```xml <result> <column_name>latency_ms</column_name> <family>additive</family> <subtype>ordinal_additive</subtype> <confidence>0.92</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 time steps 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 is PATH with sequence axis, so ordinal_additive is plausible. Column name "latency_ms" indicates an intensive physical quantity (duration/time) measured at each sequence step. The values are strictly non-negative with zeros present (13.4%), mean 99.95ms, median 19ms, wide range to 33583ms - characteristic of instantaneous latency measurements. Summing latencies across time steps is meaningless; the natural aggregation is mean/median/percentiles across entities at a fixed step, or distribution statistics along the sequence. This is ordinal_additive rather than intensive_additive because it's a PATH table where the same physical quantity (latency) is measured repeatedly along a sequence, making it additive across entities but only ordinally meaningful along time. </reasoning> <table_type_conflict>false</table_type_conflict> </result> ```