We have never seen a 10/10 score on our tests.

We're not even sure it's possible. The best we've seen is an 8/10.

Most coding tests are simply too easy. Even ChatGPT can pass many. The top candidates all get near-perfect scores and can't be compared. That's why we make the hardest software engineering challenges in the world. This allows you to differentiate between a pretty talented dev and an extremely effective one.

Candidates we've screened have worked at companies and nonprofits that have gone on to raise hundreds of millions of dollars. Customers have told us that they replaced their week-long trials with our 6-hour trials and got more signal. They were also able to screen more candidates and have less guilt, because candidates don't need to quit their jobs or take vacation to take a 6-hour test.

The tests are centered around completing real-world tasks, not trivia or puzzles. The clever dev can implement every sorting algorithm in C. The wise dev uses the standard library. Hire the wise one.

Our tests are designed to be taken open-book. Candidates are free to use chatgpt, github, and anything else they wish. There is no checkbox saying "I pinky promise not to google the answer".

We also buy up all the coding test leaks we can find. Many of hackerrank's tests and codesignal's tests have leaked solutions. These leaks make their way into AI training data. When we find our own tests in the leaks, we replace those tests immediately. We also have a dedicated grader for each specific task, so if two solutions look similar, then they'll recognize it.

First 10 minutes, then 6 hours

It only takes 10 minutes to see if someone has basic skills in a domain. We offer 10-minute tests for initial screens.

It is harder to tell if someone knows how to make a plan and execute it. An hour is not long enough. Two hours isn't either. A month would be nice, but you can get a good idea in one day. So we also make tests that take 6 hours. Of course, we recommend that you compensate candidates for the time spent on these.

We follow up with every company we work with. So far, every hire that got at least 6/10 on a 6 hour test has been rated as "more effective than the majority of all hires" by their boss.

10 minute tests offered

These are just initial screens to check if someone knows the basics. We currently have tests for:

• Typescript
• Python
• Pytorch
• C
• CUDA (C++)
• Rust
• Golang
• Linux/bash/CLI
• Red-team security
• Blue-team security

6-hour tests offered

Our main offering. There's too much work to finish it all in 6 hours, so even the speediest devs have to prioritize. If you need more details about what is inside the tests, send us an email.

• Full stack: Implement the described app from top to bottom. API, DB queries, UI, authentication. Make it fast and make it look good.
• Data munger: Take the provided giant mess and put it into a tidy DB and get the requested metrics.
• ML ops: Get the provided GPU/TPU cluster working and set up a jobrunner with authentication. Our cluster will have a few unexpected issues.
• ML training: Get the highest accuracy you can on the provided objective. Sounds simple enough.
• Security: Hack the box. Figure out who was in the box before you. Harden the box.

Contact us if you would like to screen for something else. If a few folks request the same thing and we have the knowledge required, then we'll write a new test.

Integrations

We have integrations with Lever and we're happy to support other platforms when possible. We also let you export all your data as CSV or JSON.

Live identity verification (optional)

You want to know that the right person was at the computer during the test and whether they were controlling it. Our competitors only check who was there before and after, if they check at all, and nobody checks who was controlling it. (We wonder if this might be related to the well-known discrepancy between pre-hire and post-hire performance.)

Candidates do not need to install anything; they just stream a video of their keyboard + screen from their phone. We also check audio for suspiciously helpful friends.

Grading (optional)

The 10-minute tests are automatically graded at no additional charge.

The 6-hour tests are human-graded and take about 30 minutes to grade. You can grade submissions yourself, or we can grade for you at $50/candidate. We send you a report that looks roughly like this example, but with more detail:

Pricing

• No mandatory sales calls
• No contracts
• No slice of the candidates' salary

You just pay for the number of screening links you want to generate and use them when you want. The prices are at the bottom of this page.

ML tests include the cost of hardware rental (testing an ML engineer without GPUs is like testing a fish without water).

After you checkout there will be a short delay while we verify that you are a real company. Then we will send you account info. You sign in to make single-use or reusable test links, and buy more when you run out. You can set up emails/integrations for when candidates complete screens.

Order

We give no-questions-asked refunds for unused screening links.