I believe that structure and function is correlated. As we cannot formalize function, but can formalize structure, we should clusterize architectural patterns according to their structure in hope that the same clusters are present in the function space - as function and structure are correlated.

However, that depends on the coordinates of the space - with my approach I see clusters of such properties as the ease of debugging, development by multiple teams and performance. But I have not observed other aspects, similar to the GoF classification of patterns. Which may be caused by the subset of the patterns that I use - the architectural patterns which apply system-wide and are quite independent of each other. Or maybe I did not find the correct dimension to dissect the data - this is a seminal work with next to no analysis, if only because I did not have time to dig into the results, as preparing the book takes much effort.

Still there is a proof that the system holds and even self-heals. The Polyglot Persistence cluster was discovered late in the process, when I was writing pattern evolutions. I saw that multiple issues are solved by deploying a secondary database, checked over the internet and fount the pattern name for that. And in a couple of days it grew into a cluster of about 10 patterns, ranging from read-only replicas to external indices, historical data storage, look-aside cache and evet CDN. All of them were not covered by the pattern language, but as soon as one got in, the cluster quickly increased in size. And I was pleasantly surprised to see that the system identified a missing component of itself.