The grading rubric

How you’re judged.

Placed has five interview modes, each with its own rubric modeled on what real interviewers actually listen for. This is the full breakdown — the signal, the pitfalls, and exactly what the AI grader is checking on your behalf.

Learn the patterns behind the questions
Mode 01Classic algorithmic

Coding Interview

A timed session with an AI interviewer who probes your reasoning while you code. Default 25 minutes, and you can dial it from 15 up to 45 before you start if the problem looks harder. Five criteria scored 0–10 each, graded at a top-tier SWE hiring bar — well-written but behavior-empty answers are capped at 5.

15–45 min · 25 default0–10 scale5 criteria
01

Problem Understanding & Clarification

0–10

"Do they actually know what we're solving?" Top candidates restate the problem, nail down inputs/outputs, and surface edge cases before writing a single line of code.

Good signal

  • Clarifying questions up front

    Pins down whatever the problem leaves open (input ranges, output shape, edge cases that matter) before coding.

  • Restates in their own words

    Plays the problem back to confirm understanding.

  • Surfaces assumptions explicitly

    "I'm assuming the array fits in memory — let me know if not."

Bad signal

  • Starts coding on first read

    No clarification, visible misreading of the problem later.

  • Silent assumptions

    Builds for one shape of input without checking.

  • Asks only when stuck

    Clarifies mid-implementation when it's already costly.

Common pitfalls

  • ·Skipping clarification on "easy-looking" problems — that's exactly where misreads hide.
  • ·Asking clarifying questions but not listening to the answer.
›What the AI grader actually checks

9-10 = Pinned down the problem BEFORE coding: restated it and asked about whatever the statement left open (a clear restatement alone is enough on a simple, fully specified problem). 7-8 = Understood the problem before implementing, through questions or a clear restatement. 5-6 = Restated the problem or asked a little, but left an open point that mattered unverified. 3-4 = Minimal clarification with visible misunderstandings, or none asked but the problem understood. 0-2 = No clarification; coded from a misreading.

02

Problem Solving & Algorithm Design

0–10

"Can they get to a correct, efficient answer — and do they know why it's correct?" Pattern recognition, decomposition, and complexity awareness.

Good signal

  • Traces through examples first

    Runs a small case by hand to expose structure before coding.

  • Names the pattern

    "This is sliding window" / "DP on subsequences" — and defends the fit.

  • States and improves complexity

    Starts from a brute force, optimizes with justification.

Bad signal

  • Jumps to code without a plan

    Debugs into the answer instead of designing it.

  • Ignores edge cases

    Empty input, duplicates, overflow — never surfaced.

  • Can't justify complexity

    Writes nested loops without reasoning about cost.

Common pitfalls

  • ·Forcing a memorized pattern onto a poorly-fitting problem.
  • ·Declaring "O(n)" without accounting for hidden work (hash resize, inner sort).
  • ·Optimizing prematurely before a correct baseline exists.
›What the AI grader actually checks

9-10 = Optimal approach, articulated complexity, weighed trade-offs, decomposed cleanly. 7-8 = Correct approach with sound complexity reasoning. 5-6 = Optimal approach but complexity never stated or only hand-waved; or works but not optimal. 3-4 = Brute-force only with no attempt at better, or an approach they could not explain. 0-2 = Approach is incorrect or incoherent.

03

Technical Communication

0–10

"I can solve it in my head too — I need to hear how you think." Interviewers hire the person they want to pair with, not the one who silently produces an answer.

Good signal

  • Narrates intent before typing

    States the approach and why it fits before writing code.

  • Responds precisely to probes

    Answers the interviewer's actual question, not a nearby one.

  • Holds the thread when stuck

    Keeps narrating even when the going gets hard.

Bad signal

  • Heads-down silent coding

    Ten minutes pass without a word — nothing to evaluate.

  • Monologuing

    Talks at the interviewer instead of inviting collaboration.

  • Surprise pivots

    Changes approach mid-solution without explaining why.

Common pitfalls

  • ·Going silent the moment things get hard — that's exactly when narration matters most.
  • ·Vague words ("it works", "something like this") instead of precise ones.
  • ·Arguing with interviewer push-back instead of probing their concern.
›What the AI grader actually checks

9-10 = Narrated intent continuously, responded precisely to probes, checked in at decision points, held the thread when stuck. 7-8 = Thinking mostly visible. 5-6 = Intermittent narration; interviewer had to probe. 3-4 = Mostly silent coding. 0-2 = No useful verbal signal; intent opaque.

