When optimizing for long-term maintenance, we have a few choices. I like to bet on monorepo. A monolithic repository is a simple idea. You organize the code of all your services in a single repository. It has a few advantages over using a separate repository for each service.

Reusing code is easy. Once you abstract a coherent unit of code into a module, you can then import it from anywhere.

Continuous integration runs tests against the entire monorepo, so once PR is merged you bump the version of all sub-services and there is no doubt what versions are compatible with each other. Version 1.2 of service A is always compatible with version 1.2 of service B. This is why complex projects with multiple dependencies often use monorepo as well (Babel, React, Angular, Jest). Due to the very same reason, large-scale refactorings are also feasible.

You maintain one third-party dependencies tree. It's too easy, especially with all NPM goodies, to end up with two different versions of the same library and having to sync them across different repositories manually gives me a headache. Having one main package-lock.json is a real time-saver.

Monorepo forces collaboration, it encourages having the same coding style by having a single config for your linter/code formatter/module bundler and so on.

Custom Elements are the lynchpin in the Web Components specifications. They give developers the ability to define their own HTML elements. When coupled with Shadow DOM, Custom Elements should be able to work in any application. But things don't always work seamlessly.

This project runs a suite of tests against each framework to identify interoperability issues, and highlight potential fixes already implemented in other frameworks. If frameworks agree on how they will communicate with Custom Elements, it makes developers' jobs easier; they can author their elements to meet these expectations.

Custom Elements and Shadow DOM don't come with a pre-defined set of best practices. The tests in this project are a best guess as to how things should work, but they're by no means final. This project is also about driving discussion and finding consensus, so don't be afraid to open a GitHub issue to discuss places where the tests could be improved. ✌️