(To avoid delays in label application, automatic labeling is currently only applied to images posted by users that have more than a certain number of followers. This follower threshold is also applied for auto-labeling images in quote posts.)
Due to a programming oversight, images in quote posts were not being caught by the auto-labeler. This shortcoming has been rectified: quote posts that include an image without alt text will now be labeled appropriately.
The root cause of this crash has been identified, and should now be resolved. Issues with latency in the labeling service are still under evaluation.
This service only applies labels for posts without alt text, not accounts. If you see an account that repeatedly posts images without suitable alt text, consider asking them to help make their images accessible!
This issue has been, for the moment, resolved: we are still working to determine the cause of the crash, and will post an update if any further developments arise.
As of midnight UTC 2024-03-22, the Bad Accessibility / Alt Text Labeler is currently unable to submit labels. We are investigating the issue.
It turns out that the Bad Accessibility / Alt Text Labeler is labeling posts at over 20 times the rate of the main Bluesky Moderation Service, outpacing what Bluesky was initially anticipating for this feature. Our labels are currently being delayed, to ensure that Bluesky itself remains stable.
The script that runs our autolabeling service is on GitHub:
Automatic labeling for the Bad Accessibility / Alt Text Labeler service - baatl/baatl-autolabeler
Thank you for your patience: all pending moderation requests have been addressed, and the automatic labeling service for images with no/junk alt text is now active. (In just this first half hour it's been running, we've already labeled over 4,000 posts - wow!)
Posts with heavy use of emoji and "leetspeak", while annoying, are not currently on the roadmap to be labeled by this service, so long as they can still technically be understood when read aloud. (Another party may consider writing a bot that could "un-Emojify" and "de-Leet" posts of those types.)