<result> <column_name>Accuracy_Rating</column_name> <family>additive</family> <subtype>intensive_additive</subtype> <confidence>0.8</confidence> <requires_user_confirmation>false</requires_user_confirmation> <sum_meaningful>false</sum_meaningful> <signed>false</signed> <reducers> <primary_reducer>mean</primary_reducer> <valid_reducers>mean, min, max, median, distribution</valid_reducers> <excluded_reducers>sum (summing a bounded 2-5 rating level is meaningless), product (not a ratio)</excluded_reducers> <average_semantics>arithmetic_mean</average_semantics> </reducers> <inner_outer_measure> <inner_measure_reducer>min</inner_measure_reducer> <outer_measure_reducer>max</outer_measure_reducer> </inner_outer_measure> <multiplicative_partner_derivable>false</multiplicative_partner_derivable> <reasoning>The table is a DIMENSION with no sequence axis, so the gate rules out multiplicative and ordinal_additive, and no reconstruction signals are available. The column is a bounded ordinal rating from 2 to 5, with mean 3.43 and median 4, centered away from both 0 and 1, with no zeros or negatives and a negative log_symmetry_gain. It is a per-entity score level, not an accumulating quantity, so averaging is the natural rollup and summing is not meaningful. The name has no multiplicative tokens. Confidence is not higher because a rating is not a true physical intensity, but it behaves like a level.</reasoning> <table_type_conflict>false</table_type_conflict> </result>