For those chiming in about statistics and causal inference: This paper is about causal event ordering in distributed systems. While both might benefit from some common philosophy, this subject has almost nothing to do with causal inference as studied by Pearl, Spirtes, Robins, and friends. See https://en.m.wikipedia.org/wiki/Causal_consistency

This is super interesting. Tons of people agree that machine learning needs a better grasp of causality. Something underdiscussed is that our data doesn’t generally include the information needed to automate causal inference and discovery in the large.

It’s the reason your A/B testing framework tracks which variables were the randomization units. Most tables don’t have that kind of provenance in a machine readable structured format. After all, a DAG connecting any rows in any tables at your company is a hell of a prospect.

It’s nice to read about causality tracking systems like in the article.

I was hoping this would be a result on graphs representing causal relations that showed some nice properties, such as simplicity (though that would have been pretty obvious).

Instead, it's a good introduction to causal ordering and the kind of techniques used to guarantee causal receiving between distributed nodes. The only thing I can't see is what is "novel" in the paper - the diagrams are all great visualisations, but which one was developed by the author? The trees?

To my knowledge, the author co-authored interval tree clocks as well as dotted version vectors. He also co-authored a lot of the literature on CRDTs.

It is really powerful, simple to draw yes, but it may be complex to reason on and build a network. Have a look at Judea Pearl's work : http://bayes.cs.ucla.edu/jp_home.html

I am failing to see how this work is significantly different or its contribution compared to Lamport's original work.

Really nice article. Pretty revealing.