Background

Lately I’ve been working on some projects—including a complete reimagining of this very website—and as such I’ve been interacting with Git on a daily basis. At some point I ran into some conflicts (let’s call it infighting) between various code assistants (*glares at Copilot*) and the recurrence of some workspace troubles that I have encountered over the years. So I decided that a backup solution more robust than the *copy appropriate code folders to backup folder* approach I’ve been doing for years was needed.

On a related note, since I pivoted to technical writing full time in 2024; I’ve encountered a lot of folks that have been writing for years but have either been forced into the Docs as Code approach, or just want to get more comfortable with it. I get you, and I think I can help.

Gitting Gud

OK, sorry for the pun, but it’s actually relevant here.

Before I get into the specifics of the backup solution, I wanted to address the elephant in the room: yes, if you’re comfortable with the redundancy that git provides when backing up your [pushed] code, then you don’t need an additional backup solution!

To put that into technical terms, if you’re doing something like this…

git add .
git push origin main

…then you’ve just backed up any tracked files to a remote repository. Sweet!

But. But—and this is where things get messy—there’s a lot that can happen before you push.

I mentioned coding assistants above, and I’ll pick on GitHub Copilot (a tool co-developed by OpenAI and Microsoft subsidiary GitHub) here. Copilot seems to take a different approach to other coding assistants that I’ve tried. I confess I don’t entirely understand the differences, but according to ChatGPT (lol):

“The short version: Copilot operates inside the editor’s text model, while tools like Claude Code and Gemini tend to operate around it. That difference makes Copilot uniquely capable of introducing destructive conflicts when you accept or reject suggestions.”

Furthermore, Copilot can make edits even “before you explicitly accept them” (!).

This alone might be reason enough to look for a local code backup solution, but I’ll list a few more:

• You want to push “meaty” commits that contain a full feature, bugfix, etc. rather than lots of little commits (see also: squashing commits is a pain :p)
• You don’t know what you don’t know, and just want the comfort of being able to manually revert files if something gets borked
• You’re rebasing commits (and not fully comfortable with it)

Existing Local Backup Solutions

It’s worth noting that there are lots of existing backup tools out there. (I, for one, used the free version of FBackup for years). Although I think the propagation of free cloud backup services like Google Drive and Microsoft OneDrive have obviated the need for a lot of folks to run local backups; if you already use a local backup utility then this solution may be overkill.

This Backup Solution

OK, I think we’re finally past the caveats The particulars of my backup solution (I call it “ReactiveBackup”) can be found at https://github.com/jtasse/ReactiveBackup and its README file.

But here’s a quick primer:

ReactiveBackup allows Windows, MacOS, and Linux users to configure and run code backups for 1 to N repositories on your local machine. You can use the tool to run a quick ad hoc backup, or have it run backups automatically (it’ll look for changes in x configured code folders within the last y number of configured minutes).

I hope you like the solution and find it useful. If you find bugs (I am sure there are bugs—after all this thing was the product of a few vibe coding sessions ) please report them at https://github.com/jtasse/ReactiveBackup/issues. Otherwise, you’re welcome to reach out to me with questions/comments.

Thanks!

So I would say Ms. Hao makes her case about OpenAI acting like an empire (and Sam Altman acting like an emperor) pretty well. But she actually goes beyond this, exposing the suspect behaviors of Microsoft (Open AI’s partner/frenemy), Google, Anthropic (who were supposedly the good guys until they weren’t), Meta, and even X/Tesla/pick-your-favorite-Musk-venture.

Hao is so relentless in describing the misbehavior of our beloved tech giants—yes even Google, who she points out twice tried to build water-hungry data centers in South American countries with severe drought—that I actually misremembered the name of the book as Empires of AI (plural) and had to double-check it before writing this review.

Core Values

In a book full of great quotes, the author introduces arguably its most salient of them on page 3:

“…the most successful founders do not set out to create companies. They are on a mission to create something closer to a religion, and at some point it turns out that forming a company is the easiest way to do so.”

