There is a persistent and evasive bug in the single-subject model code. This repo is an attempt to root out the issue.
We are using the C source code from Matt's thesis work as the codebase for debugging. The pulsatile package and the new libpulsatile library/package is based off of this same codebase.
Two symptoms exist so far. The first is 'sticky points' in the chains of mean width and the SD of pulse widths for essentially any dataset. The second is random noise/lack of convergence in the posterior chains of pulse locations, specifically in noisier datasets.