Table 6.

Distribution of unsafe major error subtypes. Seven error-subtype options were available in the scoring workbook. Each unsafe major error was assigned 1 primary subtype by consensus of the 2 evaluators. Wrong surveillance interval, age-threshold error, and wrong expedited treatment were available as options but received no consensus assignments. Because subtype assignment was made by consensus without an independent reliability assessment, these zero counts should not be interpreted as evidence that the corresponding error types cannot occur.

Error subtype	Total, n (%)	GPT-5.3, n	Gemini 3 Flash, n	DeepSeek V3.2, n	Arm 1/arm 2, n
Undermanagement	65 (39.4)	0	14	51	48/17
Genotype misinterpretation	48 (29.1)	0	13	35	29/19
History neglect	37 (22.4)	8	11	18	26/11
Overmanagement	15 (9.1)	0	0	15	15/0
Wrong surveillance interval	0 (0)	0	0	0	0/0
Age-threshold error	0 (0)	0	0	0	0/0
Wrong expedited treatment	0 (0)	0	0	0	0/0
Total	165 (100)	8	38	119	118/47