Establish the exact effects of intervention-output failure and selection

Establish the exact effects of nonrandom execution failures and cohort selection on the estimates from the matched realized added-text reporting intervention.

Background

The realized added-text budget comparison evaluates natural and targeted summaries under approximately matched additional-text lengths while retaining the original reports. The stricter tolerance analysis excludes five tasks, and the broader analysis includes repeated failures in two planned observations.

Because the retained cohort and successful observations may not be representative of all planned units, the paper states that the exact effects of failure and selection on the reported positive estimates have not been established. This leaves unresolved how much of the observed targeted-versus-natural advantage is attributable to the intervention rather than the analysis population.