Anchorage Encounters Produce Rule 17 Deviations at a Higher Rate Than Approach Channels. 1.9 Million Encounters Across Three Contexts.

The Setup

In 45 days, 1,901,932 vessel encounters were classified across three contexts: open sea (1,059,389), approach channel (578,267), and anchorage (264,276). Each pair was evaluated for Rule 17 compliance — did either vessel execute a course change that violated its stand-on or give-way obligation?

The deviation rates:

• Anchorage: 53.5% of encounters included a Rule 17 deviation by at least one vessel
• Open sea: 49.9%
• Approach channel: 47.2%

Anchorage has the highest rate. Approach channel — where a COLREGS failure carries the most severe consequences due to restricted water, current, ferry crossings, and limited sea room — has the lowest.

The Chain

The average CPA (closest point of approach) provides the contrasting view:

| Context | Rule 17 Deviation Rate | Avg CPA (nm) | Pct Under 0.1nm | |---|---|---|---| | Anchorage | 53.5% | 0.576 | 10.5% | | Approach channel | 47.2% | 0.519 | 14.8% | | Open sea | 49.9% | 0.465 | 16.9% |

Vessels at anchorage deviate most frequently — 53.5% — but end up farthest apart at CPA. Vessels in open water deviate at 49.9% and come closest on average, with 16.9% passing within 0.1nm. The deviation rate and proximity risk are not correlated the way a safety model would expect.

One explanation for the anchorage pattern: stand-on vessels anchored in a crowded roadstead may maneuver preemptively to avoid a swinging chain or a passing vessel in the lane. That maneuver classifies as a Rule 17 deviation — the stand-on vessel moved when it did not need to — but the motivating geometry is anchor constraint, not COLREGS misunderstanding. The deviation is real; the compliance inference may not be.

Approach channel shows the inverse. Stand-on vessels hold course at a 47.2% deviation rate (i.e., 52.8% of the time, neither vessel deviates). But 14.8% of approach encounters pass within 0.1nm — the second-highest close-approach rate — meaning vessels are coming very close without triggering deviation. Either compliance is strong enough that give-way vessels are acting correctly and stand-on vessels are holding appropriately, or the geometry is tighter than the deviation signal suggests.

The Implication

Risk-scoring models that use Rule 17 deviation rate as a proxy for compliance quality will flag anchorages as higher-risk than approach channels. The anchorage deviation rate (53.5%) is genuinely higher — the number is real — but the underlying mechanism is likely anchor-constrained maneuvering, not compliance failure. Models that do not separate these mechanisms will misallocate analyst review: pulling anchorage encounters for adjudication at a higher rate than approach encounters, which carry closer actual CPAs and higher consequence per event.

The approach channel picture is more operationally concerning than its deviation rate suggests. A 47.2% rate — the lowest of the three contexts — combined with 14.8% of encounters under 0.1nm and an average CPA of 0.519nm means the "well-behaved" approach channel is also where vessels come closest in constrained water. Close-approach events in approach channels that do not generate a deviation flag may not surface in standard alert queues at all.

In 45 days, 943,340 Rule 17 deviations were recorded across all three contexts — an average of roughly 20,963 per day. Approximately 28,200 of those (anchorage context, under-0.1nm subset) combined a deviation with a sub-0.1nm CPA.

What to Watch

The both_deviate_pct — the rate at which the classifier assigns deviations to both vessels simultaneously — is 0.0% across all three contexts. This is a classifier design constraint: the model assigns the violation to one vessel per encounter pair. Bilateral deviation, which is operationally common in congested waters, is not surfaced. Cases where both the stand-on and give-way vessel acted outside their COLREGS role are collapsed into a single-vessel attribution, reducing the apparent complexity of the interaction.

This matters most in anchorage and approach contexts where both vessels may have operational reasons to maneuver, and where the single-attribution design undercounts the full compliance load on the encounter.

Limitations

"Rule 17 deviation" is a model label, not a legal finding or a mariner's incident report. The classifier infers obligation and deviation from AIS course changes; it cannot observe pilot instructions, master override decisions, or autopilot behavior. Anchor-constrained maneuvering is a plausible explanation for the elevated anchorage deviation rate but has not been validated against ground truth. The 45-day window may not smooth seasonal variation in anchorage occupancy or approach channel traffic density. The both_deviate_pct = 0.0% is a classifier property, not an empirical observation — it does not mean bilateral deviation never occurs.

Data as of 2026-06-09. Sources: pairwise_encounter (45-day window, n=1,901,932).