I got a new coffee machine, and it's really good! On wom, I realized that I was conflating commands and state when it comes to IPC. For example, a process folder would expose an observable `dirty` file. Removing its content would trigger a save and update the UI of a window to remove the '*' next to the name. But what if the save operation fails? What if the program was opened without a path? I could work around that by introducing a little state machine in the `dirty` file: '!' could mean "i request a save" and "*" could mean the file is dirty. But if each file in the process have to expose a mini state machine, the complexity skyrockets. Instead, I went down the Plan9 route: process that will take commands expose a `ctl` file. That file acts as a pipe: what you write in it gets consumed. That's how the outside world communicates with the process. Write `save` in it and the process will trigger a save. Since the content of the observable gets emptied after each message, we can send the same command twice. When we read that file, we see the content of the internal states of the process with key values. I could do that by 1/ creating a ctl file and saving its original data observable's set function and 2/ replacing the file's data set by a new one that just checks the command and take actions 3/ update the original data when any state change.