4,433 COLREGS Encounters, Zero Analyst Reviews. Head-On Events Need 23.8 Degrees and Carry the Least Classifier Certainty.

The Setup

4,433 COLREGS encounters were classified in the past 45 days. Zero have been reviewed by an analyst.

That zero is not unusual — automated classification systems generate volumes that outpace manual review. What is notable is the pattern inside those unreviewed records: compliance deficit varies by encounter type in a direction that should be uncomfortable for any risk scoring model.

Head-on encounters (Rule 14: both vessels must alter to starboard) require an average of 23.8 degrees of starboard alteration. They carry 81.3% rule certainty. Crossing encounters (Rule 15: give-way vessel alters to starboard) require 11.4 degrees and carry 96.7% rule certainty.

The most geometrically demanding encounter type — head-on — is the one the classifier is least confident about.

The Chain

Rule 14 applies when two vessels meet on reciprocal or nearly reciprocal courses. Both must alter to starboard. The compliance deficit metric captures how far the actual track fell short of the required separation, measured in nautical miles at the closest point of approach.

For head-on encounters, the average compliance deficit is 0.471nm. The 95th-percentile deficit is 1.197nm — meaning the top 5% of head-on events fell more than 1.2nm short of the required separation threshold. That is a large tail.

Crossing encounters show a slightly smaller average deficit at 0.421nm but much higher classifier certainty (96.7%). Their p95 deficit is 0.951nm — roughly half the head-on p95. Overtaking encounters (Rule 13: overtaking vessel keeps clear) carry the smallest average deficit at 0.332nm and 95.7% certainty. These are geometrically cleaner because bearing rate distinguishes overtaking from head-on clearly at encounter initiation.

The 81.3% rule certainty for head-on events is lower because identifying a true head-on from AIS track data is geometrically difficult. Vessels approaching on near-reciprocal courses can appear as crossing situations depending on bearing rate and speed differential at the moment of classification. The classifier assigns approximately 19% probability to an alternative scenario in the average head-on record.

All 4,433 records show observed_fraction = 1.000, meaning the full encounter arc was captured. The compliance deficits are not estimation artifacts from partial tracks.

The Implication

A risk model that weights classifier certainty in its scoring will systematically underweight head-on encounters. The vessel accumulates less risk signal from a head-on event — 81% certainty, 0.471nm average deficit, 1.197nm p95 deficit — than from a crossing event at 97% certainty and 0.421nm average deficit. The most demanding encounter type generates the weakest signal downstream.

The 47 head-on records in this window are the smallest category by count. Some fraction may be misclassified — crossing situations assigned as head-on due to bearing rate ambiguity near the CPA. The compliance deficits for misclassified records are computed against the wrong required maneuver, so they can inflate or deflate the deficit unpredictably. Without analyst review, those records pass through the pipeline at face value.

The mixed-category records (64 events, 52.8% certainty) are the highest-uncertainty set — the classifier cannot confidently assign a single COLREGS scenario for nearly half its probability mass. These would be first in any triage queue. They represent 1.4% of the total.

The zero-review rate across 4,433 records means no human has evaluated whether the 47 head-on events with p95 deficits above 1.197nm are valid detections or classifier noise.

What to Watch

The head-on dataset is small at 47 events in 45 days. A month-over-month comparison would indicate whether that rate is typical. If the head-on count is stable, the focus should shift to the p95 deficit — specifically whether the 1.197nm tail is driven by a small number of repeated vessels, flag states, or geographies.

The crossing dataset (2,385 events, 96.7% certainty) is the most analytically tractable: large enough to support vessel-level and flag-state breakdowns, high enough certainty to treat individual records as reliable. A vessel-level analysis of crossing compliance deficits has not been run against this window.

Limitations

Compliance deficit captures geometric shortfall against the COLREGS-required track — not a finding of fault or regulatory violation. A positive deficit means the vessel did not achieve the modeled required separation; it does not establish intent or operator error. The 81% rule certainty for head-on events means roughly 1 in 5 records may be misclassified, with compliance deficits computed against the wrong scenario. The 47-event head-on sample is too small to support statistical breakdowns by vessel type or flag state. Encounter classification depends on AIS track quality; vessels with sparse AIS reporting intervals may produce misclassified or degraded compliance deficit values.

Data as of 2026-05-12. Source: colregs_encounters, 45-day window, n=4,433 records; analyst reviews: 0.