I provide a community labeling service for the ATProto network mainly focused on Bluesky posts. You can choose to hide or warn certain categories of content. With the labeling data, the account has also provided user lists, moderation lists and feeds of labeled content. It should be clear that not all label categories are considered problematic and that the service provides the user with the choice to see or not to see such labeled content.
Labels are created and applied if the operator feels comfortable with labeling such content.
The labeler service currently consists of me (NunnyBabbit "Aon") and a few other friends who check in once in a while. To label the Bluesky network then and now, it is clear that for additional label categories to be applied, users may volunteer for the labeler service.
Anyone is able to contribute by reporting posts specifying the labeler. For triage and sorting out queues, your help is much appreciated. If interested in volunteering, please DM NunnyBabbit "Aon" on Bluesky (@nunnybabbit.com). With the way Ozone is set up, you should have an account dedicated to moderation, else you might face issues when active blocks are in place.
To label the growing network, it is clear that automation of some sort is required. Computer vision has been deployed to classify media and to either auto-label or report for human review.
Auto-labeling could have negative implications, as mislabels are always undesirable and could cause users to feel discomfort. Auto-labeling should only be deployed if additional post context could also be used to infer a label which even then can still fail, so reports are still created for human review.
The labels applied by the service should be straight forward. However, there are a few exceptions which prevents a label from being applied. For the "Diaper" label specifically, it wouldn't be desirable to label an actual child wearing a diaper.
The labeler will not apply "!hide" or "!warn" system labels to any record or account.
A label category may only be created if at least one operator is comfortable with handling it and if there are no negative implications that may arise. This is apparent when categorizing "NSFW" or otherwise "questionable" content. Label categories should also be strictly applied based on content, not behavior of the user, as determining behavior can often be subjective and isn't well suitable for automation.
A moderation label for urine shouldn't have problematic implications, as it doesn't target protected attributes of someone such as their gender, sexuality, identity, race or body and only targets the presence of urine as defined in the label's description.
A moderation label for pup hoods (or other animalistic masks) is however considered problematic, as it is considered as an extension of identity by how a person may wear one. A moderation category to hide labeled content should not be created, the label should only be used to drive feeds.
Also see: Bluesky Labeler Guidelines
Account labels simply don't need to be applied and are considered problematic, as automation assists in reporting and labeling content, making account labels redundant and only serve to be exclusive and misleading in most cases.
Mislabeling can happen. If you feel a label was applied incorrectly, you may appeal the label by subscribing to the labeler, going to the labeled content or account and clicking on "_ labels have been placed on this account". If you aren't the account owner, you may still create an appeal by "reporting" the content to the labeler and specifying in the report that it was mislabeled, this will still receive the same amount of attention as if an appeal was created. Please note that the labeler is currently not able to directly respond back to appeals regarding their status. As a manual workaround, you may DM NunnyBabbit "Aon" to get a reason for an action taken.
The intention of the labeler is not to discourage, but to promote. Non-negative labels will have an associated curated feed and user lists.
If for whatever reason you don't want to be included in these feeds and user lists, you may DM the labeler account with a request or block the labeler account.
The feeds are not run neutrally. Posts or entire accounts may be excluded from showing in feeds depending on their content or user behavior. You may report posts or accounts to the labeler for feed exclusion if you believe something violates these guidelines.
Reasons for content or account exclusion are described but are not limited to the following list.
Additionally, the following labels from external labelers are used to populate feed exclusions.
Due to the subjective nature of account exclusion, a moderation category is not provided to hide content or accounts. Users are still able to appeal the label.
Labels prefixed by "!" are considered system labels, and cannot be used to hide content, nor prompt a user to appeal them. In the context of this labeler, they are used to control behavior of the labeler's systems when encountering them. The "!classification-forced" label will force automated image classification regardless if content labels were applied.