The Online Assessment Funnel

The developer pipeline at trading firms is far more standardised at the top than the firms' own marketing suggests. Across twenty-five of the biggest firms, the first stage is almost always the same object: an auto-graded online assessment, two to four problems, sixty to a hundred and twenty minutes, delivered by one of a handful of platforms, three of which cover most of the market.

1. 1
   Online assessment. Auto-graded, timed, no human present. This is where most candidates are eliminated.
2. 2
   Recruiter screen. Background and motivation, occasionally a light technical question.
3. 3
   Live coding. A shared editor, usually CoderPad or the platform's own pairing mode, with an engineer watching you type.
4. 4
   Onsite or virtual onsite. Three to five rounds: more algorithms, a design round, and at the C++ shops a systems round on memory, cache and concurrency.

A few platforms, one shape

HackerRank dominates: it is the assessment platform at nearly half of the top firms. Codility and CodeSignal cover most of the remainder, and the rest is a long tail of one-offs, Coderbyte and TestGorilla among them.

The platform matters less than the format, and the format is remarkably consistent. You get a small number of problems, a hard clock, a function signature to fill in, and a set of tests you cannot see. Nobody reads the code. A harness runs it and reports a pass count.

Three shapes it comes in

The mixed battery is the one candidates underprepare for. One widely reported assessment runs three to four hours in three parts: two coding problems, around twenty computer-science multiple-choice questions, and a series of timed logic games. Another opens with ten C++ multiple-choice questions on language semantics and then gives you three coding problems, all inside roughly an hour. A third pairs three medium algorithm problems with ten probability questions and ten computer-science questions. If you prepared only algorithms, you have prepared for half the test.

The take-home track rewards the opposite skills. Candidates have been asked to build a Tetris variant and then change one of its rules live in the round that followed, and to implement an order book with matching graded against roughly thirty hidden tests. At least one firm publishes that it does not use whiteboard puzzles at all, and instead sets a single design-focused question you take home for about two hours.

Which language you are allowed to use

This is the detail that ends applications before they start, because it is rarely on the job posting.

• Some assessments exclude Python outright. One accepts C++, Java, C#, C and Ruby on its coding section and nothing else, while the onsite at the same firm is language-agnostic. A Python-only candidate is filtered by a machine before reaching the rounds they could have passed.
• A few are single-language. At least one firm's assessment is C++ or nothing.
• Some let you choose between C++, Python and TypeScript, and say so on their own engineering blog.
• Some accept any language, including firms known publicly for one unusual language of their own, who will tell you plainly not to attempt it unless you already write it well.
• Most of the rest leave the choice to you, while making it obvious from the job description which language the team actually writes.

Check the assessment invitation itself for the language list before you schedule it. If the list excludes your language, you have days rather than weeks to decide whether to sit it anyway.

What this course does about it

The pure-coding stage is trainable, and it is trainable in a specific way: not by solving more problems, but by solving the archetypes these firms return to, under a clock, in code you would be willing to have read afterwards. The next lesson covers what the grader is actually measuring, which is narrower and stranger than most candidates assume.