly

Give each test copy a non-sensitive run identifier in an approved header and subject marker. Store provider message ID, recipient seed ID, send time, remote SMTP response, and receiver message ID where available. Avoid embedding customer or campaign secrets.

Search the complete monitored scope after a defined initial and final window. Check inbox tabs/categories, spam, quarantine, and provider-specific “all mail” views according to your test contract. A provider can delay processing after SMTP acceptance.

Check the monitor before classifying

Record last successful login, token expiry, API/IMAP latency, folder map, search result count, clock drift, and a control message. If the monitor cannot see a known control, mark affected results unknown and alert operations. A dashboard full of “missing” during an expired-token incident is false precision.

Aggregate without hiding uncertainty

Show counts and denominator for each state by provider and account type. Report 8 inbox, 1 spam, 1 unknown, not 89% inbox after silently excluding the unknown. Include sample size and UTC observation window.

Provider dashboards such as Google Postmaster describe aggregate sender traffic and are not real-time placement logs for individual seeds. Use them as related evidence, not as a replacement for the observed account.

The result is defensible when another operator can trace each classification to a message and monitor state. The categories are observations in controlled accounts, not guarantees about every real recipient.