The Setup
Ship-to-Ship Transfer is the motif detector that flags at-sea cargo handoffs between two vessels — the maneuver that lets cargo change custody without a port call, a document trail, or a customs stamp. Over the trailing 90 days it filed 35,377 alerts. Every one of them landed at confidence 1.000 and robustness 1.000, the detector's ceiling on both axes — no alert in the batch was ever anything less than maximum-certainty. On July 12 at 09:54 UTC, it fired for the last time. As of this analysis, 26 days later, it has not fired again.
The Chain
The silence is not a low-volume detector going quiet the way it always does. STS spent three months producing max-confidence output at a rate of roughly 390 alerts a day, then stopped completely and abruptly rather than tapering off. Two days before the stop, STS had just retaken the lead among Overwatch's five motif classes — the loudest signal in the queue. Forty-eight hours later it produced zero.
That queue never had a review pass to begin with, so the stop is invisible from inside the system. Across all motif classes, 113,248 alerts are currently sitting unacknowledged, the oldest dating back 89 days to May 10. STS wrote roughly a third of everything filed into that queue over the past 90 days and then exited it mid-stream — unresolved, unflagged, indistinguishable in the backlog from the 32,211 AIS_manipulation_dark alerts and 17,793 AIS_manipulation_spoof alerts that kept arriving in the 45 days after STS went dark. Nothing in the pipeline treats "a detector that fired 35,377 times just stopped" as an event in its own right. It is absorbed into the same undifferentiated backlog as everything else, where a stopped signal and an active one read identically: unreviewed rows in a table.
The Implication
A behavioral detector going fully silent after three months of maximum-confidence output is itself informative, independent of what it was originally built to catch. Two explanations compete and the data on hand cannot distinguish them: either the underlying activity genuinely stopped — a real behavioral shift in how cargo is being moved — or the detector broke. A detection pipeline that logs alerts but never logs detector health has no way to tell those apart, and 26 days is long enough that the distinction now matters. If the activity moved rather than stopped, it moved to a motif class this system either hasn't built a detector for or one already misclassifying it as something else.
What to Watch
Whether STS resumes at its prior rate (390/day), resumes at a materially lower rate, or stays dark past 30 days — the point at which "gap" reads as "regime change" rather than noise. Also worth tracking: whether AIS_manipulation_dark's volume shows any inflection around July 12, which would support the substitution read over the detector-failure read.
Limitations
There is no ground-truth signal here for which explanation is correct — no detector-health metric, no manual audit of a sample of vessels that were previously flagged, no confirmation that the underlying AIS feed itself didn't have a gap in this window. The 90-day and 45-day windows in this piece use different lookback periods for different comparisons; totals are not directly additive across them. A 26-day silence following 35,377 alerts at uniform maximum confidence is unusual enough to note, not long enough to call conclusively either a resolved behavior or a broken pipeline.
Data as of 2026-08-07. Source: motif_alerts, Axiom Overwatch maritime intelligence pipeline.