Elusion rate
The share of responsive documents remaining in the pile a review process discarded, estimated by sampling that pile; it measures how much relevant material is being left behind.
First used in We Ran a Blind eDiscovery Review with No Humans in the Loop
Elusion answers the question opposing counsel asks about any review: what did you miss? In the isResponsive methodology, the default target is a one-sided 95 percent upper confidence bound below 5 percent, tested alongside per-request recall and a convergence check.
The published blind review disclosed a failure on exactly this gate. The discarded pile was estimated at 4.8 percent responsive, but the upper bound was 6.68 percent, above the 5 percent target, with misses concentrated in implicitly responsive material such as chats, meeting minutes and executive presentations.
The authors concluded that the result did not support fully automated production at that standard, while recall still cleared the 75 percent defensibility floor at its lower bound. Stating the gate before the run is what made the failure reportable rather than arguable.