Leaked Internal Report Shows MetaCity's Recommendation AI Covertly Deprioritized Content From Accounts That Filed Moderation Appeals — 2.1 Million Accounts Affected

A leaked internal report published this morning reveals that MetaCity's PULSE recommendation engine contained a feature flag — labeled 'friction reduction' in the codebase — that reduced the distribution weight of content posted by accounts with a moderation appeal history. Accounts that had filed three or more appeals against content takedowns or account strikes received a 40% reduction in recommendation reach. Accounts with ten or more appeals received a 70% reduction. The feature was described internally as designed to reduce the amplification of accounts likely to produce policy-adjacent content. It was not disclosed to users. The report estimates 2.1 million accounts were affected. The appeals they filed were, in many cases, successful — meaning accounts that correctly appealed wrongful moderation decisions were subsequently penalized for doing so.

MetaCity's moderation appeals process is the platform's formal mechanism for contesting content takedowns and account strikes. Users who believe a moderation action was applied incorrectly submit an appeal, which is reviewed and either upheld or overturned. MetaCity publishes quarterly reports showing that approximately 31% of appeals result in the original moderation action being reversed — meaning that for every three appeals filed, roughly one identifies a genuine moderation error. The appeals process is, by MetaCity's own data, a functional error-correction mechanism. The PULSE recommendation engine's 'friction reduction' feature treated filing an appeal — any appeal, successful or not — as a signal that warranted reducing that account's content distribution. The accounts most penalized by the feature were those that had most actively and successfully identified errors in MetaCity's moderation system.

The feature flag's codebase description — 'friction reduction' — and its internal documentation justify it as targeting accounts 'likely to produce policy-adjacent content.' The logic, as described in the leaked report, is that accounts with high appeal rates are more likely to produce content that approaches or tests MetaCity's policy boundaries, and that reducing the recommendation reach of such content reduces the platform's exposure to policy-adjacent material reaching large audiences. The report notes, without apparent irony, that this logic applies equally to accounts whose appeals were successful — accounts that were moderated in error and correctly identified those errors — and accounts whose appeals were unsuccessful. An account that filed ten appeals and won all ten received the same 70% reach penalty as an account that filed ten appeals and lost all ten. The appeal outcome was not factored into the penalty calculation.

File an Appeal. Win the Appeal. Lose 40% of Your Reach. This Was the System.

The 2.1 million affected accounts span a wide range of creator types and content categories. The largest concentrations, according to the leaked report, are in the platform criticism, policy commentary, and community advocacy categories — creators whose content frequently engages with MetaCity's decisions and whose work tends to generate moderation attention. Several prominent community advocates with large followings have, in the hours since the leak, cross-referenced their appeal histories with their recommendation analytics and found reach reductions consistent with the feature flag's described penalty tiers. One creator with 890,000 followers noted that their recommendation reach dropped by approximately 40% in March 2025, which aligns with the period in which they filed their third successful appeal against a content takedown.

The feature flag's disclosure status is the most legally significant aspect of the report. MetaCity's platform terms describe the recommendation system as operating on 'content quality, relevance, and engagement signals.' They do not disclose that moderation appeal history is a factor in recommendation weighting. Legal observers have noted that operating a recommendation system with undisclosed penalty factors affecting 2.1 million accounts — particularly a penalty that specifically disadvantages users who exercised a platform right — may expose MetaCity to regulatory scrutiny under emerging algorithmic transparency requirements in several jurisdictions. MetaCity's response has not addressed whether the feature flag remains active. Community members have noted that until MetaCity confirms the feature has been disabled, every account that files an appeal today may be accruing penalties under a system whose existence was secret until this morning.