At present, we can optionally use --uncertainty to auto-fail all cases where we hit a pattern in any candidate transcription that indicates transcriber uncertainty, and we always detect certain patterns that might survive auto-reconciliation.
What we should really do is:
1) Study the data to identify a good set of patterns that are used in practice for transcriber uncertainty
2) Identify these patterns ahead of data reconciliation, as a part of the data cleaning, and modify the data to handle them appropriately. (It may be enough to just remove them and let the reconciler do its best with the other options, but that again would require some investigation.)
At present, we can optionally use --uncertainty to auto-fail all cases where we hit a pattern in any candidate transcription that indicates transcriber uncertainty, and we always detect certain patterns that might survive auto-reconciliation.
What we should really do is: 1) Study the data to identify a good set of patterns that are used in practice for transcriber uncertainty 2) Identify these patterns ahead of data reconciliation, as a part of the data cleaning, and modify the data to handle them appropriately. (It may be enough to just remove them and let the reconciler do its best with the other options, but that again would require some investigation.)