Documentation is the Most Valuable Thing You Do

Writing or maintaining documentation is probably the most important things that systems administrators can do. As a technical writer my perspective may be a bit skewed. While Systems Administration for Cyborgs does not address technical writing in any substantial manner, documentation is incredibly important to the work of systems administrators.

It’s feasible that systems administrators consume documentation as a class than any other class of users. Furthermore, given administrator’s distance from the development process and their role as “professional users” of software, systems administrators have the potential to generate very good technical writers. This section provides an overview of how technical writing and documentation works, and how to write high quality documentation as needed.

See also

For a wiki discussion and collection of blog posts that address technical writing, consider

The Challenge of Documentation

Good documentation is both technically complex and simple to read and understand: it both has good logical at all levels of organization and works well for reader who won’t read more than a few paragraphs. While documentation presents complex ideas and documentation resources themselves are reasonably complex works, they must:

  • thoroughly describe how a tool or product operates (i.e. what all of the knobs do,)
  • how to use the product (i.e. tutorials, installation and configuration guides,) and
  • describe possible use and configuration (i.e. use cases, practical guides, case studies.)

Often these goals and requirements conflict with each other, and present substantial challenges to maintaining quality well organized resources.

Providing documentation that focuses on and addresses the diversity of user experience is quite difficult. If you write documentation and you’re not keenly aware of your audience, stop and assess what you’re writing and how you’re approaching your task. If you’re re-reading documentation and you find it difficult to “figure out,” stop and work to understand the question you’re asking and what kind of user you are relative to the intended audience.

A poor sense of audience and disorganized goals are often the prevailing issue with technical manuals and resources, but there are other potential problems that documentation processes face.

  • Developers shouldn’t write documentation.

    Having the developer responsible for the code also responsible for documenting its use often leads to poor documentation. Developers are, rightfully, very interested in how the technology works and how the internals affect the operation. In many cases, however, when users need to know details about the internal operation of a piece of technology, there is a larger problem: either the tool isn’t functioning properly, the use case falls outside of the design of the product, or there is a larger design flaw. More than that even, developers often work on very narrow sections of products and have a hard time figuring how users actually interact with their features.

    Less abstractly, however, it’s a project management problem: documentation is a substantial task and if developers are writing documentation then they’re not writing code. While documentation absolutely essential, dedicating an engineer to writing documentation is often a difficult management decision, which means documentation ends up woefully under-prioritized.

  • Documentation is difficult to maintain, and must be continually updated.

    No documentation is perfect, just as its inevitable that some bugs slip into code or glitches into operational deployments, there are errors and glitches in documentation. Not only does software continue to develop and change, but the interaction of a host of third party tools means that user experiences can change independently of the subject matter itself.

    Once the documentation is complete, keeping it up to date, and making sure that it continues to address the current needs of users (current examples, continues to address the most common support issues,) represents a significant challenge. Not only do development teams and organizations need to dedicate time and resources to maintaining documentation, but it’s crucial (particularly on larger more distributed teams) that information flows into the documentation from changes to the product itself, to feedback from support, and notes quality assurance teams, and from the users themselves.

  • There are no shortcuts to good documentation.

    Engineers are often interested in finding ways to automate some aspect of their documentation responsibility. Generate documentation from the comments of their unit tests, or the documentation strings in their functions. Because documentation address the disconnect between users and code, it’s difficult or impossible to write code that will obviate writing documentation.

    There is a lot of room for automation in documentation: for the most part the methods for publishing and updating documentation are terrible, which makes keeping documentation up to date even more bothersome. It’s also much easier to maintain reference material and make sure that software hasn’t changed if the products you’re documenting can output all error codes, or provide a complete listing of commands, which make some of the more tedious problems of documentation less difficult.

    For instance, it might be nice to have code that produces listings of error codes, options, and APIs to make it trivial to ensure that the software hasn’t changed. Having said that, this kind of automation and testing really only covers a small percentage of cases.

Good documentation makes software and tools more valuable and more durable, and although documentation can be a great expense, in the long term its almost always worthwhile.

