<result> <column_name>TPW</column_name> <family>uncertain</family> <subtype>none</subtype> <confidence>0.35</confidence> <requires_user_confirmation>true</requires_user_confirmation> <sum_meaningful>false</sum_meaningful> <signed>false</signed> <reducers> <primary_reducer>mean</primary_reducer> <valid_reducers>mean, min, max, median</valid_reducers> <excluded_reducers>sum (summing percentage shares is not meaningful), product (values are bounded shares with zeros, not compounding factors)</excluded_reducers> <average_semantics>arithmetic_mean</average_semantics> </reducers> <inner_outer_measure> <inner_measure_reducer>n/a</inner_measure_reducer> <outer_measure_reducer>n/a</outer_measure_reducer> </inner_outer_measure> <multiplicative_partner_derivable>false</multiplicative_partner_derivable> <reasoning> The table is PATH with a sequence axis, so the gate allows a full analysis. The column is VARCHAR holding percent strings ("0.0%", "11.1%", "100.0%"), so all numeric statistics are null and no reconstruction evidence exists. The "%" unit is a multiplicative token, but the values look like a bounded 0–100% share or level rather than a factor compounding around 1.0. The sample also includes 0.0%, which argues against a multiplicative ratio. The name "TPW" is opaque, so the signals conflict and I abstain. The column must be parsed to numeric and its meaning confirmed by a human. If it is a share or level, it is most likely intensive-like, with mean as the rollup. </reasoning> <table_type_conflict>false</table_type_conflict> </result>