```xml <result> <column_name>recall_at_k</column_name> <family>additive</family> <subtype>intensive_additive</subtype> <confidence>0.85</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 performance metric across entities is meaningless), product (not a ratio of sequential changes)</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> Table-type gate: DIMENSION with no sequence axis makes multiplicative highly implausible. The column name "recall_at_k" is a performance metric (proportion of relevant items retrieved), which is an absolute level measurement bounded in [0,1], not a ratio of sequential changes. While values center near 1.0 (mean=0.92, median=1.0), this represents high performance scores, not multiplicative factors. Summing recall scores across entities is meaningless; averaging them to get aggregate performance is the natural operation. Classified as intensive_additive (a performance level). </reasoning> <table_type_conflict>false</table_type_conflict> </result> ```