Qualities of Superior Documentation

Documentation must be singular. While documentation serves different audiences and has different goals, it’s important that there be only one source of information to avoid fragmenting and potentially confusing users and readers. In pursuit of singluar, high quality documentation that provides values to readers, writers must construct documentation that is: well organized and effectively indexed

The previous section addressed some of the common shortcomings of documentation, while this section address some of the mechanical or practical things that makes “okay” or merely “good” documentation great.


Most of the time users only consult documentation after they’re already using the tool or process, and then only if they have a problem. Technical writers–hopefully–should strive to write documentation that anticipates this use. At the same time its important to write documentation that addresses other key uses cases: beginner’s getting started guides, reference material, etc.

Documentation may have unique usage patterns and audience profiles, depending on the specific content but fundamentally, documentation is no different from any other kind of informational resource. If you’re used to organizing code, or server racks, or supply closets, or even your file system, and you have methods that already work, the chances are these organizational skills will be transferable.

The titles, section heading, and overall presentation of the documentation should make it clear where all the information is within the document. If a the answer of a question, or a particular section seems like it could go in multiple locations, then this might indicate a bug in the organization of the documentation. Having the same piece of information in multiple locations makes the resource harder to maintain, and leads to situations where users find the right answer in the wrong context for the problem they’re trying to solve, which is sub-optimal. Treat these kinds of problems as “bugs” with overall organization or the scope of the current resource.

Most documentation systems have support for easy cross-referencing and linking that makes it easy to achieve this “singularity of information,” but making sure that everything is properly referenced and internally linked can be difficult.

Many writers and developers have opinions about the best way to organize documentation. Taking feedback is important, but its hard to integrate all feedback: particularly when feedback reflects specific and uncommon user experiences. In general well organized documentation has the following characteristics:

  1. Hierarchy is good and adds clarity, but too much hierarchy creates confusion.

    You get two, maybe three, levels of hierarchy in any (textual) system before it gets needlessly complex and difficult to manage. While additional hierarchy makes it easy to have more granular organization, it also increases the chance that pieces of information will become buried in trees that users may never discover.

    Writers tend to approach documentation “from the bottom up,” when writing but users almost always approach it from the “top down.” This difference isn’t bad, and it can lead to very thorough documentation, but it means that users can perceive gaps that don’t really exist. Having less hierarchy minimizes these problems and makes the documentation easier to maintain.

  2. Short, clear inductive paragraphs are (almost) always best.

    Perhaps the hardest aspect of technical writing for people with a writing background, is that you have to write documentation in a way that makes it easy to not read. Like journalistic writing, but unlike every other kind of writing, technical documents must be inductive: present conclusions and state purposes at the very beginning of the resource, chapters, section, and even paragraphs, and then delve into details and explanations.

    This is very counter-intuitive for many writers, and even for technical people who deal with users, [1] but inductive documentation is easy to read (or not read,) provides the answers that users actually need.

[1]Developers, support people, integration folks, and sales engineering types often provide examples as a way of explanation, or want documentation to include too much introduction. The folks who do this kind of work interact must often explain complex topics deductively and the inductive approach doesn’t mesh with the way they must conceptualize technical information.

Search and Indexing

Most of the time, the answer to a technical question is more easily discovered using a Google query. Having good search tools indexing the documentation makes it more useful. Having documentation that Google can index effectively, is even better.

However you provide search, the more important aspect of making documentation user-friendly revolves around making sure that the documentation is useful for people who enter using search tools. This means linking and cross-referencing redundantly, assuming that users will not always read the previous section, and spending more time than you think really possible on the reference materials is often extremely worthwhile.

If your documentation tools can produce indexes of your reference material these kinds of information aggregations can be quite useful for people using documentation to answer questions.

Rich Semantics

This is a build and presentation issue for documentation, but being able to add some level of semantic markup to documentation almost always helps improve clarity and reduce redundancy. In this case, semantic markup, refers to systems that make it possible to annotate text with tags and identifiers that make documentation easier to index and process pragmatically. Good markup in turn makes it easy to index and generate novel views of the documentation and of reference material, which improves the effectiveness of search tools.

Its easy to go overboard on semantic markup and support for semantic markup varies from system to system. The key is to make the annotations unobtrusive for writers and readers. In the end this is largely a documentation production issue, but better documentation tools and processes can make documentation easier.

