Screen applicants for free with Triplebyte

Triplebyte Screen is a fast, accurate way to measure programming ability in your applicants. Rather than focusing exclusively on a small number of algorithm problems (like other popular engineering screening services), Triplebyte Screen uses a large library of questions covering programming logic, system design, CS fundamentals, front-end engineering, iOS and Android. Like the GRE and GMAT, the test is adaptive. It selects the optimal questions for each candidate given the role they are applying for and how they've performed on the test so far. We built Triplebyte Screen by training it on 80,000 engineers over the last 3 years. The result is stunningly accurate. In our cross-validation, 90% of candidates who score a 5/5 on the test go on to pass interviews with senior engineers, while less than 5% of candidates who score a 1/5 or 2/5 go on to pass interviews.

We tested both - multiple choice and free form. We found that candidates prefer multiple choice questions and complete multiple choice tests at a much higher rate than free form tests. Additionally, multiple choice tests are just as predictive as free form versions.

We score candidates on a 5 point scale. We provide overall scores that are based on both accuracy and speed as well as a breakdown of scores for each topic that the test covers. When deciding whether to interview candidates, we generally recommend the follow:

• For 1/5 and 2/5 scores, the applicant is probably not a strong engineer
• For a 3/5 score, the applicant is of average ability — factor their background and experience into your decision
• For 4/5 and 5/5 scores, the applicant is a strong engineer and definitely worth talking to

This will vary based on your own pipeline quality, but on average, 70% of applicants score a 1/5 or 2/5. 15% will score a 3/5, 10% a 4/5 and finally, less than 5% of applicants will score a 5/5.

We've refined the test until it is highly predictive: applicants scoring a 5/5 on the test pass our interview 90% of the time.

The test is currently running with hundreds of calibrated questions. It gets larger every week.

Absolutely. Our model is recalibrated every night and we add new questions weekly. The test is calibrated on the hundreds of technical interviews we run every week. The test gets more accurate as we scale.

Our model takes into account more than whether a question was answered correctly. Time spent answering a question is also a factor, for example. We've found that multiple easy questions are often more predictive than a few hard questions. Easier questions are also less stressful and more fun for candidates, increasing overall completion rate, and an easy question can break up a run of hard questions and help candidates perform at their best on the harder questions. Overall, our model asks the question that we'll learn the most from, not the trickiest question.