Extremis Is the Dominant Rule 17 Path. Every One of the 669,182 Closed Within 185 Meters.

The Setup

In 90 days across 2.89 million tracked vessel encounters, the Rule 17 handoff trigger that fired most often wasn't give-way failure. It was extremis.

Extremis — the threshold where collision is imminent and the stand-on vessel is permitted, then required, to act — triggered 669,182 Rule 17 deviations in the dataset. Give-way inaction, the more commonly discussed failure mode, triggered 202,729. Extremis is 3.3 times more common.

The Chain

Rule 17 of COLREGS defines what happens when two vessels are on a collision course. The give-way vessel is supposed to maneuver early. The stand-on vessel is supposed to hold course and speed until the give-way vessel acts. If the give-way vessel doesn't act, Rule 17(a)(ii) and eventually 17(b) kick in: the stand-on vessel may, then must, take avoiding action.

There are two observable paths to that endpoint. In the first, give-way inaction is the trigger: the give-way vessel fails to act early enough, forcing the stand-on vessel to deviate. This is an early-warning failure — Rule 17 fires because the protocol broke down at the first step. In the second — extremis — the geometry has already deteriorated to collision-imminent before any Rule 17 logic applies. Both vessels are maneuvering simultaneously under 17(b).

The data confirms which path dominates: extremis, 3.3 to 1 over give-way inaction.

More important than the count is the geometry. Extremis-triggered encounters produce an average DCPA of 0.036nm. The median is 0.029nm — 54 meters. Every single one of the 669,182 extremis-triggered encounters closed within 0.1nm (185 meters). The give-way inaction path produces an average DCPA of 0.111nm — 3.1 times wider — and only 46.2% of give-way inaction encounters reached the sub-0.1nm threshold.

The average intervention window — measured from risk-window-start to CPA — is 468 seconds across all extremis encounters. Roughly 7.5 minutes. That window is consistent across all three contexts: 491 seconds on the open sea, 445 seconds in approach channels, 476 seconds in anchorages.

One context stands out on rate: anchorage encounters had the highest Rule 17 deviation frequency of any setting. 53.21% of anchorage encounters triggered a Rule 17 deviation, versus 49.15% on the open sea and 47.20% in approach channels. The 172,598 extremis events in anchorage zones produced average DCPA of 0.038nm — essentially identical to open-sea and approach geometry.

The Implication

When extremis dominates the Rule 17 trigger distribution, the collision-avoidance system is regularly waiting until the last available moment before acting. The geometry — median 54 meters of separation at CPA — leaves no margin for sensor error, helm-response latency, or a third vessel in the picture. The 7.5-minute average window sounds generous; at 10 knots combined closure speed, two vessels cover 0.1nm in 36 seconds.

The 669,182 extremis encounters don't represent near-misses in the legal or casualty sense. They represent encounters that resolved within a geometric envelope that is defined by immediate mutual action rather than orderly sequential avoidance. The stand-on vessel moved because waiting was no longer an option, not because the give-way vessel acted too late.

That distinction matters for regulatory framing: give-way inaction generates a clear COLREGS violation attribution. Extremis doesn't — Rule 17(b) authorizes exactly what happened. The 669,182 extremis events are, in a technical sense, compliant outcomes. They are also the closest outcomes in the dataset.

What to Watch

The anchorage rate is the most actionable signal here. Anchorage zones produce Rule 17 deviations at the highest rate (53.21%) and extremis events that close to the same geometry as open-sea encounters, despite nominally lower vessel speeds. Traffic separation schemes, approach channel monitoring, and near-miss reporting systems are typically weighted toward underway encounters in confined water approaches. The 172,598 anchorage extremis events suggest the monitoring emphasis may be misaligned.

Limitations

This dataset captures encounters where AIS position data was continuous enough to compute DCPA and classify the rule17_handoff_trigger. Encounters with incomplete AIS tracks are excluded — which may mean the extremis count is conservative if tracking gaps correlate with close-quarters maneuvering. TCPA calculations show negative values for most records, indicating CPA has already occurred at record time; the geometry numbers reflect historical encounter resolutions, not live predictions. Encounter confidence scores average 0.314 for extremis events (versus 0.329 for give-way inaction), meaning the extremis classification carries more uncertainty. Average intervention window (468s) is measured from risk-window-start to CPA as recorded; actual mariner decision windows are narrower once situational awareness acquisition is accounted for.

Data as of 2026-07-07. Source: Axiom Overwatch pairwise_encounter, 90-day window ending 2026-07-07.