```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, sum (cross-entity only)</valid_reducers> <excluded_reducers>product (not a ratio), geomean (not a ratio), sum (along sequence - summing instantaneous latencies across time is meaningless)</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: PATH table with sequence axis allows both multiplicative and ordinal_additive. Column name "latency_ms" is a physical intensity (duration) measured at each sequence step, not a ratio or rate. Center of mass (mean=8228, median=6635) is far from 1.0, ruling out multiplicative. The column represents an instantaneous measurement at each step - summing latencies across timestamps is meaningless (you don't add "10ms at t=1" to "12ms at t=2"), but averaging or taking percentiles is sensible. This is the classic ordinal_additive pattern: a physical quantity on a PATH table where cross-entity aggregation uses sum but along-sequence uses normalized reducers. </reasoning> <table_type_conflict>false</table_type_conflict> </result> ```