Replication, reproducibility and machine learning - reference request

Are there any papers where machine learning (NLP) was used to assess if papers would be reproducible? I remember something I think it was from a researcher in Harvard but I cant seem to find it.