Writing Good System Overviews

April 1, 2019

Over the years at Nuna, our data platforms have grown from small pipelines which served simple analytics, to large, polyglot pipelines which can help patients receive personalized doctor recommendations. As we’ve scaled, our inline documentation and README’s for specific libraries have been helpful, but haven’t been sufficient to help new hires understand how our whole system fits together. In order to help our “Newnas” get up to speed faster, we’ve written a different type of documentation: a system overview, detailing how the whole system comes together.

What’s a System Overview?

No matter how well you’ve architected your system, at the end of the day, it’s all just text. For someone to be productive in your system, they need a mental model of how the system works in order to translate the text they’re reading into higher level concepts. That’s where system overviews come in.

System overviews are documents designed to create a mental map of how a system works. When done well, they help provide the context new hires need to start delivering code quickly and with confidence. When neglected, or written poorly, system overviews can cause confusion for new members on the team as they work to understand the cause of an issue.

Building these mental maps may seem hard, but it only requires three key steps.

Step 1: Design a block diagram to provide a birds-eye view of the system

The first step in designing a mental map of the system is to create a visual model. Create a block diagram where the major components are blocks and the relationships between them are expressed as arrows. This diagram helps developers build a baseline understanding of the system and begins to demystify, and contextualize, the terms their teammates’ use when talking about their work.

It can be tempting to over-specify the details of this block diagram, but it’s best to strive for brevity and clarity rather than completeness. It’s more helpful to have a concise, rather than cluttered, diagram. You’ll be able to better express the additional details in the next component of the system overview.

Step 2: Write descriptions for each block

The block diagram introduces a rough mental model of how the system interacts, but doesn’t provide the supporting detail explaining how each piece operates. Understanding these details is essential to achieving a greater awareness of how the system functions.

This documentation doesn’t need to be an exhaustive description, but it should answer some key questions about each component. These include:

The answers to these questions should be brief (2 paragraphs is a sweet spot) and allow people to look at any class and start to understand roughly what it does and how it fits into the system. Including these write-ups in your system overview helps others have a better idea of where their bugs may be hiding, or where to incorporate a new feature.

Step 3: Create a concrete example of the system in action

Close your system overview by creating a worked example to show step-by-step how the pieces in your system solve a problem. This last step helps to cement your readers’ understanding of the system, and turn theory into practice.

Good examples contain a few key traits:

  1. They are real: Resist the temptation to create an example filled with names like
    . Names like Foo and Bar are convenient, but they don’t help to build a plausible mental model. A contrived, but relatable, example is much better.
  2. They include real, small code examples: Your system overview should contain code that demonstrates how to solve your example problem. This code should be long enough to help readers understand your descriptions, but shouldn’t go into too much detail.
  3. They touch the whole system: If something appears in your block diagram, it should also appear in your example so your reader can develop a complete mental model.
  4. Keeping system overviews up-to-date

    Once you’ve completed these three steps — designing a block diagram, writing descriptions of each block, and creating a worked example — your system overview is in a good place. That’s not to say that our work is done; we still need to keep your system overview up-to-date as the system changes.

    New users to your system are the best monitors of existing documentation, so have them review it as part of their ramp-up. After they’ve read it, have them explain what they learned and ask questions. Afterwards, revisit the system overview, and add any necessary detail to clarify any pieces that may be out of date.

    Our system overviews are constantly evolving, but they’ve proven valuable in helping to establish a high-level perspective of how our systems work. While they require some time up-front to write, that investment has more than paid off in terms of easier onboarding as our systems become more complex. A system overview is never done, but we’ll keep evolving them as we update our systems and find new ways to improve patients’ quality of care.

    Note: This initially appeared on the Nuna Blog I'm just cross-posting it here

    Discussion, links, and tweets

    Hey! Thanks for reading! If you like what you read and want more, you can follow me on Twitter.