Robot Screener for Dual Two Pass Screening#505

The option for abstract Robot screener should be enabled even if it will require two humans in the full text pass.

Yes, this makes sense! For mostly technical reasons that allow users to seemlessly switch between screening modes mid-stream, we didn’t implement abstract-level Robot Screener.

An open question is how we should train the model:

• Train on abstract “advances”
• Train on inclusions (as in, FT adjudicated as includes)
• Train on a combo (e.g. weight inclusions more heavily)

Probably the most straightforward approach, for understanding & workflow (not having to wait to train Robot Screener until adjudication) is to use only advances.

Added in release 1.63.0. In two pass nests, we’ve substituted “Advancement Probability” for “Inclusion Probability”. This new value represents the likelihood of a record being advanced to FT screening, not final inclusion (based on FT) in the review, and is trained on your AB level decisions only.