—Sam Altman, 2013

This quote immediately resonated with me, as I once worked for a company—I won’t name it, but I’m sure it wouldn’t be hard to figure out which one—that attracted me and hundreds (thousands?) of others on the strength of its religion culture. The company celebrated its core values constantly, and even awarded an “employee of the week” for the employee that best embodied one of the values for the given week. Even earlier in the company’s history, I was told of a crisis where employees remained so devoted to the company’s vision that they temporarily forewent their salaries while the company weathered a challenging cashflow situation.

So what happened to that company? Well, the problem was that they had record profits. *cough*, sorry. The problem was that they had record profits and some overeager executives/private investors, and this led them to make the “tough decision” to lay off dozens of people for the first time in the company’s history. I left the company shortly after the layoffs.

Not so Nonprofit

In Empire of AI, Karen Hao recounts the earliest days of OpenAI, back when it was actually structured as a true nonprofit (technically OpenAI still has a nonprofit component, but as the book describes it the nonprofit arm of OpenAI is little more than decorative these days). She talks about how the founders, including Ilya Sutskever, Elon Musk, and Sam Altman truly believed in the idea of creating AGI (Artificial General Intelligence, or an AI system that can perform just about any task as well as a human) and leveraging it to benefit all of humanity. Someday, they believed, AGI could “solve climate change”, “cure diseases”, and usher in an era where everyone has access to UBI (universal basic income).

But here we are in 2025—ten years after OpenAI was founded and despite hundreds of billions of dollars invested—and what does OpenAI have to show for their efforts?

• No AGI
• So many data centers that are accelerating climate change and, oh yeah, making our electric bills go up
• 0 diseases cured
• No UBI
• ChatGPT is kind of neat?

The Untainted Vision Economy

There is a part of me that wants OpenAI to succeed, to give them a chance to make good on their original promise of spreading around a little prosperity to everyone. But I’m not holding my breath.

In Empire of AI, Karen Hao paints the “Safety” researchers—I’m thinking of “existential safety” in the sense of “trying to make sure AGI or superintelligence doesn’t kill us all—as the good guys. The prime example would be Dario Amodei, who, along with a number of like-minded folks, left OpenAI to found Anthropic amid concerns that OpenAI no longer truly cared about safety. However, Hao points out that over time Anthropic’s view of safety has become harder to distinguish from OpenAI’s.

This brings to mind a quote:

“All of this has happened before, and it will all happen again.”

-initially I recalled this from Battlestar Galactica (2004), but apparently it originated with J.M. Barrie’s Peter Pan (1954)

We cannot count on bold visions like the one that began with OpenAI to magically come to fruition of their own accord. In so many cases like that once-unicorn of a company where I worked, like with Google, and Microsoft, and well, America; the present reality has drifted far afield from the founding ideology. Both from internal pressures to succeed against the “other”, and external ones to grow and innovate at 10x the speed of the competition; entropy seems to push our psychological and economic systems in the wrong direction, practically by default.

But if we follow Karen Hao’s lead. If we call out the tech fiends on their bullshit. If we stop letting the Sam Altmans wriggle out of yet another ridiculous lie/prediction. If we stop the construction of another data center that will take our power, our water, and accelerate climate change. If we champion the usage of smaller, more lightweight AI models and use them instead of the gigantic language models like those powering ChatGPT.

Then we just might be able to choose our own future (instead of AGI choosing it for us).

Attributions

• My copy of the book, https://play.google.com/store/books/details/Empire_of_AI_Dreams_and_Nightmares_in_Sam_Altman_s?id=r6M4EQAAQBAJ&hl=en_US
• Featured image/thumbnail for post, https://helios-i.mashable.com/imagery/articles/04yKMcFP7f9KJhiI7uJHfll/hero-image.fill.size_1248x702.v1747420776.jpg
• Book cover image, https://upload.wikimedia.org/wikipedia/en/3/3f/Empire_of_AI_book_cover.jpg
• “All this has happened before” quote, https://www.thestrangeloop.com/2014/all-of-this-has-happened-before-and-it-will-all-happen-again.html

If you’re like me, the em dash (—) represents an irresistible (lazy?) way to break up sentences. But, frustratingly, it’s not super easy to use in every context. For example, MS Word and Google Docs will automatically convert 2 en dashes (--) into an em dash. But what if you’re in Notepad or a browser?

