X will now tell you that you were labelled, just not what it cost you

X is testing a settings page called "under the hood" that lets a small group of users download account data and look for the "visibility-limiting labels" applied to their posts ^[1][2]. It shipped alongside a release that open-sources more of the ranking code behind the "for you" timeline, which X's VP of Product Keith Coleman described as "an unprecedented level of transparency into the X algorithm" ^[3][4].

The detail that decides how useful this is sits in X's own GitHub notes: the export does not give per-post statistics, and instead shows aggregate statistics from the last month about visibility-impacting labels on an account and its posts ^[5]. That is a specific and consequential design choice. A creator who watched a post underperform cannot take this file and establish that the post underperformed because of a label, because the file does not join labels to per-post outcomes ^[1]. Anything older than the trailing month is outside the window entirely ^[2]. The most common complaint on the platform, that a particular post was throttled, remains unfalsifiable in either direction.

The delivery format compounds it. In an example video shared by the account @cb_doge, the download arrives as a .json file that initially reads as a wall of code ^[6]. Read closely and it is legible: two posts on that account carried two separate NSFW labels, which hid them from non-followers and underage users ^[7]. So the mechanism is disclosed. The magnitude is not.

The code release is more informative than X's earlier algorithm dumps, which Engadget notes revealed very little about how the platform actually works ^[8]. The repo enumerates labels for NSFW content, spam, violence, impersonation and hateful conduct, plus narrower ones such as a "for emergency use only" incident-response label with a custom notice ^[9][10]. Engadget reports no labels tied to political content, though a "civic integrity" flag restricts a post to its author's profile only, applied to posts detected in response to a user report as violating X's Civic Integrity policy, which the company says is reserved for critical escalations ^[11].

Two categories of material are still held back: information X says could be used to game the system, and the code covering ads and other systems outside the "for you" timeline ^[14][15][3]. That is short of Elon Musk's recent statement that every line of code touching the X system would be open source and third-party audited sometime in August ^[13]. X says unnamed third-party recommender system experts previewed and pressure-tested the release, without disclosing how they were chosen or whether their findings will be published ^[12]. A spokesperson told Engadget that future updates could add labels related to brand safety ^[16].

What to watch: whether the export ever gains per-post attribution or a longer history, whether the third-party reviewers' findings surface, and whether the promised brand-safety labels arrive before the ads code does ^{[5][12][16][15]}.