You are a Research Scientist at Meta leading execution for a new Instagram Feed ranking model intended to improve meaningful interactions for users in the US and Brazil. The project involves a cross-functional team of 11 people: 4 research scientists, 3 ML engineers, 1 data scientist, 1 product manager, 1 TPM, and 1 privacy engineer. Leadership wants the model launched before the end of the quarter because Feed engagement has been flat for two consecutive quarters.
A disagreement has emerged two weeks before launch readiness review. Your manager wants to ship the model broadly on the original date because an executive update to the VP of Product already references the launch. A peer ML engineer disagrees and argues the model is not ready: offline gains are strong, but online holdout results show a 0.8% increase in comments and a 1.6% increase in reshares, while also increasing creator complaint rate by 6% and latency by 18 ms. The peer recommends delaying launch by three weeks to retrain with stricter integrity features and optimize serving.
The VP of Product wants an on-time launch tied to quarterly goals. Integrity wants tighter protections against low-quality engagement bait. Infra wants latency to stay within SLO for Instagram Feed ranking. Your manager prioritizes execution credibility and does not want to miss the committed date.
The launch date is in 14 days. You have $120,000 of remaining experiment and labeling budget, no additional headcount, and only 2 weeks of reserved capacity from the Feed serving team. Any model change after week 1 requires rerunning privacy review and launch approval. Brazil must be included in the first launch wave because it is a priority market.