Windows

Pure keystrokes (if you have a num pad)

In Windows, you can insert an em dash by doing the following:

NOTE: this solution requires that you have a number pad on your keyboard

1. Hold down Alt
2. Type 0151
3. Release Alt

Emoji picker

You can also press Windows key + period to bring up the ’emoji and more’ interface and insert an em dash via the ‘Symbols’ tab.

Espanso

These options are fine, if a bit gross. But if you’re open to installing the espanso app on Mac, Windows, or Linux, there is another way.

Steps

• Download espanso and install it.
• Find out where the match configuration lives in your environment.

NOTE: in Windows, you can find it by navigating to %appdata%/espanso/match and opening base.yaml. I am not sure about where to configure matches for other operating systems, so see the docs for those.

• Within the base.yaml file, configure a match.

NOTE: below is the match config I am using, but you can use whatever you want (just be sure to avoid collisions with other keystrokes where possible)

1. (OPTIONAL but recommended) heading for the match
2. trigger: when these keystrokes are entered while espanso is running they will be replaced by the text defined for the replace value
3. replace: this is the value that will be substituted for the trigger value.

Enjoy!

EDIT 6/30/25: Here’s a copy/paste-friendly version of the config that you can add to your base.yaml file, with a entry for an en dash added as a bonus.

NOTE: apparently yaml (or VS Code reading yaml) doesn’t display em dashes differently from en dashes (as seen in the replace values below), but it does preserve the character information.

If you’d like to verify this yourself, you can copy the characters out of the yaml and paste them somewhere like Notepad or a browser.

  ## Em dash
  - trigger: "---"
    replace: "—"

  ## En dash
  - trigger: "-+"
    replace: "–"

This is part two in a series of posts describing the value and implementation of Docs as Tests.

Intended Audience

My last post was very much focused on the business value of Docs as Tests and [hopefully] building some excitement in the minds of technical writers. But in this one my goal is to provide enough technical detail to whet the appetite of more technical folks (think developers, DocOps, etc.) while still bringing along those of you tech writers (or documentarians, if you prefer) that dared to dream after seeing the demo from last time.

By the way, it’s important I acknowledge that some documentarians write code and some software engineers write docs—I will do my best to meet you where you are!

Technical Approach

I fully admit that trying to create an instructional technical post for such a diverse set of readers is a bit like threading a needle, but I am going to try anyway So I’ll start with some suggestions as you read this post:

• For folks with less technical experience: this might feel a bit like onboarding at your last job (but hopefully more fun!), as I am going to introduce you to the tools, commands, etc. that you’ll need to run the demo from the last post (and more) yourself. If at any point you feel overwhelmed, that’s fine—you are welcome to email me or take a break and come back in an hour, a day, or a week.
• For folks with more technical experience: I’ll try to structure things so you can easily skip over sections you might know (e.g. – installing VS Code).
• For everyone: if you need help or have questions/feedback regarding this post; go ahead and email me or leave a comment below. On the other hand, if your questions/feedback relate to Doc Detective, visit the Doc Detective Discord or LinkedIn page.

Goals for this post

1. Discuss where the testing we’re doing fits with automated software tests
2. Install prerequisites
3. Run tests

Where Are We?

Before getting into the technical stuff I want to paint a picture of where Docs as Tests lives in the software development and documentation space. If you have any familiarity with Continuous Integration / Continuous Delivery (or CI/CD), you’ve probably seen a diagram like the one below.

CI/CD Diagram (click to collapse/expand)

Image source: https://www.primotech.com/ci-cd-pipeline-everything-you-need-to-know/

Basically, the software (and docs) development process goes from left to right . Now for our purposes, we want to focus on the test step.

CI/CD Diagram (test step)

But let’s zoom in a bit.

Types of Tests

A word of caution: this diagram is by no means a complete picture of all types of testing (especially on the software development side). Also, if you think about it too much you’ll see there’s a bit of an apples to oranges comparison going on

It seems fair to say that software testing is more mature as a discipline than Docs as Tests—after all, it has been around for much longer. This is reflected in the diagram above where I show the various flavors of software testing.

