
Meta's Instagram Reels team is preparing a 10-week launch of a new ranking improvement intended to increase short-term engagement without hurting creator satisfaction. You are the Engineering Manager for a team of 8 engineers, 1 data scientist, and 1 product manager. The project is high visibility because the VP wants the change included before the next quarterly business review, and several partner teams depend on your launch date.
The Product Director wants a measurable lift in Reels session depth. The Integrity team is concerned the new ranking may amplify borderline content. The Infra team wants to avoid a meaningful increase in online inference cost. The Reels creator team cares most about creator distribution fairness and is pushing for additional diagnostics before launch.
You have 10 weeks total, a fixed budget of $180,000 for experimentation support and labeling, and no additional headcount. The model depends on a feature pipeline owned by a separate ML platform team that can only commit 1 engineer for 25% of their time. The launch must begin with a 5% traffic ramp and cannot increase p95 ranking latency by more than 8 ms or inference cost by more than 4%.