How to Read Documentation

Documentation should provide “signposts,” that instruct the reader on the internal organization and structure of the text. Notice these and use them to navigate the documentation. If the documentation isn’t signposted, or even if it is, consider the following strategies:

  • Read tables of contents and indexes.

    Tactical reading is a good strategy for finding the answer: Good comprehensive documentation is a marker of great software and systems, at any given moment the answers to your questions are not comprehensive, and having a good overview of the text can help you figure out how to find your answers.

    Also I think technical writers, on the whole, work really hard to build useful indexes of their content. So you should take advantage of them.

  • When in doubt, use search tools to find the answer to your questions. Increasingly, the authors of documentation expect that people will use search tools to find the answers to their question, and this is often the most efficient means of finding the answers to questions.

  • Scan through topical documentation and examples, but focus your attention on the reference materials.

    Once you have a sense of how things work and how to approach problems, most of the answers to your questions will be in the reference material anyway. These section contains information on particular behaviors.

How to Write Documentation

When you write documentation you should:

  • focus on the reference materials,
  • make sure that the organization is clear and that the content is cross referenced, and
  • make sure that you focus on being as clear and forthright as possible.

While it’s unusual that technical writers aren’t surprised by what they when they write documentation, the reading experience should contain no surprises. There are a number of less-obvious things that you can do, as a writer, that may not be obvious.

  • Use structured conventions to make reading easier.

    Lists, tables, and a restrained use of document hierarchy can make your texts easier to read and scan for easy reading.

  • Use short clear sentences.

    Longer complex sentences take a long time to read and are hard for readers to parse. In some cases, complex sentences are neccessary for introducing complex concepts and relationships, avoid these kinds of sentences when possible.

  • Reduce redundancy.

    Documentation should be comprehensive, but not at the expense of being repetitious. Include information once and then reference and link about to it in later sections. Redundant information is hard to update and makes it difficult to maintain consistent information across the resource.

  • Within reason, whitespace increases readability.

    Readers tend to bounce between paragraphs, and if your text is full of 100 word paragraphs, readers will miss important details. It’s possible to go to far, which can have a deleterious effect, but generally this is a good starting point.

  • Always provide context and explain examples.

    Never assume that the code can or will speak for itself, or that you can present a code example without:

    1. explaining what the effect of the sample will be,
    2. where it runs, and
    3. why it has the effect that it has.
  • Clearly mark feature change notification.

    Note “version added,” “deprecation,” “version removed,” and “version changed,” events with related features to prevent confusion and frustration.

  • Cross reference extensively.

    Manage documentation with software that allows you to link to other related or discussed topics. Cross referencing should be explicit and frequent, and should help users find relevant and related information. Cross referencing can help reduce redundancy, which in tern increases clarity.

Work Bottom Up

Resist the temptation to start writing documentation “from the beginning,” both in the sense of beginning to construct a technical resource by writing the introductory material, and by fully explaining and contextualizing processes and concepts before documenting them. Introductory and materials are absolutely necessary for documentation resources, but it’s often better to write them when they can introduce active content rather than when they are merely aspirational.

Similarly, while context is absolutely essential it’s easy to get lost in providing context: too much context leads to redundancy and documentation that’s difficult to read. This is understandable. The writers and maintainers of documentation primarily relate to the context and explanation of the text, while users of the documentation are primarily interested in the processes, reference material, and examples.

The best way to mitigate these problems is to work backwards and construct documentation from the reference materials, procedures, tutorials, and examples and then add a thin layer of connective text that ties the resource together logically. There also comes a point, somewhere in the middle of the documentation process, when it becomes impossible to proceed without a proper introduction: write the introduction then. Introductions are necessary when you have more than a handful of major topics, and you need something to provide additional structure.

Reference material can be daunting to create, not to mention boring, but having a good reference and index makes it possible to build documentation that’s easier to maintain, and reduces overall redundancy by centralizing core information within the resource. Furthermore, writing documentation in a non-linear manner tends to reduce the impulse to build larger scale narratives that make it difficult to use and approach tactically.

Use Inductive Structure

Writing education focuses on deductive forms: you introduce a premise, provide supporting arguments and eventually arrive at a more narrow conclusion. Technical writing, like newspaper-style journalism, prefers an inductive style. In inductive writing, you present a conclusion, and then present evidence and arguments to support that conclusion as needed until you’ve conveyed all necessary information. -30- Present the most important information first with less important information following logically as needed.