But Docs as Tests is catching up fast! When you look at the Venn diagram in the Docs section of the diagram, do you see anything familiar? You might recall (or not, which is fine!) how in the last post I shared an image listing some of the tools we have available for testing various parts of our docs.

Below is that image [slightly altered for correctness/emphasis] again. The takeaway here is that despite Docs As Tests being the new kid (compared to software testing), we already have lots of tools we can use for our tests!

All right, that’s enough mental positioning—let’s get your environment ready to do some testing!

Prerequisites

I am going to list the tools that you’ll need and where to get them. I believe they are all available in Windows, Mac, and Linux environments, but let me know if you run into issues and we’ll figure something out.

Also, I know that people might prefer different tools. For example, the newer Cursor IDE has gained some popularity. Of course, if you are using a different IDE than VS Code, I suspect you already have the tools listed below (or equivalent) installed and can skip down to cloning my repo

OK, let’s get started.

Tools

• Download and install Git.

Don’t be intimidated by the Git installer: it has many screens with different options (or at least the Windows one does), but the defaults ought to be fine in most cases. Like good defaults should be!

• Download and install Visual Studio Code (VS Code) (or your preferred IDE)

• Download and install Node.js

I can vouch for Node.js major version 22 working fine with the tools we will use in these tests, but thankfully you can install multiple versions of Node if you need to use an older version for some reason.

Code

• Clone my docsastests-demo repository

This process will put a copy of my docsastests-demo repo on your machine. I’ll provide steps that should work in any environment.

• Open a command prompt or terminal of some kind
• Navigate to the directory where you want to store the code repository (note that this is totally arbitrary). I believe there are some slight differences in syntax depending on the OS and/or terminal, but let’s suppose you keep (or want to keep) your git projects within the C:/git directory. In this case you would run the command:

cd c:/git

• You’ll know you’re in the right place if you see something like following left of the blinking cursor in the terminal:

c:\git>

• OK, now go ahead and copy the following command and run it in your terminal:

git clone https://github.com/jtasse/docsastests-demo.git

• You should see Cloning into 'docsastests-demo'… and some other messages appear in your terminal. If there are no errors, you can move on to installing dependencies.

Dependencies

Dependencies are software (often referred to as packages) that we can include in a project if we don’t want to write the code for the corresponding functions ourselves. For example, since I do not want to write a tool that can auto-detect assertions in markdown and test them, we are going to install Doc Detective as a dependency.

One nice thing about dependencies is that it is generally quite easy to install them once they are defined in your project.

Further reading: if you’d like to see the dependencies for the docsastests-demo repo (or any other node-based project), open the package.json file at the root of the repo.

• OK, now open a terminal (or use an existing one) and navigate to the root of the now-cloned docsastests-demo repo. For example, if you put it in the location I used above, the command would be:

cd C:\git\docsastests-demo

Tip: modern IDEs like VS Code tend to have built-in terminal functionality. Although it’s not required to use it, you might find that it cuts down on context-switching and helps you ensure you are in the right place (when running commands that are directory-sensitive)

• Now run the following command:

npm install

• You should see some activity in your terminal. This command can take a long time to run in bigger projects, but we only have a few packages so it shouldn’t take more than a couple minutes.

NOTE: after the package install is complete you might see a message in your output that looks like this:

19 vulnerabilities (6 moderate, 13 high) 

Unless you plan on creating and releasing production software from your current project/code, these vulnerabilities won’t affect you.

Structure of the docsastests-demo project

We’ve installed all the stuff! However, before we jump into testing a quick tour of the project is in order.

You can think of docsastests-demo as a Frankenstein application that has one purpose—helping you learning how Docs as Test-ing (ugh, I know) works. After you go through the demo you are free to copy it, extend it, or delete it as you see fit. But for now, here’s a reference image and corresponding docs for how things are laid out:

NOTE: the content of this project workspace screenshot will become out of date almost immediately, as some folders (like the test results) will update with every test. However, I will try not to change the structure of the project more than is necessary

1. product: folder containing the files for the web application we are testing
2. readthedocs: folder containing our Read the Docs implementation
3. source: this is the folder where I am storing markdown files to be tested (we will use the highlighted account-setup.md file for our first test)
4. results: this is the folder where Doc Detective will store test results (you can change this in .doc-detective.json)
5. .markdown-doctest-setup.js: contains required setup code for markdown-doctest.
6. .doc-detective.json: the configuration file for Doc Detective (see the Doc Detective config docs for more info)
7. package.json: this file manages packages for the project (see the npm docs for more info)

Running the tests

UI Test (Doc Detective)

In this test we are going to use Doc Detective to run the test described in the demo portion of the previous blog post. After confirming that the test fails, we are going to make the required change in our docs and run the tests again to verify the issue is resolved.

