How We Match Problems to Companies

When AlgoArena says a problem resembles what a company asks, that is a measurement, not an impression. This post explains what we measure, and just as importantly, what we refuse to claim.

What we claim

That a company's interviews lean on some algorithmic patterns more than the industry norm, and that we can surface practice problems matching that emphasis. When we say a company over-indexes on trees, we mean tree problems make up a larger share of their reported questions than they do across employers generally.

What we do not claim

That any specific problem was asked at any specific company. We hold no such evidence, and every problem in our library is an original written for AlgoArena rather than a reproduction of anyone's interview question.

We also do not claim a profile for a company where the evidence is too thin to support one. That restraint is deliberate and it costs us coverage.

Where the data comes from

Publicly reported interview data: candidate write-ups, published pattern analyses, and what companies say about their own hiring process. Reports are aggregated into a topic distribution per company, weighted so that questions reported more often count for more.

It is self-selected data. People who report an interview are not a random sample of people who sat one, and we treat the results accordingly.

Why we compare against a baseline

The largest topics are the least useful. Array problems are roughly a sixth of every company's questions, and that share barely moves between employers. Ranking by raw popularity would return nearly the same problems for everyone and the feature would be worthless while looking arithmetically correct.

So every figure is relative: how much a company favours a pattern compared with the average across all employers we measure. That difference is the whole signal.

Companies we will not profile

For roughly half the companies in our data there are too few reported questions to tell a genuine preference from coincidence. Two tree problems out of four reported questions is not evidence a company favours trees, in the same way three heads in four coin flips is not evidence of a loaded coin.

We test every company's mix against what random chance would produce at that sample size. The ones that do not clear it are not given an invented profile. They are grouped with companies whose interviews genuinely resemble theirs, and labelled that way.

Where this is weakest

Most companies are not very distinctive. The honest finding from our own data is that a handful of employers have a clearly different emphasis and the rest look broadly similar to each other. Where two companies genuinely ask similar things, we show similar problems, because that is what is true.

Interview processes change. A profile reflects what was reported up to when we last rebuilt it, not what a company will ask you next month.

A second, weaker kind of claim

Everything above describes our algorithmic library. A growing share of interviews now ask candidates to build a small working system rather than solve a puzzle, so we write those separately rather than pretending an algorithm problem is the same exercise. Some of them carry a company name too, and that label means something different from the one on an algorithm problem. It is worth being plain about the difference rather than letting the two look equally solid.

A company label on an algorithm problem comes from the statistics above: a measured distribution, compared against a baseline, tested against what chance would produce. A company label on a build-a-system question does not. It says only that this employer has been publicly reported to ask questions of this shape, from a handful of accounts rather than from a distribution we can test. We would not build a filter on evidence that thin, so we have not.

What the label is for is honesty about where the format came from. The question itself is ours, written from scratch, and the point of naming the shape is so nobody has to guess whether we copied someone's interview. We did not.

How to use it

Treat a company filter as a way to weight your practice, not as a syllabus. The value is in noticing that one employer leans on graph traversal while another leans on intervals, then spending your time accordingly. Anyone promising the exact questions you will be asked is selling something we deliberately are not.

---

Company names and logos are used to identify the employers whose publicly reported interview patterns are described here. AlgoArena is not affiliated with, endorsed by, or sponsored by any company named on this site.