While there are exceptions, approach everything inductively: from the overall structure of a resource to the structure of individual articles and sections to the structure of paragraphs and sentences. These organizations make it easier for readers to discern meaning and find the answers to the their questions. You may think of the inductive approach as a way of “making documentation easier to not read,” which may be difficult to stomach if you have a conventional writing background, but for documentation this is exactly the right approach.

Consider Audience

The target audience for documentation defines the shape of the text and the overall organization of the resource. Different classes of users have different kinds of problems with technologies and processes, ask different questions of their documentation, and have different usage patterns based on their level of experience and proficiency.

Good writing and considered organization leads to documentation that is useful to multiple audiences. Indeed, proper indexing, signposts, and tables of contents make documentation more useful to more different kinds of readers by leading them to the most relevant pieces information more quickly.

Nevertheless, different audiences need different levels of detail and will approach their use from different perspectives. These requirements are difficult to reconcile. The question technical writers must ask themselves about audience is not “does this resource contain everything that it must,” but rather “does this resource contain information that it doesn’t need or that will confuse users and make it more difficult for users to answer their questions.”

You can discern part the kind of documentation that users require based on your own experience as a user. The design of the software or solution defines another portion of the documentation requirements. However, the most important information comes from the users themselves in the form of support questions, and product inquiries.

As a result, when users ask the same question on mailing lists, in support tickets, and on chat rooms, figure out what pieces of information will allow users to answer their questions themselves. Then, look at the areas of the software or solution that people are both most interested in and least likely to intuit independently. Use these use cases as the starting point of your documentation, and grow outward from this base.

Documentation needs to be user focused. Unfortunately, in most cases users who are new to a concept or solution do not always know the “right” questions to ask, and more experienced users won’t always know to ask the right questions that beginners need to answer. Using secondary indications of users’ needs, questions, and confusion is a great way to ground documentation projects to ensure that the documentation is relevant, focused, and functional.

Approach Documentation Like Code

Documentation is just another user interface and another way of solving the same “business process” issues that scripts, software, and systems solve. Documentation requires:

  • an architecture (i.e. an outline,)
  • it’s iterative (i.e. revise-able,) it requires code review (i.e. editing,)
  • it’s a massive project that requires some degree of technical scaffolding and support (i.e. tools, frameworks, build systems, etc.,) and
  • requires testing (i.e. quality assurance.)

If you plan, support, and organize your software and systems/infrastructure projects, it is only reasonable to plan and support documentation projects in the same way.

This isn’t to say that you should entirely automate the authorship process (i.e. generating documentation from unit or other tests or from comments within the code), [2] after all at some level people, not machines, are responsible for all code. This isn’t to say that documentation shouldn’t be tooled and that you shouldn’t automate some aspects of the documentation process. Build tools, indexing, and change management [3] are great candidates for automation.

In its “source form,” documentation often resembles the software source: both exist in a collection of plain text files, both have a structure, and both have a build or rendering process that translates the source into a usable product. As a result, most tools that programmers use to manage their work and products are useful for writing documentation, including:

  • version control systems,
  • automated testing frameworks,
  • code review tools,
  • issue tracking databases,
  • text editors, and
  • build automation tools.

The right toolkit for your project depends on many factors, and three are a number of different approaches to documentation management that can improve your ability to manage documentation projects. Documentation tooling, should integrate into the systems that you already use to manage your work on other projects. While some projects manage documentation in the same source control repositories as their core source code, and/or use the same issue-tracking systems as the main projects the goal of “treating documentation like code,” is not to strictly integrate documentation and engineering process. Rather you should strive to reuse the tools you already have experience with. Consider the following cases:

  • Treat issues with documentation like issues with software: file bug reports.

    The average problem with documentation is quite small compared to the average size of most issues with software. As a result the software that most software developers use for issue tracking tends to be too heavyweight for most issues. [4] However, users of documentation–at least for some kinds of systems and processes–may already be familiar with issue trackers systems, which makes them more ideal for providing a venue for collecting feedback.

    Depending on your workflow and other responsibilities it may also make sense to use ticketing systems as a way of managing and batching documentation tasks, or distributing documentation tasks among a team.

