Great article! Some additional tools I would like to reference:

**"Rule 5: Specify software versions"**
- could be too tedious and not feasible in scenarios where you would like to retroactively reproduce some past environment. On Debian (and NeuroDebian-) systems snapshots of APT repositories allow to "freeze" to a specific date. nd_freeze tool (from neurodebian-freeze in (Neuro)Debian) invocation could be placed as the first command in Dockerfile to run (shipped within all NeuroDebian containers) which would switch to use APT repositories in the state for that date. So it would mimic "real-life" scenarios where we update from state on one date to another, and could be used to reproduce some past environment. E.g. see example: https://github.com/ReproN... produced by neurodocker (see next, which supports nd_freeze)

**"Tools for container generation"** - https://github.com/ReproN... (despite the name it is useful beyond neuro domain) - create Dockerfile or Singularity recipes from a single command line invocation, which would adhere to best practices

**"Use version control"** - DataLad or git-annex directly could be used to store not only recipes but containers themselves. https://github.com/datala... DataLad extension also helps to "register" and use those containers with DataLad. This way you could keep **everything** (code, Dockefile recipes, data files themselves, and container images) under version control

**Rule 7: Mount datasets at run time** - totally agree. But it hinders reproducibility since then outside data resources might change, be unavailable, have different paths on different systems. Have a look at YODA principles to self-contain everything within a "analysis dataset" itself -- input data, containers, etc, so all the paths to be mounted always reside "within" the "analysis dataset". https://github.com/myyoda... . More on that could be found in DataLad handbook: http://handbook.datalad.o... , and here is a sample DataLad dataset with containers for neuroimaging which enforces "total compartmentalization" for singularity containers: https://github.com/ReproN...