```xml <result> <column_name>avg_non_null</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 an average 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> Gate prior: DIMENSION table with no sequence axis makes multiplicative highly implausible. Column name "avg_non_null" explicitly indicates this is an average (an intensive level), not an extensive quantity. The values (150-891, mean 229.9, median 156) are absolute levels centered away from both 0 and 1, with no zeros. Summing averages across entities is meaningless; the natural rollup is to re-average them (mean of means). All signals converge on intensive_additive with high confidence. </reasoning> <table_type_conflict>false</table_type_conflict> </result> ```