• Begin by opening the account-setup.md file, which lives at the path readthedocs/docs/source/account-setup.md. (This file is also highlighted in my workspace screenshot above).

NOTE: paths can be a bit confusing depending on the use case, but I usually mean “path if starting from the workspace root”. If I mean something else, I will try to call it out.

• OK, so you should see markdown match what I have here:

(Optional – VS Code Only) see detected Doc Detective tests before they run

• I didn’t want to lengthen the critical path of this post, but if you have a minute to spare I think you’ll find the Doc Detective extension useful. To get it, do the following:

1. Click the extensions icon on the left sidebar (it looks like a block being added to 3 existing blocks)
2. Search for Doc Detective
3. Install the extension via the pane that appears (not shown)

• Once the installation is complete, you can access the Doc Detective extension by clicking its icon in the sidebar

• You should now see the tests that Doc Detective found in the account-setup.md file.

NOTE: if you do not see any content, go back and make sure you have account-setup.md open

• Before we can test our docs against our product, we need to bring the product page (located at product/index.html) online. To do this, open a terminal and run the command:

node server.js

• You should get a confirmation in the terminal that the server is running:

NOTE about terminals: when you run a process like a Node server in a terminal, the process essentially takes that terminal over. So if you want to run other commands without interrupting the server, you need a new terminal instance.

• First, open a new terminal instance.

(Optional) VS Code Terminal Management

ANOTHER note about terminals: you could manually run a new instance of Git Bash, PowerShell, etc. outside of your IDE and cd to the project directory; but it’s worth mentioning that VS Code does have a nice menu for adding and managing terminals.

• Now, from the new terminal (not the one running the Node server), execute the following command:

npx doc-detective --input readthedocs/docs/source/account-setup.md

NOTE: npx is a Node command that lets you run…other commands. Further reading in the npm docs.

• This test might take a couple minutes, and you’ll see a lot of activity in the terminal. Keep waiting until you see the following:

1. Indicates where the test results for this run have been saved.
2. Shows that the terminal is ready to accept new commands.

• At this point we could just scroll up in the terminal to see the test results, but because of all the output in the terminal I find it’s easy to get lost there. Instead, let’s go ahead and open the results file directly.

Tip: even within the results file, there is still a lot of content! To jump to the results we care about the most, do a find (Ctrl + F) in the file and search for:

"steps": [

OK, we made it to our test results! But rather than relying on screenshots, let’s review each JSON result one block at a time…

NOTE: if you see any discrepancies in your test results compared to mine, feel free to reach out to me.

Reviewing Test Results One by One

NOTE: if you installed the optional Doc Detective extension in VS Code, the structure of these results should look familiar

Let’s start with the first block:

Style note: apologies for the white-themed code blocks appearing within an otherwise-dark theme (JSON does look a little better in this block style than a bare-bones dark code block). If anyone knows how to change this behavior in the WordPress block editor I would love to hear it

                {
                  "result": "PASS",
                  "resultDescription": "Opened URL.",
                  "goTo": {
                    "url": "http://localhost:3000/"
                  },
                  "stepId": "67f92b95-bae1-4c15-b9b8-4d5f3c4030b6"
                }

Nice! With this test, Doc Detective:

1. Saw the URL in the first step of our account-setup.md file
2. Attempted to access the URL and succeeded
3. Reported a “PASS” result for this test

Let’s look at the next test:

{
                  "result": "PASS",
                  "resultDescription": "Returned 200",
                  "checkLink": {
                    "url": "http://localhost:3000/"
                  },
                  "stepId": "9681b914-1cb4-42d2-8e9e-0af1d21602f6"
                }

If you’re not familiar with HTTP response codes, a 200 means that the request has succeeded. To me, the previous test result (i.e. – successful navigation to a web page) seemingly renders this test unnecessary; but there could be a reason Doc Detective runs this test separately, and it is definitely not doing any harm!

