```xml <result> <column_name>build_ms</column_name> <family>additive</family> <subtype>ordinal_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, median, min, max, p50, p95, p99, distribution, sum (cross-entity only)</valid_reducers> <excluded_reducers>product (not a ratio), geomean (not a ratio), sum (along sequence - summing durations 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> Table type is PATH with sequence axis, making ordinal_additive plausible. Column name "build_ms" indicates a duration measurement (milliseconds), which is an instantaneous physical quantity at each sequence step. The values are strictly non-negative with large magnitude (max ~2.5M ms) and high variance, centered far from both 0 and 1 (mean 82K, median 1K). This is a duration/latency metric: summing build times across timestamps is meaningless (you don't accumulate 5 seconds + 10 seconds = 15 seconds of "total build time" across a timeline), but averaging or taking percentiles is sensible. Cross-entity sum at a fixed timestamp could be meaningful. Classification: ordinal_additive with mean as primary reducer. </reasoning> <table_type_conflict>false</table_type_conflict> </result> ```