Currently, the transcription checks are handled entirely in Slack. The reactions on the check messages "store" which checks have been handled already and if there were issues in the transcriptions. Bubbles looks through the most recent messages to remind the mods which transcriptions need to be checked.
I propose to move this functionality to Blossom, storing the checks in the DB. This would allow us to generate interesting stats: How many checks have been performed by each mod? What portion of transcriptions are checked? What portion of checked transcriptions have issues? How good is the quality of transcriptions from a specific volunteer?
However, this poses the question how we would structure the review process and how we store the reviews. I think it would make the most sense to have a web interface which shows the currently unreviewed transcriptions and the transcriptions that the mod has reviewed. It could directly show a transcription preview and the MD source along with the transcribed image and the sub's rules. The mod can then tick off the transcription or leave a comment which will be posted by the bot automatically to give feedback to the user.
Bubbles could then check periodically if the user replied to the message or if the transcription has changed and then notify the mod. This would also have the advantage that the mod doesn't need to check periodically if the transcription has been fixed. If the issues have been resolved the mod can approve the transcription.
This system also has several more advantages:
It could be used for the welcoming process as well
The transcription in the database can be updated, so we won't have as many broken transcriptions in there anymore
We could block new users claiming new posts when they have a review pending
Of course there are also a couple of disadvantages:
Probably requires a significant development effort
Needs a lot of API calls to check the pending transcriptions
We need to think of a good model on how we store the reviews in the DB
When new reviews are pending, Blossom could still ping them in Slack and add the reactions automatically when a mod reviewed them, then the mods don't have to check the website looking for the checks.
Currently, the transcription checks are handled entirely in Slack. The reactions on the check messages "store" which checks have been handled already and if there were issues in the transcriptions. Bubbles looks through the most recent messages to remind the mods which transcriptions need to be checked.
I propose to move this functionality to Blossom, storing the checks in the DB. This would allow us to generate interesting stats: How many checks have been performed by each mod? What portion of transcriptions are checked? What portion of checked transcriptions have issues? How good is the quality of transcriptions from a specific volunteer?
However, this poses the question how we would structure the review process and how we store the reviews. I think it would make the most sense to have a web interface which shows the currently unreviewed transcriptions and the transcriptions that the mod has reviewed. It could directly show a transcription preview and the MD source along with the transcribed image and the sub's rules. The mod can then tick off the transcription or leave a comment which will be posted by the bot automatically to give feedback to the user. Bubbles could then check periodically if the user replied to the message or if the transcription has changed and then notify the mod. This would also have the advantage that the mod doesn't need to check periodically if the transcription has been fixed. If the issues have been resolved the mod can approve the transcription.
This system also has several more advantages:
Of course there are also a couple of disadvantages:
When new reviews are pending, Blossom could still ping them in Slack and add the reactions automatically when a mod reviewed them, then the mods don't have to check the website looking for the checks.