In particular for within session, the default strategy is k-fold validations.
Because of this, the variance in the paired t-test is probably over optimist.
1 suggestion from @aquemy would be to apply Nadeau & Bangio corrected permutation testing: https://proceedings.neurips.cc/paper_files/paper/1999/file/7d12b66d3df6af8d429c1a357d8b9e1a-Paper.pdf.
In particular for within session, the default strategy is k-fold validations.
Because of this, the variance in the paired t-test is probably over optimist.
1 suggestion from @aquemy would be to apply Nadeau & Bangio corrected permutation testing: https://proceedings.neurips.cc/paper_files/paper/1999/file/7d12b66d3df6af8d429c1a357d8b9e1a-Paper.pdf.