Programming skills assessment for hiring

What the 83 tests cover

Most tests cover one named technology. Web development is the largest group, with 27 tests on front-end and back-end frameworks, from React and CSS to Django. Software development has 19, which includes the coding tasks and practice topics such as Clean Code. AI and machine learning has 9, among them PyTorch. Data and analytics has 7, such as Pandas and Apache Spark for Data Engineers. Cloud and DevOps has 7, including Microsoft Azure. Mobile development has 6, from Flutter to Android Development Using Kotlin. Databases has 4, such as MySQL. A few tests on shell scripting, cryptography and Salesforce's Apex language round out the group.

The 69 knowledge tests are multiple choice, built from short code snippets and scenarios. Almost all take 10 minutes; Pandas takes 12.

The other 14 tests are coding tasks: the candidate writes a program, and hidden test cases check what it outputs. Twelve are algorithm, data structure and debugging tasks, such as Coding: Arrays and Coding: Graphs. Two are SQL tasks run on SQLite.

How the coding tasks work

Each coding test is a single task with one time limit. Coding: Algorithms, Entry Level takes 15 minutes. The intermediate algorithm and data structure tasks take 35 minutes each, as does Coding: Debugging, the one advanced task.

In the algorithm tasks, the candidate chooses from six common languages, and every task is checked to work the same way in each. The debugging task gives the candidate unfamiliar code and a written specification. They find where the code goes wrong and repair it without breaking what already works.

The two SQL tasks ask for a query from a short business brief. SQL Coding: Database Operations is entry level and takes 20 minutes: filtering, simple calculations, grouping and sorting. SQL Coding: Querying is intermediate and takes 30 minutes against a database with several tables.

All 14 coding tests ask for a computer, so the editor is comfortable to use. So do Pandas and Data Analysis Using R. The other 67 tests also work on a phone.

Which tests suit which roles and levels

Match the test to the stack in the job description, and the level to the seniority. Of the 83 tests, 75 are intermediate.

The two entry-level tests are both coding tasks. Coding: Algorithms, Entry Level suits a junior developer, graduate or intern who writes code. SQL Coding: Database Operations suits junior back-end developers, data and reporting assistants, and QA testers who read from small databases.

The six advanced tests are for senior specialists. Advanced Networking in Azure and its AWS and Google Cloud versions suit network and cloud architects. Computer Vision and Multi-Task Learning (MTL) suit machine learning engineers and applied scientists. Coding: Debugging suits experienced engineers who maintain or review other people's code.

The intermediate tests name mid-level roles. React suits front-end and full-stack developers. Node.js and Express.js suit back-end developers who build APIs. React Native suits mobile developers. Pandas and Data Analysis Using R suit analysts and data scientists who code. Coding: Algorithms, Intermediate suits mid-level engineers who write production code.

How to combine them

A good set for a developer pairs what they know with what they can build. Take one or two knowledge tests for the stack, at 10 minutes each, and add one coding task that fits the work.

A front-end developer could take React and CSS, then Coding: Arrays. A back-end developer on Python could take Django and REST API Design, then SQL Coding: Querying. A data engineer could take Apache Spark for Data Engineers with SQL Coding: Querying, since both name data engineers. A junior developer could take Coding: Algorithms, Entry Level with SQL Coding: Database Operations, about 35 minutes in all.

A 35-minute coding task is the longest part of a set, so one task is enough unless the role needs two distinct coding skills.

For tools around the code, such as Git, Docker or PostgreSQL, see the software skills tests in the test library.

Each score is a signal for your team, not a verdict. There is no pass mark, and no result moves a candidate on or out by itself. A coding result reads best beside a conversation about how the candidate approached the task.

How the tests are kept fair

The fairness notes for these tests aim at one thing: the answer should turn on programming skill, not on which version, language or company a candidate learned on.

In the coding tasks, candidates pick the language they know best, and each task is written to behave the same way in every language offered. The debugging task plants the same faults in every language, and a language-specific trick is never the fault to find. Every rule the hidden cases check, such as case, spacing or ordering, is stated in the task. The SQL tasks state the table layout, the column meanings and the exact result expected.

In the knowledge tests, code is short, neutral and stated against current releases. React questions have the same answer in React 18 and React 19, and MySQL questions state the version and settings they assume. Microsoft Azure describes services by what they do as well as by name, so a recent rename does not trip anyone up.

Scenarios use everyday data such as orders, bookings and task lists, with names from many backgrounds and no idioms, sports or local knowledge. Money appears as plain numbers or neutral units. Scratch paper is allowed on 29 of the tests.

See Fairness and accommodations for how candidates request support, and How we build our tests for how questions are written and reviewed.