Incident teardown ยท northstargoods.com ยท 9 June 2026

The send behind the spike was the smallest of the week.

Gmail complaint rate doubled on a day the account sent 46% less mail than its baseline. If you are watching volume for a cause, this is the day you find nothing.

Peak spam rate
0.230% (9 June)
30-day baseline
0.115%
Mail sent that window
39,901/day vs 73,667 baseline
Recovered in
5 days, to 0.085%

Less mail, worse rate

The account sent 46% less through this window than its 30-day baseline โ€” 39,901 a day against 73,667 โ€” and the spam rate doubled anyway. Authentication was stable throughout.

This is the shape that defeats the usual instinct. Complaint rate is complaints divided by delivered, so a smaller send to a worse-matched audience beats a large send to a good one, every time. Cutting volume does not fix a rate problem; it can make it worse by shrinking the denominator.

Which send

Google published four feedback-loop identifiers for the window. Two resolved to real Klaviyo campaigns. The larger of the two:

CampaignDeliveredShare of that dayGoogleโ€™s spam ratio
Reactivation ยท Offer ยท US14,94537%0.14%
Reactivation ยท Offer ยท FR4,66112%published, no ratio given

Just under 15,000 emails โ€” a third of a quiet dayโ€™s volume โ€” carrying a published spam ratio above Googleโ€™s recommended line on its own.

The day before mattered too

Two campaigns sent on 8 June, each about 30% of that dayโ€™s volume, sit immediately before the window opens. Google published no identifier for either, so we cannot name them the way we can name the 9 June sends. Klaviyo did record two spam complaints against one of them inside the window.

That is likely, not identified, and the difference is the whole discipline. Timing plus volume share is a reason to look; a feedback-loop identifier carrying the message ID is a reason to act. Collapsing the two is how deliverability advice becomes superstition.

What this costs you to know

Nothing, if you only need the first half. Google publishes the rate, the curve and the requirements for any domain you verify, straight in Postmaster Tools. The part that took a licence to write is the join: 502 feedback-loop identifiers decoded on this account, 376 of them resolved to a campaign or flow by name.

The remaining 63 are all-digit tokens that resolve to nothing. We show them as unidentified rather than guessing, because a confident wrong name is worse than an honest gap.