1,824 Nav-Status Dark Events in 90 Days. Critical-Tier Gaps Average 20.9 Hours — Shorter Than Medium-Tier's 35.1. The Anomaly Score Barely Moves.

The Setup

In the 90 days ending August 18, the nav_status subtype of dark_events fired 1,824 times — AIS gaps flagged specifically on navigational-status grounds, each stamped with a risk_tier (high 912, medium 319, unassigned 258, critical 185, low 150) and a kinematic_anomaly_score meant to quantify how anomalous the gap looks kinematically. 1,818 of the 1,824, 99.7%, have never been touched by adjudication_status — the review queue behind this subtype is functionally untouched.

The Chain

If risk_tier tracked kinematic_anomaly_score, the tier averages should step down cleanly from critical to low. They don't: critical events average 0.811, high 0.823, medium 0.786, low 0.775 — the entire ladder spans just 0.048, and the top tier scores below high. What does separate the tiers is gap_duration_hours, and it doesn't separate them monotonically either: high-tier gaps average 15.6 hours, critical-tier gaps average 20.9, medium-tier gaps average 35.1, and low-tier gaps run 62.7 — critical sits shorter than medium, not longer. Geographically the events are not scattered: all 1,824, 100%, carry a nearest_hotspot_id, averaging 65.3 nautical miles away.

The Implication

risk_tier on this subtype is not a readout of kinematic_anomaly_score — the two move almost independently, and where they do interact, duration doesn't even order tiers the way "critical > medium" would suggest. Whatever assigns tier here is answering a different question than the anomaly model, and an analyst triaging the queue by tier alone is prioritizing on a variable close to orthogonal to the score built to measure anomalousness. With 1,818 of 1,824 events still unreviewed, essentially none of this queue has been checked against either signal.

What to Watch

Whether the 6 nav_status events that have been adjudicated cluster in one tier or one score band — a first read on which signal, if either, adjudicators actually trust. Whether risk_tier gets revised if kinematic_anomaly_score is recomputed later. Whether the other dark_events subtypes — gnss_spoof, teleport, impossible_kinematics — show the same tier/score decoupling or whether it's specific to how nav_status gaps get tiered.

Limitations

This is a cross-sectional average comparison, not a per-event regression — a stronger relationship could exist within subgroups that group-level means wash out. risk_factors and kinematic_meta are null on every row sampled here, so the mechanics behind tier assignment aren't visible from the data itself. Ninety-day window only; earlier nav_status events aren't included.

Data as of 2026-08-21. Source: dark_events (subtype, risk_tier, kinematic_anomaly_score, gap_duration_hours, adjudication_status, nearest_hotspot_id), Axiom Overwatch.