Confirm the difference is real

Check seed-monitor health, message correlation, send times, and SMTP acceptance for each provider. Repeat the controlled run. A provider showing unknown because its account token expired is not a placement difference.

Report raw counts and account types: “3/3 authorized Yahoo seeds observed inbox; 1/3 personal Gmail seeds observed inbox, 2 spam.” Do not combine consumer, enterprise, and gateway accounts under a single provider label if their paths differ.

Compare the received evidence

For each provider, inspect trusted Authentication-Results, Received path, From/return-path/DKIM identities, ARC where relevant, TLS indication where available, message size/MIME, links, and timestamps. Confirm the sender platform used the same account and expected outbound pool.

Look at SMTP responses even when accepted. Deferrals, throttling, and route changes can reveal a provider-specific issue that a final placement label hides.

Add official provider data

For Gmail, review current Postmaster compliance, spam rate, authentication, feedback-loop, and delivery-error data for the appropriate domain and period. Google states this data is delayed, can be absent at low volume, and applies to personal Gmail traffic. For Yahoo, compare current authentication, complaint, unsubscribe, DNS, and audience practices with official sender guidance.

Do not apply one provider’s threshold or dashboard definition to another. Keep public blocklist results separate from the receiver’s private reputation.

Form ranked hypotheses

Examples include provider-specific complaint history, authentication alignment for one route, recipient-cohort expectation, volume/rate change, provider account/pool selection, or message transformation. Assign confidence and evidence to each. “Provider hates our template” is not a testable root cause.

Change and verify

Contain any permission or suppression issue immediately. Otherwise change one supported factor, send a fresh paired test, and watch the provider’s normal feedback window. Keep a rollback.

The investigation is complete when it explains what evidence differs and what next test follows. Some receiver decisions remain proprietary; label them unknown rather than reverse-engineering a universal rule from a small seed panel.