    When using an issue tracking system for documentation the following are particularly important for a successful experience.

    • Ensure all tickets are actionable and discrete.

      Before filing an issue on documentation:

      1. ensure that you can envision a solution to your problem: (i.e. “document this feature or process,”) or
      2. that you are expressing a problem with a specific aspect of the documentation (i.e. “I found this confusing,” or “this section is unclear or difficult to find.”)

      Documentation tickets can be pathologically open ended, with subjects like “improve the documentation of these related features” or “clarify documentation of outputs.” These kins of issues are difficult to address and harder to figure out when you’ve actually resolved the issue.

    • Respond to all tickets in a timely way.

      Even if your publication schedule does not feasibly support regular updates in response to tickets, it’s‘ important to respond to and triage tickets as they’re filed. Providing a response to ensure that readers know that their feedback is valuable will pay off in the long run. The initial contact provides the opportunity to assess the issue, figure out it’s severity and importance, and potentially readdress the un-actionable and non-atomic quality of the issue itself.

    • Don’t be afraid to close tickets that are irrelevant.

      While there’s no need to maintain a low ticket count, having too many unsolvable open issues makes it difficult to use the ticketing system to actually track ongoing improvement, enhancement and maintenance of the documentation.

      Often, users and engineers will attempt to add or request that you add information to the documentation that may actually make the resource less useful and more confusing for users. While all requests merit consideration, some requests do not merit action, and it’s acceptable to close tickets or rewrite tickets as needed.

  • Track versions of documentation using source control systems.

    Version control makes the editing and publication (i.e. release) processes manageable. In the same way that version control systems make it feasible for more than one developer to work on a single piece of software at a time, version control allows multiple writers to edit and contribute to documentation at the same time.

    In the last five or ten years, version control has become more advanced and much easier to use, and if you produce documentation and expect that more than one person may ever work on the documentation, then it’s good practice to use version control. Which version control tool you use is not especially important.

  • Manage document generation and publication processes using build systems. (i.e. GNU Make.)

    While build automation systems (i.e. Make/GNU Make/SCons/Ant/Ninja) are all targeted at different use cases, and none are explicitly designed for building documentation, being able to reproduce the publication process for your documentation and thus be able to revise, edit, and improve the documentation on an ongoing basis is crucial. Furthermore, build-like processes are sufficiently regular and unchanging that it makes sense to record the process for in a script or Makefile.

  • Get technical review on all changes: use code review for all [5] changes.

    If nothing else, make sure that someone other than the original author reads documentation before providing it to users or prospective users. The review process should ensure that the documentation is both factually correct and stylistically clear and consistent. Ideally more than one person should read the documentation, because it’s difficult to read a piece of writing for both style and form concurrently. However, some review is always better than no review.

    Make sure that you have review for all changes to the documentation. While you might be able to make some small changes without affecting the meaning of your text, often a collection of incidental changes can affect the meaning or clarity of a document in aggregate.


Generated documentation (i.e. derived from comments, “doc strings,” or similar) is particularly lacking with regards to audience participation. While easy to produces and potentially complete, fails because the resulting documentation is entirely declarative, aimed at no audience, and does little more than summarize the code in English. While this kind of documentation has its utility and place, it’s not useful for users. Any question that such documentation could answer is probably better addressed by reading the code.

While engineers, product architects, and software designers often have all knowledge required to write documentation, their efforts tend to focus on functional aspects if the program, and ignore the practical. In most cases “how something works,” is only relevant insofar as it affects decisions that users must make about deployment and configuration, and the expected behaviors of the product or outcomes of the process.

[3]Because it’s difficult to build documentation automatically, making sure that documentation changes to reflect the changes in the product or process is sometimes difficult.
[4]To be fair, issue tracking is far from a “solved problem,” and issue trackers are universally despised for their poor design, difficult administration, and cumbersome user interface.
[5]Well, maybe not all changes, but most substantive changes are probably worth getting some level of review.


Every piece of information should exist in one and only place. Furthermore, this location should be obvious, and your target audience should be able to look at the organization of your resource and find most pieces of information by browsing the resource and its indexes. In most cases, users will have a different access pattern, but keeping browsability and audience in mind when developing organization leads to very high quality and useful resources.

