A hiring test that actually works for senior engineers

Why the standard loop fails at senior level

Algorithm screens measure how recently someone practiced algorithm screens. A staff engineer with fifteen years of production experience and two kids has not practiced; a bootcamp graduate with three months of free time has. You will systematically reject the more valuable candidate.

Unpaid take-homes fail differently: the strongest candidates already have offers and will not spend eight unpaid hours on yours. What you get back is a filtered sample of people with time, not people with judgment.

The four stages we use

• 1. Screen (45 min, senior engineer) — Walk through two systems they built, in depth. What broke, what they'd change, why the trade-offs were made. Seniors give specifics and own mistakes; imposters give architecture-diagram summaries and pass blame.
• 2. Paired build (90 min, their editor, their tools) — A realistic bug or small feature in a codebase they've never seen. AI assistants allowed and expected — you're evaluating how they navigate unfamiliar code, form hypotheses, and verify, not whether they memorized an API.
• 3. System design on their own system (60 min) — Not 'design Twitter.' Ask them to redesign a system they described in stage 1, given a 10x constraint. Real constraints, real numbers, real reasoning about failure modes and cost.
• 4. References (two, from the last 24 months) — Managers or tech leads, not peers. One question does most of the work: would you hire them again, and for what specifically? Hesitation is the signal.

The scoring rubric

Score four dimensions 1-4 each and require a 12 to advance: technical depth in their claimed area, debugging method under uncertainty, communication with a non-expert, and ownership evidence (they shipped it, maintained it, and lived with the consequences).

The single strongest predictor we've observed is stage 2 debugging method. Candidates who read before typing, state a hypothesis, verify it cheaply, and narrow systematically perform well on real work regardless of stack. Candidates who pattern-match and change three things at once do not.

Red flags and false negatives

Real red flags: cannot name a decision they got wrong; describes every previous team as the problem; the system-design answer is a list of technologies with no constraints attached; references are all peers and all recent.

Common false negatives worth guarding against: nervous performance in the first fifteen minutes, unfamiliarity with your specific framework (a strong senior picks it up in two weeks), and a resume of unglamorous domains. Some of the best engineers we've placed spent a decade in insurance or logistics.