Hi there.
@Katie - Thanks for you kind comments, and for sharing the link. I'll be happy to answer any questions you might have... Preparing an idealised example for presentation at conference is one thing, doing it in anger with a real-world project can sometimes be slightly more squirrely!
I'm an advocate for using rmarkdown notebooks for analysis. It encourages the analyst to be explicit about assumptions etc. and to "leave a trail" that they can follow later. Too often the "WHY" isn't addressed in code comments... I had a recent discussion with a colleague who was about to embark on some analysis and simulations for clinical trial design using Bayesian inference, and Notebooks seem an ideal way to capture prior choice (what is the prior based on?), model assumptions (where does it predict well, what are the limitations), simulation scenarios (null model, realistic, optimistic cases), and sensitivity analyses.
Writing a fully reproducible, fully automated, fully parameterised report i.e. one that builds a consistent, coherent and correct report from an arbitrary set of data, through an arbitrary model to inferences is a TALL order for rmarkdown reporting. I'm thinking of notebooks as more granular reporting... i.e. results for THIS model...
Happy to discuss further, if that helps.
Mike