71,663 Approach Channel Encounters Ended With Give-Way Failure. Average CPA: 0.247nm — the Closest of Any Context in the Dataset.

The Setup

In the 45 days ending June 2, 2026, 516,368 vessel encounters were logged in approach zones — the constrained navigational corridors leading into and out of major ports. Within that population, 71,663 encounters carried a giveway_inaction trigger: one vessel had give-way obligation and did not maneuver, forcing the stand-on vessel to take avoiding action.

Those 71,663 events averaged 0.247nm minimum range at closest point of approach. That is the smallest average CPA of any encounter type across all three context zones in the dataset.

For reference:

| Context | Trigger | Events | Avg Min Range | Avg Conf | |-------------|-------------------|-----------|---------------|----------| | Approach | giveway_inaction | 71,663 | 0.247nm | 0.437 | | Approach | extremis | 210,631 | 0.342nm | 0.397 | | Open Sea | giveway_inaction | 159,918 | 0.272nm | 0.360 | | Open Sea | extremis | 466,163 | 0.311nm | 0.346 | | Anchorage | giveway_inaction | 27,265 | 0.264nm | 0.358 | | Anchorage | extremis | 92,269 | 0.400nm | 0.329 |

Approach channel giveway_inaction is closest. It is also the most confidently detected of any combination in the table.

The Chain

The COLREGS framework assigns stand-on and give-way obligations based on bearing geometry and vessel type. Give-way failure triggers when the obligated vessel does not maneuver within the classifier's detection window and the encounter continues to close — forcing the stand-on to deviate. In open water, the stand-on vessel has time and sea room: open-sea giveway_inaction events average 0.272nm, 10% further than approach.

In approach channels, the geometry changes. Channel width restricts available course changes. Traffic density compresses the spatial buffer between events. Pilot stations and VTS coverage is highest in approach zones — which is the likely explanation for the elevated confidence score (0.437): the encounters are more reliably observed because the zone is more heavily instrumented.

But higher observability does not mean more maneuvering room. When give-way failure occurs in an approach channel, both vessels have fewer degrees of freedom to resolve it. The stand-on maneuver that in open water might produce a 0.272nm CPA produces 0.247nm in an approach channel.

The 234,074 approach encounters with no handoff trigger — events that resolved without intervention — averaged 0.723nm CPA, nearly 3× the giveway_inaction average. The no-trigger population passed at comfortable distances. The failure population closed to within a quarter-mile.

Approach extremis events (210,631 at 0.342nm) show that when both vessels recognize the conflict and act together, outcomes are better. Mutual extremis in approach channels produces 0.342nm average CPA — 38% further than single-vessel give-way failure. Coordination helps even in constrained water.

The Implication

Port risk models that use encounter count or encounter rate as primary inputs will score approach channels as moderate-risk: highly managed, with VTS, pilots, and radar. The giveway_inaction CPA data suggests a different framing. When approach zone encounters fail — when give-way obligation is not met — they fail closer than any other zone.

A model that uses the zone-level average for approach encounters (which pulls the distribution right because 45.3% of approach encounters have no trigger and average 0.723nm) will understate the severity of the failure tail. The 71,663 giveway_inaction events represent 13.9% of all approach encounters — not a rare edge case.

The encounter type that produces the worst CPA outcomes is not extremis in open water, where most screening models focus. It is give-way failure in constrained channels, where both the geometry and the failure mode are most dangerous.

What to Watch

The giveway_inaction rate in approach zones. Currently at 13.9% of approach encounters (71,663 of 516,368). If this rate increases — more vessels failing to yield in constrained channels — the CPA distribution will tighten further. Open sea is currently at 15.6%; approach is 2 percentage points lower, meaning approach channels have marginally fewer failures per encounter but worse outcomes when they occur.

Any event-level geographic clustering in approach giveway_inaction would identify specific corridors where give-way failure is disproportionate. High-traffic chokepoints with concentrated approach traffic — the Houston Ship Channel, Delaware Bay, Puget Sound approaches — are the most likely candidates.

Limitations

min_range_nm and encounter_conf are computed fields derived from AIS position interpolation. The approach/open_sea/anchorage context_tag is assigned algorithmically based on proximity to port boundary and anchorage definitions. Boundary edge cases may classify some near-port open-water encounters as approach events and vice versa.

The 234,074 null-trigger approach encounters (45.3% of approach) are not failures — they are normal transits. Including them in the zone-level average produces a misleading overall mean. All CPAs cited in this analysis refer to the specific trigger populations, not zone averages.

Confidence scores reflect detection reliability, not event severity. A high-confidence event is one the classifier observed well; it is not necessarily more dangerous than a lower-confidence event that happened to be detected at a sharper angle.

Data as of 2026-06-02. Source: Overwatch pairwise_encounter, 45-day window. Total encounters across all context zones: 1,779,327. Approach zone total: 516,368.