In most cases, if you find that you must include a piece of information in your documentation that seems as if it could appear in two places, you have uncovered a more important problem. In essence you have the following options:

  • Change the structure of the resource.

    Create a new section, page, or tutorial if you need to add something to cover a use case or interaction path that doesn’t have sufficient space in the existing documentation. Alternatively, you may sometimes merge sections sections to create a more logical organization that incorporates the existing information and the new information.

    At the same time avoid situations where reorganization result in needlessly deep hierarchies of section headers or sub-documents. While structure and hierarchy often make organization more apparent and easier to use, every level of hierarchy caries a cost: only add levels in extreme situations.

  • Remove the potentially confusing piece of information.

    When information “feels” like it fits in multiple locations, you may have an organization that is unclear or too expansive. Rewriting or deleting some of these sections can often make the “one right place” for the new piece of information apparent.

  • Insert cross-references.

    Most documentation systems have some facility for fine-grained links between sections. If a note or piece of information fits in more than one place, consider putting that information in one place and inserting links from all of the other relevant places.

    Often the greatest challenge with cross referencing is ensuring that all of the places that need a link have it. Additionally, the cross refraining approach must remain consistent over time. Formal, recorded, policies help and prevent drift.

  • Duplicate the information.

    For small chunks of information, in some situations, duplicating information may be the only reasonable solution. This is a strategy of last resort, but with the proper cross-references, and notes to improve future maintenance, duplicated data sometimes leads to a more positive user experience.

Sometimes the information leads to an apparent and obvious organization. In other situations the best way to resolve these conflicts may not be clear and you may have to tolerate an non-ideal solution. Above all, value consistency, even if you’ve accepted a compromise on some point. For longer projects, have an explicitly recorded approach for dealing with harder organizational problems. Document maintainability is extremely important.

Consider the following strategies:

  • Separate the “physical” organization from the “logical” organization of the resource.

    Use indexing, tables of contents and file including, URL mapping solutions you can present users with a well organized document, that follows a more simple representation in the editing interface. In many cases, for documentation, the requirements of “physical” representation are much more simple than the “logical” or user facing organization: placing a layer of abstraction between the two can improve both.

    In addition this separation may be helpful providing multiple independent views into the same larger resource, which allows you to address multiple audiences and publication venues with comparatively minimal effort.

  • Be explicit. Record the organizational decisions you make early on for future reference and use inline comments to include pointers to other parallel sections to ease the maintenance burden.

    All documentation requires some ongoing maintenance; however, any particular document or section is at best edited or reviewed once or twice a year. Having “meta” documentation that describes decisions that you made regarding the resource’s organization makes this process easier and more consistent. At the same time, inline comments can also make it easy to understand “why things are the way they are,” and provide helpful reminders of other related changes.

  • Use glossaries and reference material to centralize information in one place and use the “content” of the resource to hold procedural and usage information.

    This only works well for some kinds of documentation, but it’s a generally powerful approach. It’s easy to index and search reference material and because reference materials don’t require any narrative padding, it’s easy for users to get answers from reference material.

    Strong and complete reference material permits procedural and usage documentation to omit background information because of cross referencing. The result is an easier to organize, easier to write, and easier to use documentation resource.

The Value of Documentation

Documentation is valuable as reference and standardization material and because it creates value for organizations and products:

  • Good documentation tends to reduce the volume of support requests and erroneous bug/issue reporting. Additionally, good comprehensive documentation makes responding to the remaining support requests easier.
  • Good documentation gets knowledge out of people’s heads and into a shared format. This increases reliability, because it reduces a dependence on people’s memories and notes, which ma not always be accessible, or may be inconsistent.
  • In absence of a specification, documentation can help define and regulate the user experience as development proceeds.
  • Without documentation users may be unaware of features and behaviors, which reduces their value. If users never learn about or take advantage of new features and functionality, then development resources are essentially wasted.

This doesn’t mean you should write documentation in situations where organizing a system or environment is more practical and efficient, or that standardizing a process with a script or program doesn’t have it’s place. However, for complex problems where users interact with your interfaces or processes, documentation is often the answer. Take pride in documentation, treat it seriously, and the return will be great.