== baseline
curated   agent-worklog          n=  8 distil_act= 100% self_act=  88% savings=  5%
curated   coding                 n=  8 distil_act= 100% self_act= 100% savings=  6%
curated   data-analysis          n=  8 distil_act=  75% self_act= 100% savings=  5%
curated   devops                 n=  8 distil_act= 100% self_act= 100% savings=  7%
curated   finance                n=  8 distil_act= 100% self_act= 100% savings=  7%
curated   ops/sre                n=  8 distil_act= 100% self_act= 100% savings=  6%
curated   research               n=  8 distil_act=  75% self_act= 100% savings=  9%
curated   support                n=  8 distil_act=  88% self_act=  88% savings=  9%
curated   web-research           n=  8 distil_act= 100% self_act= 100% savings= 85%
synthetic api-json               n= 16 distil_act=  44% self_act=  88% savings= 80%
synthetic k8s-incident           n= 16 distil_act=  94% self_act= 100% savings= 39%
synthetic logs                   n= 16 distil_act= 100% self_act= 100% savings= 61%
synthetic metrics                n= 16 distil_act=  75% self_act=  88% savings= 39%
synthetic rag-synthesis          n= 16 distil_act=  31% self_act=  94% savings= 59%
synthetic sql-rows               n= 16 distil_act=  56% self_act=  81% savings= 42%
synthetic stacktrace             n= 16 distil_act=  31% self_act= 100% savings= 49%
synthetic support-chat           n= 16 distil_act= 100% self_act= 100% savings= 58%
== v3
curated   agent-worklog          n=  8 distil_act=  62% self_act=  75% savings=  5%
curated   coding                 n=  8 distil_act= 100% self_act= 100% savings=  6%
curated   data-analysis          n=  8 distil_act=  75% self_act=  88% savings=  5%
curated   devops                 n=  8 distil_act= 100% self_act= 100% savings=  7%
curated   finance                n=  8 distil_act=  88% self_act=  88% savings=  7%
curated   ops/sre                n=  8 distil_act= 100% self_act=  88% savings=  6%
curated   research               n=  8 distil_act=  75% self_act=  88% savings=  9%
curated   support                n=  8 distil_act=  50% self_act= 100% savings=  9%
curated   web-research           n=  8 distil_act= 100% self_act= 100% savings= 85%
synthetic api-json               n= 16 distil_act=  50% self_act=  94% savings= 80%
synthetic k8s-incident           n= 16 distil_act= 100% self_act= 100% savings= 39%
synthetic logs                   n= 16 distil_act=  81% self_act=  88% savings= 61%
synthetic metrics                n= 16 distil_act= 100% self_act= 100% savings= 39%
synthetic rag-synthesis          n= 16 distil_act=  38% self_act= 100% savings= 59%
synthetic sql-rows               n= 16 distil_act=  88% self_act=  94% savings= 42%
synthetic stacktrace             n= 16 distil_act= 100% self_act= 100% savings= 49%
synthetic support-chat           n= 16 distil_act=  38% self_act=  88% savings= 58%
