What the DSAT Filters For (It Isn't Syntax)

Every applicant to Miatz passes through the same gate: the DSAT, a timed, integrity-monitored aptitude test. People assume an engineering school's entrance exam tests programming. Ours barely does — and never tests syntax recall at all.

That's not leniency. It's a position about what matters now, held firmly enough to build an admission gate on it.

Why syntax recall is a dead signal

For decades, knowing things like the exact incantation for a JavaScript closure or the flags on a tar command was a usable proxy for experience — annoying to acquire, so possessing it meant you'd put in hours. That proxy is dead. Models now hold every language's syntax simultaneously and produce it on demand. Testing humans on it in 2026 is like testing navigators on their memory of street names after the invention of the map app: you'd be selecting for the skill your tools just absorbed.

Worse, syntax questions actively mis-select. They favor recent grinders over deep thinkers and punish exactly the career-switchers and polyglots who often carry the strongest judgment.

We test for the things that make AI more powerful in your hands — because everything the model already does well, it will do for you.

The three filters

The DSAT measures three traits. Each maps to a specific demand of AI-era engineering.

Reasoning. Can you hold a system in your head and run it forward? Questions give you unfamiliar rules — a made-up protocol, a strange data structure, a policy with edge cases — and ask what must follow. No prior knowledge helps; that's the point. This predicts the core act of [driving AI](/mysty): a model hands you a plausible plan, and you must simulate its consequences before approving it. Passengers pattern-match on plausibility. Drivers execute the plan mentally and notice where it breaks.

Product sense. Given a goal and constraints, can you tell which requirement is load-bearing and which is decoration? DSAT items pose fuzzy, human situations — a feature request that contradicts itself, two stakeholders wanting incompatible things — and ask what you'd build, or refuse to build, and why. AI collapses the cost of building; it does nothing to tell you what's worth building. The judgment about 'what' and 'why' is precisely the part that stays human, so we filter for the raw material of it.

Attention to detail. The quiet one, and in our view the most predictive. Some questions contain small, consequential discrepancies — a spec that says 'at least' where the example implies 'more than', an off-by-one hiding in a boundary. Not tricks; exactly the class of discrepancy that fills real postmortems. GitLab's database was deleted by a right command on the wrong host. Knight Capital died by a reused flag on one server of eight. And reviewing AI output is detail work in its purest form: the code is always plausible, always confident, and wrong in one quiet place. The reviewer who reads carefully is the whole safety system.

What we deliberately don't ask

No language trivia. No framework questions. No 'what does this obscure operator do'. If a question could be answered instantly by the tools you'll have on the job every day, it has no business on an admission test for that job.

We also don't require a CS degree, prior professional experience, or any particular stack history. The DSAT is designed so a sharp tester, a self-taught tinkerer, and a lapsed physicist all get a fair shot at showing the three traits. The [founding cohort is free](/signup); the currency of admission is aptitude and seriousness, not biography.

Why a gate at all

Because the program assumes it. Everything past the gate — mastery gates instead of calendars, blunt War Room grading, propose-and-approve work with real audit trails — is built for people who can reason carefully and read closely. Admitting someone the program will grind down isn't kind; it's a wasted seat and a discouraged person. Selectivity here is quality control on the promise we make.

And there's the honest flip side: a gate protects the cohort. Learning accelerates around serious peers. Every person who passed the DSAT is one more colleague whose Saturday review, forum answer, and incident teamwork are worth your attention.

If you've read this far and felt more intrigued than intimidated, that's the profile. [Take the DSAT](/signup). It's about ninety minutes, and you'll know quickly whether we're your kind of hard.