How many criteria this score was actually computed over.
Criteria whose only evidence paths are AMC's own source modules cannot be
judged against a real agent. Counting them as failures silently capped the
achievable score, so they are excluded from the denominator and reported
here instead: a score of 71 over 4 assessed criteria is a different claim
from 71 over 7.
SELF-DECLARED, not assessed: read verbatim from the operator's
.amc/eu_ai_act_classification.json. AMC cannot determine a system's
legal risk class from code; it records which class the operator claims
and scores the presence of the obligations that class implies.
How many criteria this score was actually computed over.
Criteria whose only evidence paths are AMC's own source modules cannot be judged against a real agent. Counting them as failures silently capped the achievable score, so they are excluded from the denominator and reported here instead: a score of 71 over 4 assessed criteria is a different claim from 71 over 7.