The document discusses rethinking the use of automatic speech recognition (ASR) software in courts. It proposes that ASR designed for courts should embrace errorful recognition, as normal ASR is not suitable due to the usual presence of errors. It suggests using techniques like caricatured scenarios and actor-network theory to reveal assumptions and understand what elements in the courtroom are obdurate (resistant to change) and plastic (changeable). The goal is to design a new type of ASR that respects obdurate traditions but targets plastic work practices, with errorful recognition integrated into workflows and speech archived for later use rather than transcribed verbatim.