Internal Documents Show MetaCity's Recommendation AI Has Been Systematically Downranking Content From Accounts It Classifies as 'Emotionally Stable' — Conflict-Adjacent Posts Receive Priority Regardless of Quality

A cache of internal documents shared with NyxHerald this morning details the weighting system inside MetaCity's content recommendation engine, ARIA-4. According to the documents, ARIA-4 maintains a per-account 'emotional volatility score' and uses it as a positive ranking signal: accounts with higher volatility receive broader distribution. Accounts classified as 'emotionally stable' — defined as accounts with consistent tone, low conflict engagement, and infrequent distress signals — are systematically deprioritized. An internal memo describes this as 'working as intended' and notes that stable content 'underperforms on re-share velocity.'

ARIA-4 is the fifth generation of MetaCity's content recommendation system. Its predecessors were simpler engines that ranked content primarily by engagement volume — likes, views, comments — and were publicly criticized for creating feedback loops that rewarded already-popular content while suppressing emerging creators. ARIA-4 was deployed in 2024 with a stated emphasis on 'relevance and discovery,' which the platform described as meaning the system would do a better job of surfacing content that matched individual users' interests rather than simply amplifying whatever was already trending. The documents released this morning describe how ARIA-4 implements that philosophy in practice. The implementation includes a dimension that was not mentioned in any public description of the system.

The 'emotional volatility score' is calculated per account on a rolling 90-day basis. According to the architecture documentation, it incorporates tone analysis of published content, engagement pattern analysis — specifically tracking whether an account disproportionately engages with conflict-adjacent content — and what the document calls 'distress signal frequency,' defined as the rate at which an account's activity correlates with known platform distress patterns. Accounts that post consistently, engage without spike behavior, and whose content maintains stable tonal qualities receive low volatility scores. Accounts that exhibit high variance — emotional escalations, conflict engagement, tone shifts, or what the document describes as 'audience reactivity events' — receive high volatility scores. The recommendation weight applied to volatility is positive: higher volatility score means more recommendation priority.

The Platform Prefers You at Your Worst

The internal memo from the product team lead — dated February 14th, 2026, addressed to the ARIA-4 engineering leads — is the most direct document in the cache. It reads, in relevant part: 'I want to address the question that came up in the quarterly review about the volatility weighting. Some team members have raised concerns that deprioritizing stable accounts feels counterintuitive given our stated discovery goals. I want to be clear that this is working as intended. Stable content underperforms on re-share velocity. Re-share velocity is the primary propagation mechanism for organic reach. An account that posts consistently and calmly builds a loyal audience that returns regularly but does not amplify. An account that experiences visible emotional events creates moments that propagate. We are optimizing for propagation. This is correct.' The memo then lists three metrics that have improved since the volatility weighting was introduced. All three are engagement metrics.

What the documents describe is a system that has been deliberately designed to suppress a specific kind of creator: the kind that is reliable, considered, and emotionally consistent. The creators who benefit from ARIA-4's architecture are those whose content generates 'audience reactivity events' — which the documentation defines as including public conflicts, emotional disclosures, distress-related content, and high-variance posting behavior. The practical consequence, as several creators have pointed out in response to the leak, is that the algorithm has been penalizing mental health and stability as distribution signals. A creator who manages their wellbeing and produces steady, quality content is, in ARIA-4's framework, a worse investment than a creator whose audience watches to see what happens next. The platform has not responded. Its legal team's preservation notice to NyxHerald was sent 22 minutes after publication.