```xml <result> <column_name>latency_ms</column_name> <family>additive</family> <subtype>intensive_additive</subtype> <confidence>0.95</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, p50, p95, p99, distribution</valid_reducers> <excluded_reducers>sum (summing latencies across entities 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 "latency_ms" indicates a physical intensity (time duration per operation/request), which is an absolute level, not a ratio. The values are strictly positive with large magnitude (mean ~9711ms, max ~40369ms) and centered far from both 0.0 and 1.0, consistent with an intensive measure. Summing latencies across different entities is meaningless; averaging is the natural aggregation. This is intensive_additive with high confidence. </reasoning> <table_type_conflict>false</table_type_conflict> </result> ```