```xml <result> <column_name>time_to_insights_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 durations 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: PATH table with sequence axis allows full analysis. Column name "time_to_insights_ms" indicates a duration measurement (milliseconds), which is an intensive property - a latency/duration at each sequence step, not an accumulating total. The values are strictly positive with large magnitude and high variance (mean 15.5M, median 719K, max 693M), centered far from both 0 and 1, consistent with absolute duration measurements. High log_symmetry_gain (5.95) suggests log-normal distribution typical of latency metrics. This is NOT a cumulative time (which would be real_additive with sum meaningful), but rather instantaneous duration measurements where averaging across entities is the natural rollup, not summing. </reasoning> <table_type_conflict>false</table_type_conflict> </result> ```