Netflix Open-Sources oci-agent, an Actor-Critic Loop for Causal Inference
Netflix open-sourced oci-agent, a human-in-the-loop actor-critic workflow that automates the toil of observational causal inference, from covariate balance checks to sensitivity analysis, while leaving question-framing and verdicts to humans. In a retention case study it caught early-adopter bias a bare LLM missed, cutting a flawed estimate by 75%. It's been orchestrating 100+ analyses per month at Netflix.