DJs and DJ mixing have been a part of our cultural landscape for decades, representing the art of selecting and transforming recorded media. DJs use techniques like - time-stretching, - jumps or loops, -layering multiple tracks together, - and other effects. DJ mix reverse engineering is the process of breaking down mixes into the tracks used, and analyzing the creative choices made by the DJ. More specifically, we focus on DJ mix transcription. Given a recording of a mix and the recordings of its constituent tracks, our goal is to determine when and how each track was played. We use a matrix-based model of the mixing process. This allows us to capture all temporal transformations and all mixing gain changes in a so-called "activation matrix". We estimate this matrix’s values using Non-negative Matrix Factorization, or NMF. To handle the long durations typical of DJ mixes, we developed a multi-pass algorithm, which performs the NMF at progressively finer time resolutions. The method is promising, allowing precise extraction of mix parameters. But future work is needed to handle a wider range of transformations, such as transposition, compression, or filtering.