# Preserve out\-of\-network discovery for non\-adult posts from adult\-labeled creators

[arquieta88-art]

The current visibility architecture appears to allow account-level NSFW labels to suppress out-of-network recommendation eligibility for an author’s posts, even when an individual post itself contains no adult content.

This creates an undesirable coupling between creator identity/history and post-level content eligibility.

I propose introducing a post-level discovery override for policy-compliant, non-adult posts from adult-labeled accounts, while keeping all existing protections for actual adult content.

The goal is not to increase distribution of adult media to users who have not opted into it. The goal is to allow adult creators to participate normally in discovery when they publish content that is independently classified as safe for general recommendation.

Current behavior

The published Under the Hood configuration documents account labels such as:

• NsfwHighRecall
• NsfwHighPrecision
• NsfwNearPerfect
• NsfwAvatarImage
• NsfwBannerImage
• NsfwAdmin

Several of these state that the effect is:

The account’s posts are hidden from recommendations to non-followers.

The published visibility-filtering tests also demonstrate author-level NSFW states that allow in-network serving while dropping the post in TimelineHomeRecommendations.

This means that account-level classification can potentially determine recommendation eligibility before the system meaningfully distinguishes between:

1. an explicit adult post,
2. a suggestive but policy-compliant post,
3. a normal photo,
4. a fitness/lifestyle post,
5. or a completely non-sexual text post.

Why this is a problem

Adult creators are not necessarily adult-content-only creators.

A creator may legitimately publish:

• explicit media,
• non-explicit photography,
• fitness content,
• humor,
• commentary,
• lifestyle posts,
• text-only discussions,
• creator-industry content,
• or other general-audience material.

If every post inherits the same out-of-network restriction solely because the author has an NSFW account label, the system removes the incentive for adult creators to correctly separate and label their content.

A creator who carefully marks explicit media and publishes safe content separately receives little discovery benefit from doing so if both categories are ultimately treated the same for recommendation eligibility.

That is undesirable both for creators and for X.

Product incentive problem

X should ideally encourage adult creators to:

1. correctly identify adult media,
2. use sensitive-content controls consistently,
3. keep explicit and non-explicit content clearly separated,
4. and produce content that can safely participate in broader discovery.

A blanket author-level OON restriction creates the opposite incentive.

If a clean post cannot participate in discovery anyway, there is less algorithmic benefit to careful post-level classification.

A post-aware system would reward correct labeling and improve safety simultaneously.

Proposed behavior

For accounts carrying an NSFW author label:

Adult post

If the individual post receives a post-level adult-content label such as:

• NSFW_HIGH_RECALL
• NSFW_HIGH_PRECISION
• NSFW_TEXT
• NSFW_CARD_IMAGE

or another applicable adult-content classification:

Preserve the existing adult-content visibility restrictions.

No change is requested here.

Non-adult post

If the post:

• has no post-level NSFW classification,
• contains no explicit adult media,
• contains no sexually explicit text classification,
• passes ordinary visibility and safety checks,
• and is otherwise eligible for Home recommendations,

then allow the post to enter out-of-network candidate retrieval/ranking despite the author’s account-level NSFW classification.

The author label may remain available as a feature for risk assessment, but it should not necessarily be a hard Drop condition for independently safe posts.

Possible implementation approach

Instead of:

author_is_nsfw -> OON Drop

use a conditional structure conceptually similar to:

author_is_nsfw && post_is_nsfw -> existing NSFW policy

author_is_nsfw && post_is_clean -> normal recommendation pipeline

Additional conservative safeguards could include:

• requiring high-confidence clean classification,
• excluding accounts with severe enforcement labels,
• limiting eligibility to logged-in adults initially,
• using a rollout experiment,
• monitoring report / block / mute rates,
• and automatically reverting a post to restricted serving if subsequent classifiers or reports identify adult content.

This would preserve safety while removing unnecessary author-level overblocking.

Relationship to Issue #45 / PR #17

Issue #45 already proposes valuable improvements around restriction lifecycle, transparency, TTLs, reassessment, and automated NSFW account labeling.

This proposal is complementary.

Even with perfect account-label expiration and reassessment, there is still a separate architectural question:

Should a valid NSFW author label automatically make every individually non-NSFW post ineligible for out-of-network discovery?

I believe the answer should be no.

Account-level labels can remain useful for safety and classification without functioning as a permanent content-agnostic recommendation ban.

Why this can improve recommendation quality

The recommendation system already predicts viewer-specific probabilities for positive and negative actions.

Allowing clean posts from adult creators into ranking would permit Phoenix to determine whether a particular viewer is likely to engage positively with that specific post rather than rejecting it exclusively because of the author’s historical category.

This preserves personalization.

It also gives adult creators a legitimate route to audience growth through safe content while maintaining strict controls on explicit media.

Suggested experiment

A controlled experiment could compare:

Control

Current author-level NSFW OON restrictions.

Treatment

High-confidence non-NSFW posts from NSFW-labeled authors are allowed into OON ranking.

Measure:

• reports,
• blocks,
• mutes,
• Not Interested,
• follows,
• replies,
• shares,
• dwell,
• overall recommendation satisfaction,
• and adult-content classifier false-negative rate.

If negative safety signals do not materially increase while positive engagement and follows improve, the treatment would indicate that post-level eligibility is more precise than blanket author-level suppression.

Expected benefits

For users

Fewer irrelevant adult recommendations while still allowing safe content from creators they may enjoy.

For adult creators

A legitimate discovery path that does not require abandoning adult content entirely.

For X

Better incentives for accurate sensitive-content labeling, more diverse creator participation, more precise recommendation filtering, and less reliance on coarse account-level suppression.

Principle

Adult-content policy should follow the content whenever technically possible, not permanently define every piece of content produced by the creator.

An adult creator publishing an adult post should receive adult-content treatment.

An adult creator publishing a clean post should have an opportunity to compete as a clean post.

That distinction would make the recommendation system both safer and fairer.

I think a good idea that accounts that post adult content correctly have more visibility than account that systematically abuse the system gray areas it incentivise to label thing right and penalises how doesn't.
I disagree with adult creators having clean post, many times they can abuse normal hot topics to get visibility and lure to adult content, if they want to participate in normal post fairly a separate/new account should be used, other way the system can be exploit.

[arquieta88-art]

I agree with your first point: creators who correctly label adult content should have stronger incentives than creators who systematically try to avoid classification.

Where I disagree is the idea that an adult creator should need a separate account in order to participate in normal discovery.

That assumes that the creator’s identity should determine the classification of every post, rather than the content of the individual post itself.

An adult creator can also be a fitness creator, photographer, comedian, writer, musician, or simply a person sharing everyday life. If a non-adult post is independently classified as safe, policy-compliant, and suitable for recommendations, I don’t think it should automatically become ineligible because other posts from the same author are adult.

The abuse case you describe is real, but I think it can be handled more precisely: the system can detect misleading funnels, repeated policy abuse, deceptive linking behavior, or accounts using clean posts primarily to circumvent adult-content restrictions.

That is different from assuming that every clean post from every adult creator is an exploit.

Requiring a second account also fragments identity, audience, reputation and social graph, and creates an unnecessary product burden for creators who legitimately publish mixed content.

I would rather see X distinguish between:

• a compliant adult creator publishing correctly labeled adult content,
• a clean post from that same creator,
• and a creator intentionally using clean content to evade adult-content enforcement.

Those are three different behaviors and ideally should not receive the same treatment.