Software Engineer Interview Evaluation Criteria: Why Coding Ability Is Scored Last, Not First

What Google Actually Hires On — and the One Surprise in the Order

Bock laid this out directly in an interview that's since been widely cited: Google hires against four attributes, not one. General cognitive ability — the capacity to learn and reason through unfamiliar problems, which Bock ranks above raw expertise (“what we hire for is not so much expertise or experience as learning ability”). Emergent leadership — stepping in when a problem needs solving and stepping back once it doesn't, with no interest in whether that happened on a resume-worthy title. “Googleyness” — a deliberately fuzzy bucket covering intellectual humility, comfort with ambiguity, and conscientiousness. And role-related knowledge — the actual technical skill being tested when you solve the coding problem.

• General cognitive ability — learning ability and reasoning under novelty, ranked as the attribute Google weighs most
• Emergent leadership — stepping up and stepping back as a situation requires it, independent of job title or tenure
• “Googleyness” — intellectual humility, comfort with ambiguity, conscientiousness; deliberately not reduced to a checklist
• Role-related knowledge — the technical depth needed to do the job — explicitly the least important of the four

That ordering is the part almost nobody preps for, because the coding round is the only one of the four that feels like a test with a right answer — the same instinct that drives most candidates to go silent and work alone during it, which is exactly the habit that costs points on the round that actually is graded on more than correctness.

“Role-related knowledge — the thing being tested when you solve the coding problem — is, in Bock's own words, “by far the least important thing we screen for.””

Why GPA and Test Scores Don't Predict Who Gets Hired

The same source makes a second claim that cuts against almost everyone's prep instincts: GPAs and test scores, Bock says, don't correlate with on-the-job success at Google. His explanation is structural, not dismissive — academic settings reward producing a specific, pre-known right answer under exam conditions. Real engineering work rewards something closer to the opposite: reasoning well through a problem nobody has handed you the answer key for.

That's a direct, if uncomfortable, answer to a question a lot of candidates ask themselves right before an interview: does my transcript matter here? For the specific evaluation being run in a software engineer loop at a company built around this framework, the answer is close to no — the thing actually being scored is how you reason live, not what a transcript says you once knew.

The same question, asked from the other side of campus placements in India, gets a similarly structural answer — see what the research says about CGPA as an eligibility gate versus a hiring signal, and the pattern holds for college tier too.

The Same Four Attributes, Scored Against a Different Bar at Every Level

None of those four attributes are scored against a fixed bar. They're scored against the bar for the specific level you're being evaluated for, and that bar moves a lot between levels. At Google, an L3 (Software Engineer II, roughly 0–2 years) is mainly judged on completing assigned work inside a defined scope and ramping fast. An L4 (Software Engineer III) is expected to own significant features independently — but is typically not assessed on system design at all, unless the loop is specifically probing for readiness at the next level up. L5, the first level the role itself calls genuinely senior, is where technical leadership across a team and influence on adjacent teams actually enters the scorecard.

This produces a specific, avoidable failure mode: a candidate interviewing for L5 who answers a system-design question with L4-level scope — correct, but narrow — can get leveled down after the debrief, even when every individual answer was technically right. The problem was never correctness. It was scope that didn't match the bar for the level in the room.

The same scope-over-correctness pattern shows up one role over, in a place most candidates don't expect it — a strong engineering manager system-design round that still fails the interview because the leadership round is scored on its own axis, not averaged against the technical one.

It's also the reason there's no single answer to “what does a data scientist interview test” — four genuinely different formats exist there too, and each one fails a candidate independently.

“A correct system-design answer scoped for the wrong level isn't a wrong answer. It's a lower level — decided in the debrief, not in the room.”

Amazon Scores a Completely Separate Axis, in the Same Round

Amazon's version of this makes the split even more explicit. Every onsite loop includes a Bar Raiser — an interviewer, not identified to the candidate in advance, trained specifically to keep the hiring bar consistent across teams, usually with no stake in whether this particular team fills its seat. The Bar Raiser leads the final debrief and can, in practice, hold up a hire even when every other interviewer wants to move forward.

What makes Amazon's loop structurally different from a pure coding gate: every interviewer grades a candidate on technical competency and, separately, on specific Leadership Principles — each interviewer is typically assigned two or three of Amazon's named principles to probe for and score on their own. That behavioral grading isn't a warm-up before the “real” technical interview. Reporting on Amazon's process is explicit that behavioral evaluation carries more weight there than at most peer big-tech companies, and the final recommendation — Strong Hire, Hire, No Hire, or Strong No Hire — gets filed by each interviewer independently, on both axes.

Which means the stories you'd rehearse for a dedicated behavioral round are being scored inside the same loop as your coding rounds, on equal footing — here's how to actually build a working set of Leadership Principles stories rather than discovering which ones you're short on mid-interview.

How to Find Out What Your Loop Actually Scores, Before You Walk In

You don't have to guess which of these four attributes, at which level, is going to decide your outcome. A few checks, done before you start prepping, narrow it fast:

• State a specific target level out loud to your recruiter, and ask what scope is expected at it — anchoring low, or staying vague, is how a correct answer still reads as under-scoped.
• Ask directly how many rounds the loop has and what each one is scored on. This is a normal scheduling question, not an awkward one, and the answer tells you far more than guessing from the job title.
• If the company names a framework — Leadership Principles, a competency list, a published rubric — treat it as the literal scoring sheet, not boilerplate values language on a careers page.
• Rehearse the behavioral and leadership answers with the same seriousness as the coding round, out loud, start to finish — not as a fallback for if the technical rounds go badly.

That rehearsal is the part that actually closes the gap, and it takes more reps than most candidates budget for — here's what the research says about how many it actually takes before an answer stops sounding rehearsed.

Key Takeaways

If you remember only a few things, remember these:

• Google hires software engineers against four attributes — general cognitive ability, emergent leadership, “Googleyness,” and role-related knowledge — and role-related knowledge (the coding round) is explicitly the least important of the four, per former Google SVP of People Operations Laszlo Bock.
• GPA and test scores don't correlate with on-the-job success at Google, by Bock's own account, because academic settings reward a known right answer and real engineering work doesn't.
• The bar for every attribute moves with the level being evaluated. At Google, system design generally isn't assessed at L4; a candidate targeting L5 who answers with L4-level scope can be leveled down even with a technically correct answer.
• Amazon scores a separate axis — named Leadership Principles — alongside technical competency in the same loop, with a trained Bar Raiser protecting the bar and each interviewer filing an independent recommendation.
• You can find out which attribute and which level your loop is actually scoring before you walk in: name your target level, ask how the rounds are scored, and rehearse the behavioral round as seriously as the coding one.