```xml <result> <column_name>ingest_elapsed_ms</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, p50, p95, p99, distribution</valid_reducers> <excluded_reducers>sum (summing elapsed times 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> Table-type gate: DIMENSION with no sequence axis makes multiplicative highly implausible. Column name "ingest_elapsed_ms" indicates a duration measurement (milliseconds), which is an intensive property - each entity has its own elapsed time. Center-of-mass (mean=834, median=152, max=28719) shows absolute magnitude far from both 0 and 1, with high variance typical of performance metrics. Summing elapsed times across different entities is meaningless (you cannot add "entity A took 100ms" + "entity B took 200ms" = "300ms of what?"); the natural aggregation is mean/median/percentiles to characterize typical performance. This is a classic intensive_additive level. </reasoning> <table_type_conflict>false</table_type_conflict> </result> ```