Next!

                {
                  "outputs": {
                    "element": {
                      "text": "Email"
                    }
                  },
                  "result": "PASS",
                  "resultDescription": "Found an element matching selector. Found element by text.",
                  "find": "Email",
                  "stepId": "79fe9fb3-0843-4711-94fa-bc757493c9e5"
                }

This is an example of Doc Detective being quite clever. Here’s what it did:

Let’s start using DD as shorthand for Doc Detective, which delivers (at least) a double-dose of testin’

1. DD saw that email in account-setup.md was wrapped in double-asterisks (so **Email**).
2. DD does not think (as far as I know?), but if it did, it made a note of the above and thought “I’m going to go and look for an email field on the product page” mentioned in step 1.
3. DD determined that the product page *indeed contained an email field, and reported that it “Found an element matching selector” (i.e. – Email)
4. Recorded a PASS result for this test

*CAUTION: I wanted to call out here that Doc Detective appears to require a field label (in this case, <label for="email">Email</label>) in order to pick up Email or other fields. If you are testing a web application and its syntax does not align with this approach then you might not be able to use DD autodetect in the relevant areas.

Next test.

{
                  "outputs": {
                    "element": {
                      "text": "New Password"
                    }
                  },
                  "result": "PASS",
                  "resultDescription": "Found an element matching selector. Found element by text.",
                  "find": "New Password",
                  "stepId": "7b331142-38da-46f3-b9ef-feef72202593"
                }

This test was very similar to the test for the Email field, with DD looking for a New Password field instead. It passed once DD found the following label in index.html: <label for="password">New Password</label>

All right, we’ve made it to the last test of this particular demo!

                {
                  "result": "FAIL",
                  "resultDescription": "No elements matched selector or text.",
                  "click": "Create Account",
                  "stepId": "d942f059-bb0c-4f6a-862f-d948d976538c"
                },

In this test, DD searched for an element with the text “Create Account”, but couldn’t find it. So the test failed.

In a real world example, this might manifest as the Development/Engineering team changing the text of a button without informing the Docs team that the corresponding documentation needed to be updated. But we have tests now, and we just caught it before the release!

NOTE: I do see another "Create Account not found" failed test that appears identical to the one above. I am not sure why there is another one, but I am going to ignore it for now and move on.

Making & Verifying the Fix

OK, so Doc Detective found a discrepancy between the docs and the product. Awesome! But what happens next?

While it’s possible that the change in the product was a mistake, it’s probably more likely that the Product team wanted to change the button language to what it is now (i.e. – Sign Up instead of Create Account) and they just forgot to tell the Docs team. (I mean hopefully they forgot and it’s no all out war between departments! ).

Accordingly, it would probably be a good idea for your Docs team to reach out to Product to confirm that this was indeed a misunderstanding. Let’s assume that it was, in which case there is no further action required by the Product or Engineering teams for this particular case.

Obviously if the Docs team is implementing Docs as Tests, at some point one would expect a dialogue between the Docs team and Product, Engineering, etc. That is outside the scope of this (already long) post, but Manny Silva’s Docs as Tests book does a good job addressing it!

But now that we know what the “blessed” language from Product looks like, let’s go ahead and update our docs.

• First, open readthedocs/docs/source/account-setup.md in your IDE.
• Now, on line 6 where it reads: 4. Click **Create Account**…

• …change it so it reads 4. Click **Sign Up**. (Be sure to save your changes).

• Now go ahead and run your tests again:

npx doc-detective --input readthedocs/docs/source/account-setup.md

• When the test run finishes, you should see something that makes you very happy: all items passed!

But just to be sure, let’s take at the formerly failing test which is now passing:

                {
                  "outputs": {
                    "element": {
                      "text": "Sign Up"
                    }
                  },
                  "result": "PASS",
                  "resultDescription": "Found an element matching selector. Found element by text.",
                  "find": "Sign Up",
                  "stepId": "933314a7-0851-4319-ad00-355b024f9390"
                }

Bravo—you just resolved a discrepancy between the docs and the product before it ever got out the door!

