Abstract Deciding what to code for in archived online corpora from sociolinguistic recordings is important, but enriching such recordings to include more variable situations is an important prerequisite if these corpora are to serve as a resource for the analysis of situational or stylistic variation, still a neglected variable in sociolinguistic theory and analysis. Accordingly, much of this paper is devoted to illustrating the value of recording speakers in different situations, using examples of Situational (primarily triggered by changes in interlocutors) and Metaphorical (primarily triggered by changes in topic) Style Shifting from the literature. Some of these are Serendipitous, while others are Orchestrated, the result of deliberate efforts by researchers to record speakers with different addressees or in different situations. The richest examples, involving Orchestrated Serendipitous Situational and Metaphorical shifting, come from recent work by Devyani Sharma and Ben Rampton, which is briefly summarized. In my conclusion, I turn to the issue of what to code for in archived sociolinguistic corpora, suggesting that we start with some of the key Hymesian components, like setting, scene, participants and perhaps purposes, key and local norms of interaction. We should of course feel free to draw on Preston's and other frameworks and more recently emphasized elements, like agency, while balancing a concern for completeness and ethnographic richness with a concern for feasibility and practicality.