SilverTorch retrieval overhaul (Facebook + Instagram Reels)
Meta collapsed the retrieval stage of its recommendation stack into SilverTorch, a single unified neural model that combines the item index, eligibility filter, scoring layer, and user tower into one network. Applied first to Reels ranking across Facebook and Instagram. This is one of the largest architectural changes to Meta's recommendation system in years.
Practical impact: content that used to reach an audience via loose category matches ('interested in wellness → served spa Reel') now needs tighter semantic relevance. Reels that clearly describe what they are in the first 2 seconds (both visually and in caption) are getting more consistent distribution. Reels with vague hooks are losing reach. For behavioral health and healthcare: lead with the specific condition or population you serve within the first frame.