API Test (markdown-doctest)

Up until now, our testing has been all about the UI. But I wanted to conclude this post by dipping into another area, and why not API testing?

In this test we’ll use markdown-doctest to find an issue with one of our code snippets. As with our previous test, we’ll fix the issue (this time in our code snippet) and verify our test passes.

• All right, since we don’t need a local web server for this test, you can (optionally) stop the Node server you ran in the previous test by opening the terminal where it’s running and pressing CTRL + C.
• Now go ahead and open the file readthedocs/docs/source/js.md in your IDE.

1. Remember that this is a markdown file (even though it is comprised mostly of JavaScript)
2. In markdown, 3 backticks (‘`’) followed by a language identifier begins a code block (or code snippet).
3. This is the most important line for our testing. Keep an eye on that URL

Cool. We already did our setup earlier, so let’s go ahead and run the test!

npx markdown-doctest

The output here is a bit misleading on account of some combination of A) the JavaScript I told Copilot to write for our snippet and B) the expected usage for markdown-doctest. But that’s OK, as we can still get a meaningful result. As I’ve highlighted below, the test reported that it encountered an HTTP response code of 404 (not found) when trying to do a fetch from the endpoint at https://jsonplaceholder.typicode.com/todoz/1.

• Wait a minute, what’s a TODOz list? Let’s go back to our readthedocs/docs/source/js.md file and fix the code snippet.
• Change the URL on line 7 from https://jsonplaceholder.typicode.com/todoz/1 to https://jsonplaceholder.typicode.com/todos/1 and save your changes.
• Your js.md file should now look like this:

• Now run the test again

npx markdown-doctest

Look, no 404! Essentially, we just verified that after fixing a typo in one of our code snippets, the test suite is now able to reach the correct endpoint without errors. And because we caught it before the docs were released, we didn’t have to hear about the developers who called Support asking why their TODOz apps weren’t working.

Great job!

What now?

Since you’ve now tested a UI and API using different tools, you can officially say you have Docs as Tests experience in those areas And seriously, give yourself credit for all of the work it took to get here. Last I checked, the widget that indicates “reading time” for this post was at 20 minutes, but I suspect it took you at least 2-3 times make it through all of that setup and testing (probably more for the less technical folks ).

But obviously it doesn’t end here. For my part, I’m going to move on to making a video that covers this and the last post. I know some people vibe with learning from videos more than reading, and I also have a large video-shaped hole in my portfolio

Code Coverage

After that, there is a topic that I think deserves some attention: code coverage. You’ve probably noticed that I have a strong preference for running tests that are auto-detected using tools like Doc Detective. Aside from the fact that these tests have high return on investment, they also scale with your docs automatically.

However, if code coverage of documentation-based tests is taken into account, it becomes easier for us to have confidence that we are “covering” all of our products with these tests. And, crucially, this can help us prevent our hand-crafted tests from becoming out of sync with our docs.

That’s it for now. Follow me on LinkedIn if you’d like to be notified when I publish a video or another blog post.

Cheers.

Docs as Tests: the series! OK the name needs work, but this is my first post in what I expect to be a series of [at least 2] Docs as Tests posts (and probably some videos as well)

Table of Contents

1. The Book
2. The Promise
3. The Goods
4. And You Showed Me the Tools Why?
5. What Success Looks Like
6. Demo
7. So what?
8. And from the devs come testing riches
9. What’s Next

EDIT 6/30/25: I have created a companion video that illustrates the points of this post (and more). Just want to see a demo of Docs as Tests in action? Use this link to jump straight there.

The Book

Welcome! If you’re here, you probably heard about this phenomenon called “Docs as Tests”, which Manny Silva coined and then went on to write the Docs as Tests book.

First off, let me be clear that I am not affiliated with the Docs as Tests book, but as a technical writer I read it and I thought it was really cool and definitely worth reading. (I can vouch for the Kindle version being well put together, but there is a paperback version as well—either way, it’s worth your time!).

