Title: Delayed and reapplied sensitive labels suppress all-ages posts
Problem
Sensitive/NSFW labels are sometimes applied hours after posting, including to text-only posts with no adult content. A label can also be reapplied after it was removed following review. In one observed case, an account-level label was applied even though the author had made no new posts for at least an hour.
The open-source repository shows that the system includes:
This creates a serious false-positive risk: an incorrect post-level or account-level classification can suppress unrelated, all-ages posts from recommendations or search.
Scope
This is not a request to stop labeling genuinely adult content. It is a request to stop collateral damage to all-ages work. Current enforcement appears to over-index on recall—catching as much potentially adult content as possible—while under-indexing on precision.
Requested changes
- Judge each post primarily on its own content rather than allowing account history or author-level scores to determine the treatment of unrelated posts.
- Do not reapply a label that was removed following review unless there is new, strong, post-specific evidence.
- Raise and recalibrate pNSFWText and author-level NSFW filtering thresholds to reduce false positives, especially on text-only posts.
- Separate content warnings or interstitials from automatic recommendation and search exclusion.
- Publish false-positive rates and precision/recall metrics, broken down by language, media type, and label.
- Make successful appeal outcomes durable and provide a clear reason and status when a label is reapplied.
Creators cannot operate reliably under a system that changes a post’s or account’s status hours later, flags text-only all-ages work, and provides no durable appeal outcome.
Title: Delayed and reapplied sensitive labels suppress all-ages posts
Problem
Sensitive/NSFW labels are sometimes applied hours after posting, including to text-only posts with no adult content. A label can also be reapplied after it was removed following review. In one observed case, an account-level label was applied even though the author had made no new posts for at least an hour.
The open-source repository shows that the system includes:
This creates a serious false-positive risk: an incorrect post-level or account-level classification can suppress unrelated, all-ages posts from recommendations or search.
Scope
This is not a request to stop labeling genuinely adult content. It is a request to stop collateral damage to all-ages work. Current enforcement appears to over-index on recall—catching as much potentially adult content as possible—while under-indexing on precision.
Requested changes
Creators cannot operate reliably under a system that changes a post’s or account’s status hours later, flags text-only all-ages work, and provides no durable appeal outcome.