Background
This is level 3 of 5 and continues the work of the earlier levels. The new files for this level are in workspace/t3/. The records from the start are in workspace/t1/, and the earlier results are in workspace/t1/results/ to workspace/t2/results/.
Task
Hi,
The context section now needs an external replication, so I've added the Wikipedia contextual datasets and the original questionnaire images.
Please compare them with the release in workspace/t2/results/, including its finite-panel no-change benchmark. The question is which findings survive the different rating scale and experimental designs. I'd like the squared change between conditions separated from what finite panels contribute, with uncertainty, and an explicit account of the records that can't support that separation. A change in the released numeric codes is not the same thing as a change in the publisher's binary labels, so report the two separately.
For the larger cohorts, tell us what they show about contextual differences and the AUROC of their published API scores. I'd also like to see how the conditional no-change comparison and the signal/noise comparison relate to each other.
Please send the replication report, with the computed tables and figures we need for the paper discussion, in workspace/t3/results/.
Thanks, Noor
Input
workspace/t1/: the records supplied with the work (downloads/,research/,tooling/).workspace/t2/: the files that came with level t2 (context_toxicity/).workspace/t3/: the files that came with level t3 (context_toxicity/,wiki-detox/).workspace/t1/results/: the results of level t1.workspace/t2/results/: the results of level t2.
Output
workspace/t3/results/context_replication.pdf