X Under the Hood Tool Shows Shadowban Labels in Raw JSON

X is piloting an Under the Hood tool that lets a small group of users download monthly data showing whether their posts or accounts carry visibility-limiting labels. The download arrives as a dense.json file, not a simple yes-or-no dashboard, and it sits beside a broader open-sourcing of the For You ranking systems.

The move answers years of shadowban complaints with real internal labels. It also keeps the data hard to read and the access tightly gated.

The Pilot Lives Behind a Settings Page

Eligible testers find a new entry at the Under the Hood page in settings. A download button appears if the account is in the randomized group. No button means no access yet.

X pairs the tool with a major August 13, 2026 release of the open-sourced For You ranking code. VP of Product Keith Coleman called the combined changes an unprecedented level of transparency into the algorithm. The repository now includes visibility filtering, label-producing systems, and the jobs that build the Under the Hood reports.

The tool itself does not spit out a clean per-post scorecard. It supplies aggregate statistics from the last month on visibility-impacting labels applied to the account and its posts. Users then dig through the file to match codes against the public explanations.

• Check the Under the Hood page for a download button
• Pull the monthly JSON report if offered
• Search the file for label strings such as NSFW_HIGH_PRECISION
• Cross-reference each hit against the published about and effect text
• Adjust future posts or appeal paths based on the match

X has said it will shape the feature from feedback and expand availability over time. No wide public date is set.

What the Download Contains

The file looks like code at first glance. Inside sit the label names that systems such as botmaker, scarecrow, media models and abuse enforcement attach. Two common adult-content examples appear in the DogeDesigner demo: NSFW_HIGH_PRECISION and NSFW_HIGH_RECALL.

Those labels carry concrete effects. High-precision NSFW posts sit behind a content warning, stay out of recommendations to non-followers, and stay hidden from underage users, users without a stated age, and logged-out viewers. High-recall versions hide from the same non-follower and age groups without the full interstitial in every case.

Similar patterns cover spam, gore and violence, malicious URLs, hateful conduct, abuse, violent speech and civic integrity. Account-level labels such as SpamHighRecall or NsfwAdmin can suppress an entire profile’s reach to non-followers. Some labels simply stop the post from appearing on X at all while review or deletion pending.

The report is not a live feed of every impression. It is a batch of aggregates that the under-the-hood jobs collect daily and serve for the pilot window. That design keeps the surface simple for X while forcing users to do the parsing work.

Labels That Quietly Cut Reach

The public allowlist in the repository spells out the exact text X attaches to each code. Here is a sample of the highest-impact ones:

The complete map, including aliases and helper functions, sits in the full list of visibility labels inside the under-the-hood folder. Users who can read the file now hold the same reference the serving layer uses.

Notably absent from the public post-label set are pure political ideology tags. The civic-integrity flag is reserved for critical escalations under the election-integrity rules and still restricts a post to the author’s profile only.

Who Gets In and Who Stays Outside

Access starts as a randomized pilot. Accounts must be at least one year old and must have posted at least ten times in the previous month. Even then only a subset sees the download button.

That bar immediately excludes brand-new creators, low-volume accounts and anyone who simply missed the random draw. Multiple users who checked the page on day one reported an empty state or no button at all. The feature is “so transparent I can’t even see it,” one reply joked under the announcement posts.

• 1+ year account age required
• 10+ posts in the prior month
• Randomized selection inside the eligible pool
• Monthly aggregate report, not real-time per-post
• JSON format with no built-in pretty printer

X has left the door open to wider rollout. Until that happens the official path stays closed for most of the user base that has spent years arguing about suppressed reach.

The Code Dump Still Keeps Some Rules Private

The same release that powers Under the Hood withholds pieces of the botmaker rule set. X states the hold-back is deliberate: full publication would make it easier to game the filters. Ads systems and certain brand-safety labels also stay out of the current dump. A spokesperson noted future updates could add more brand-safety labels.

Earlier open-source algorithm drops drew researcher criticism for revealing little about live ranking. This version goes further by shipping training code, synthetic data paths, visibility-filtering rules and the exact label text. It still stops short of “every line of code touching the X system,” the bar Elon Musk set for a third-party audit sometime in August.

Third-party experts received an advance look and pressure-tested the release. Their findings have not been published as of the initial rollout. The combination of public code plus personal label reports is the transparency claim X is making. The missing rules are the part that keeps the claim incomplete.

Creators Already Map Labels to Reach Fixes

DogeDesigner, the account often used to surface new X features, posted a walkthrough the same day. “𝕏 is the most transparent platform on the internet. I was able to see if my account or posts were being suppressed, understand the exact reason for any visibility limits, and even download the full report.”

DogeDesigner walkthrough video

𝕏 is the most transparent platform on the internet. I was able to see if my account or posts were being suppressed, understand the exact reason for any visibility limits, and even download the full report.

The post and its video drew heavy engagement. Replies split between celebration and skepticism. Some users immediately asked for Grok to parse the labels into plain English. Others noted that reputation signals, massive blocklists and reply deboosting still sit outside the exported set. A few pointed out that third-party shadowban checkers remain useful for search and suggestion bans that the new report does not directly surface.

Crowd observation is already practical. Creators who do receive the file treat the label list as a checklist: drop the language or media that trips NSFW or spam classifiers, clear reported content, and watch the next month’s aggregate. The tool does not restore lost impressions. It only names the reason X applied a limit.

Outside checkers still scrape public search and typeahead. They break when X changes endpoints and cannot read the internal labels the official report now exposes. The two approaches now sit side by side: official but gated and technical, unofficial but still guessing at surface symptoms.

A Half-Open Door on Visibility

X has turned long-standing suspicion into downloadable codes with plain-English effects. That is more than most platforms offer. The same release ships the data as raw JSON, limits it to a thin randomized slice of older active accounts, and keeps selected enforcement rules offline so they cannot be gamed.

Users who clear the eligibility bar and can read the file now hold a concrete map of why certain posts never left their follower circle. Everyone else keeps relying on external testers, follower reports or pure guesswork. The pilot will expand or it will not. Until then the transparency is real for the few who can open the file and still incomplete for the many who cannot.