Benefits and Leveling the Playing Field

The CSAT and Practice[+] are designed to enable candidates who deserve a place to stand out, with a keen eye towards the underprivileged, and to aid institutions to identify them.

Our experience and conviction are that it’s vital to allow each candidate to shine at what they are good at and at what they like. Otherwise, a deserving candidate can fall through shortcomings of standardized or multiple-choice tests, school performance, interviews, or total test scores. We believe it is important to recognize that there is no single aptitude or set of aptitudes that qualifies a candidate for a subject, especially one like Computer Science. The latter is very broad. Candidates with different strengths, e.g. pure maths versus algorithmic thinking, can equally excel in different areas of computer science. We want to enable all deserving candidates to succeed.

CSAT questions require reasoning rather than mathematical or computing recipes. Its questions are varied and exotic. As such, the CSAT is not a test one can typically optimize for in the way they can for standardized tests. CSAT questions are carefully designed with the following guiding ideal in mind: talented individuals who did not have much opportunity to train extensively should find it as challenging as highly trained individuals. The purpose is to identify deserving candidates irrespective of background or privilege.

We want a deserving candidate to stand out by employing solid reasoning. We went further and created the Practice[+] platform specifically to teach and build strong reasoning skills around non-trivial mathematical and computing problems.

Predicting undergraduate performance

Since 2015, the CSAT has been used at the University of Cambridge for admissions to Computer Science. Among known predictors, the CSAT has had the highest correlation with later undergraduate performance in the Cambridge course (aka Tripos), achieving a correlation coefficient of 0.492. For reference, the rule of thumb for predictors of Tripos performance has been that “a score of ~0.4 or above indicates that a model is one in which high confidence may be placed, and any score above ~0.2 is significant” [1].


The CSAT asks candidates to answer only a subset of questions, and they can choose which questions to answer, i.e. the questions they feel most comfortable with. This allows them to show- and institutions to see what they like and what their strengths are. They don’t have to answer algorithm questions if they enjoy pure maths more, or vice versa.


The knowledge required is from the syllabus covered in school by the time the CSAT is sat. There is enough choice for everyone regardless of the school curriculum. The CSAT strives to offer maximum choice, and also to signpost questions so candidates can determine if they have covered the required material for it.


CSAT questions are typically more varied and exotic than those in school exams, standardised tests, etc. They require reasoning and depth of thought rather than applying mathematical recipes. The CSAT does not test breadth of knowledge. Its questions prompt for constructing a line of reasoning, many times from first principles (trying to avoid requiring a “divine intervention”, which is more typically found in high level olympiads). Our guiding ideal is for deserving individuals who had less opportunity to train extensively to find it as challenging as highly trained ones. Our goal is for the CSAT to identify deserving candidates irrespective of background or privilege.


We mark everything written and then take into account only the best answers. For each question attempted, we look at the entirety of the written solution to identify the candidate’s reasoning. We ignore algebraic mistakes or silly mistakes in the line of reasoning. They may have proven that they understand a concept but used it in the wrong place and/or for the wrong thing, and that should be fine. Our (minutious) marking process strives to determine if their reasoning would lead to a correct solution, even if they stopped midway through; sometimes we even try to work out new alternative solutions prompted by incomplete workings that are outside our model solutions, in order to award marks. In the end, we build a profile of the best solutions, and measure the candidate’s performance on a number of distinct aptitudes. This is meant to advantage them, such that their choices of questions reveal what they are good at.


The CSAT is not meant to be an easy test, or a general aptitude test, as is apparent from the papers. This is to the candidate’s advantage. It is not uncommon for quality candidates taking easy tests to make silly mistakes (rushing the easiest things to get to the juicy things), making these costly. Silly mistakes are also costly in multiple-choice tests. More importantly, easy tests don’t allow quality candidates to stand out, as a large majority scores highly. Easy tests are suitable for shortlisting, while CSAT’s goal is to select deserving candidates. General aptitude tests do not go in depth into relevant technical/scientific skills, offering less opportunity to show true potential. The CSAT offers a chance to show fundamental aptitudes in relevant technical areas with questions of the candidate’s choosing.


Below are papers based off previous CSAT years, also covered on Practice[+].

Preparation and Practice[+]

The Practice[+] platform is a great place to learn and build solid reasoning skills for non-trivial mathematical and computing problems. However, it is not meant to be the only preparation or your first preparation. First, ensure you have covered the prerequisite school material.

If you haven’t ventured much outside of school material then it may be worth ramping up before attempting the CSAT, so that its exoticism and difficulty won’t create the wrong impression. You’d be surprised by how much you can learn and improve with some patience and perseverance. As the CSAT questions are diverse and exotic, and require reasoning rather than exercising knowledge, directly similar material may not be readily available online. For base knowledge one could start with year-11/12 UK A-level exams (past papers here, here and here).

Ramping up, it is a good idea to practise on questions from other competitions or exams. Helpful resources include the MAT, the UKMT (e.g. start with the IMOK and Senior Kangaroo then work towards the MOG), the STEP Correspondence Course (not ignoring the “warm down” questions). You may also practise on STEP papers in parallel with Practice[+]; you will notice a difference in style and targeted aptitudes between the two.