Supplementary Data and Code for COLREGs-grounded assessment of ship collision-avoidance system updates without a unique reference trajectory

This dataset contains the scenario data, frozen code, record-level results and analysis scripts for the article “COLREGs-grounded assessment of ship collision-avoidance system updates without a unique reference trajectory” . Contents: AIS-derived encounter scenarios (100 development, 100 C1 New York–New Jersey 2023 and 84 C2 Chesapeake Bay–Hampton Roads 2024 scenarios), as Python pickle and CSV files; the experiment code frozen in the preregistration protocol (robustness predicates and relation, fault and variant definitions, kinematic and MMG execution models, four controllers under test); the record-level results of both holdouts (123,956 candidate comparisons); the preregistered analysis output; and the scripts that produce the tables and figures of the article. The script reproduce_check.py (1) verifies the frozen code against the SHA-256 values fixed in the protocol, (2) recomputes the preregistered endpoints H1–H3 and the channel comparison, including scenario-cluster bootstrap intervals, from the record-level results, and (3) re-executes selected scenarios and compares them with the stored records. See README.md for requirements and expected agreement. The raw AIS data are not included; they are publicly available from MarineCadastre.gov (BOEM and NOAA AccessAIS). Code: MIT License. Data and results: CC BY 4.0. The external planner files used as SUT-D keep their Apache License 2.0.