DISCLAIMER: I will try to make it extra clear where possible, but you can basically assume everything I am going to say in this post was either pulled directly from the Docs as Tests book or heavily inspired by it.

The Promise

After reading the Docs as Tests book, I am confident that the Docs as Tests discipline can solve many business problems related to documentation. In the book, Manny goes into great detail how this bit of tech can help you test with docs for that part of the stack.

The Goods

However, if you’re not a software engineer (or perhaps even if you are, as I was!), you might be intimidated by the amount of tools mentioned in the book. Some examples:

NOTES:
* I added Pytest because I have used it for integration testing
** Doc detective is kind of a unicorn in that it seems to be capable of testing in multiple areas

I did my best to break the tools down by test surface area, but I suspect some of these tools are more versatile (i.e. – they can test more areas) than the graphic would suggest.

And You Showed Me the Tools Why?

Oh, right I’m trying to make you more comfortable

First, before going anywhere with Docs as Tests, one helpful strategy is to redirect your overwhelm into curiosity. Another way to put this is to acknowledge you are going to have “What the heck is that?” moments, and focus on the stuff that you can understand.

Second, if there’s one thing I learned as a developer it’s that scope matters. A lot. Don’t bite off more than you can chew!

In the book, Manny suggests (and I agree completely) that you need to decide which areas (e.g. – UI, API, etc.) that you want to begin improving with documentation testing. (You might have multiple areas that can be improved, but you should start with a small win and go from there).

What Success Looks Like

You may have already clocked it based on the tooling discussion above, but one frustrating thing about Docs as Tests is that a successful testing implementation for Company A might look very different from Company B’s implementation. In other words, it depends on your environment and answers to questions like:

• Do your docs describe the behavior of a UI, an API, or both?
• Do you have code samples that developers can run?

NOTE: this isn’t a complete list of guiding questions, but hopefully these are enough enough to get your docs testing gears turning

Demo

Before going any further, I wanted to mention that Manny Silva recently showed off v3 of his Doc Detective tool in this video, which includes some examples like the one I’m about to show.

Although there is often a place for a deep dive showing how something works (stay tuned for more Docs as Tests content here), I also think there’s value in distilling one potential approach down to a few screenshots. So let’s begin!

First, consider the sign-up page below for a product:

Now suppose that we have the following documentation (in markdown) instructing end users how to sign up using this page:

1. Navigate to [Product Home](https://jtasse.github.io/docsastests-demo/index.html)

2. Enter your address in the **Email** field.

3. Enter a password in the **New Password** field.

4. Click **Create Account**.

Have you spotted the problem yet?

No worries if not. Let’s see what Doc Detective can find.

"result": "FAIL",
"resultDescription": 
     "No elements matched selector or text.",    
     "find": "Create Account",

Oh no, it looks like that blasted development team has done it again! As the above Doc Detective test result shows, the test failed because the text of the button on the product page no longer matches the documentation.

So what?

Isn’t this cool?! OK, I hear you, failing tests are a bummer. But think about what this means (again, just repeating the logic from the Docs as Tests book):

You can use tools like Doc Detective to catch discrepancies between your product and docs before the product even releases!

And from the devs come testing riches

DISCLAIMER: below is probably an oversimplification of how software testing works (and how messy it can be), but hopefully it communicates the power of what documentarians can now access

Let me try to put the above revelation into a different context. I worked as a software developer for about 7 years. For probably every product I worked on, myself and other devs would write tests that would run every time a new change is introduced to the software.

If a test failed, it meant that a developer had made a change that didn’t play nice with some existing functionality. They broke something.

I don’t recall documentarians/technical writers ever having access to anything like that. Sure there are lots of tools to check spelling, grammar, style; and those are still important.

But now technical writers armed with tests like the one I demoed above can say:

* you might want to say it a little nicer than that, as developers are people just trying to get their stuff done, too (see: the Agile Prime Directive)

What’s Next

If you’re feeling brave, you can clone the GitHub repository that I used for the demo in this post.

But fear not if you stopped learning about cloning with Multiplicity (Michael Keaton’s best work?). I intend to do a deeper dive into how Doc Detective (and maybe other testing tools) work in my next post.

I also plan produce a related video or two, if that’s more your speed.

Until then, happy testing.