Twitter's massive write-time and nearly-instantaneous Tweet fanout system is well-known. This talk will discuss the architecture and challenges involved in the shift towards read-time determination and collection of Tweets for users, incorporating disparate content relevance and recommendation systems. With additional layers of software business logic composed into the core Twitter experience, this has necessitated the adoption of new contracts between teams and services, a re-examination of our principles about resiliency, and reconciliation between competing forms of content selection.