Incident teardown · northstargoods.com · 11–13 June 2026
Three variants of one campaign. Google ranked them.
The spam rate more than doubled while volume barely moved and authentication never wavered. Google published twelve feedback-loop identifiers across the window; ten resolved to real Klaviyo sends, and three of them were the same campaign aimed at three different audiences.
- Peak spam rate
- 0.283% (13 June)
- 30-day baseline
- 0.117%
- Google's line
- 0.10% recommended, 0.30% policy
- Recovered to
- 0.085%
First, what it was not
Two explanations get reached for before the evidence does, and both are checkable, so we check them first.
It was not volume. The account sent 70,011 a day through the window against a 64,720/day baseline — 8% more mail. An 8% increase in sending does not produce a 142% increase in complaint rate.
It was not authentication. SPF, DKIM and DMARC all stayed within their prior range for the whole window. Nothing broke in the plumbing.
What Google actually published
Postmaster reports a spam rate per day, and separately a set of opaque feedback-loop identifiers — strings like 01KTX771HZJK9HCA1VCTZJZ0GR — each with its own spam ratio. Google will not tell you what they are. They are Klaviyo message identifiers, and once you hold both sides you can join them.
Twelve identifiers appeared across this window. Ten matched a real campaign or flow. Three of those ten were the same promotion, sent on 12 June to three different audiences:
| Audience variant | Delivered | Share of that day | Google’s spam ratio |
|---|---|---|---|
| Win-back · lapsed 180 days | 47,413 | 31% | 0.13% |
| Prospects · list import | 22,991 | 15% | 0.09% |
| Main list · engaged | 35,949 | 23% | 0.05% |
Same offer, same day, same sending domain. The win-back segment complained at 2.6 times the rate of the main list, on the largest send of the three.
Why that matters more than it looks
Read at the campaign level, this is one promotion with a slightly elevated rate, and the obvious response is to send fewer promotions. Read at the variant level, it is a single audience misbehaving inside an otherwise healthy send — and the fix is that segment, not the campaign, not the calendar.
One of those two readings costs a marketing programme its promotional cadence. The other costs one segment a review. Google published the difference; it just published it as a hex string.
The flow underneath
Also inside the window, the abandoned-cart flow ran at 4.2× its normal daily volume — 2,735/day against a 650/day baseline. Flows do not appear in a send calendar, so a window like this is routinely blamed on the campaign that happens to sit in it.
What we would not claim
A feedback-loop identifier that contains a campaign’s ID and carries a published spam ratio is strong evidence that Gmail users reported that message. It is not proof that this campaign alone moved the domain’s rate. Three sends were named, a flow surged, and the window covers three days. We label what is identified, what is merely likely, and what is only possible, and we do not collapse those into a single confident sentence.
The window closed at 0.085%, back under Google’s recommended line.