In psychology we tend to follow the general logic of falsificationism: we separate the 'context of discovery' (how we come up with theories) from the 'context of justification' (how we test them). However, in studying human interaction, separating these two contexts leads to theories that have low ecological validity: they do not generalize well to our real lives outside of the lab. Our proposed improvement is to use key research practices of formal inductive methods such as conversation analysis (CA) during the process of discovering new regularities by first looking at natural data without being prejudiced by theory. This may remind readers of the questionable research practice of HARKing (Hypothesizing After The Results are Known). We argue that using carefully calibrated forms of 'legal HARKing' systematically and transparently in our exploratory research methods, we can achieve more robust and ecologically valid theories.