04

Code Quality & Craft

0–10

"Would I want to review this in a PR?" The final artifact matters — variable names, structure, and the ugly cases are all signal.

Good signal

  • Readable names and structure

    Short functions, intention-revealing names, no mystery vars.

  • Idiomatic

    Uses the language's strengths rather than fighting them.

  • Handles edge cases explicitly

    Null / empty / boundary inputs are addressed.

Bad signal

  • Copy-paste duplication

    Same three lines appear four times with no extraction.

  • Magic numbers and cryptic vars

    `i`, `j`, `tmp`, `x2` with no context.

  • Off-by-ones left in

    Submits without tracing boundaries.

Common pitfalls

  • ·Optimizing for brevity at the cost of clarity (clever one-liners).
  • ·Leaving debug `console.log`s or commented-out code in the final submission.
  • ·Forgetting to return, or returning the wrong thing.
›What the AI grader actually checks

9-10 = Readable, well-structured, idiomatic, no dead code, naming carries intent. 7-8 = Solid with minor stylistic slips. 5-6 = Works but messy — several issues a reviewer would call out. 3-4 = Quality problems hurt readability. 0-2 = Barely-working code, or scaffolding without substance. Code that passes every test at the best known complexity scores at least 7.

05

Testing & Verification

0–10

"Show me you don't just trust your own code." The single most common missing signal in 2025–26 SWE loops. Interviewers want to watch you walk through your own solution.

Good signal

  • Dry-runs example inputs

    Traces happy path and at least one edge case line-by-line.

  • Proactive edge-case hunt

    Names empty / single / duplicate / overflow cases unprompted.

  • Debugs methodically

    When something breaks, isolates cause instead of guess-and-check.

Bad signal

  • Declares "done" without walking through it

    Hands in code they haven't verified.

  • Confuses "compiles" with "correct"

    No real trace, just absence of syntax errors.

  • Only tests happy path

    Boundary cases untouched.

Common pitfalls

  • ·Running the code but not reading what it actually output.
  • ·Claiming correctness confidently without evidence.
›What the AI grader actually checks

9-10 = Proactively traced through example inputs, handled the edge cases that matter for this problem, debugged methodically. 7-8 = Traced at least one example end-to-end; considered an edge case where the problem has one that matters. 5-6 = Some verification, mostly prompted. 3-4 = No real testing. 0-2 = Did not verify; made confident claims without evidence.

Overall rating thresholds

Insufficient

overall < 4.0

Developing

overall 4.0–6.4

Proficient

overall 6.5–8.4

Exceptional

overall ≥ 8.5

Your final rating is the mean of your per-criterion scores (0–10), rounded to one decimal.

Start a Coding Interview
Mode 02Code with AI — graded like a real AI-enabled round

GenAI Coding

You solve a problem with an AI assistant, the way the AI-enabled rounds at Meta and Google actually run. Default 25 minutes, dialable from 15 up to 45 before you start. You're scored on the SAME four competencies as a traditional coding interview — the AI changes how the evidence is gathered, not what's being measured. How you prompt is evidence, not a score: nobody is hired for writing a tidy prompt. Hard caps apply to observable behaviours like never running the code or submitting AI output verbatim.

15–45 min · 25 default0–10 scale4 criteria
01

Problem Solving

0–10

"Did they decompose this themselves, or delegate the thinking?" The work you do before the first prompt is the work being scored.

Good signal

  • Decomposes first

    Breaks the problem into subparts before prompting.

  • Names constraints

    Input sizes, edge cases, and what they rule out.

  • Justifies the approach

    "Edges are unweighted, so BFS gives the shortest path."

Bad signal

  • Asks the AI what to do

    "How should I solve this?" as the opening move.

  • No constraint thinking

    Never asks what happens at scale or at the edges.

  • Approach by accident

    Builds whatever the first response happened to suggest.

Common pitfalls

  • ·Letting the model pick the algorithm and calling it your decision.
  • ·Coding before you can state the approach in one sentence.
›What the AI grader actually checks

Context built across multiple turns counts equally to up-front framing. 9-10 = Broke the problem into meaningful subparts before prompting; named constraints and edge cases; chose an approach and said why. 7-8 = Clear grasp of the problem and a defensible approach; minor gaps in edge-case thinking. 5-6 = Understood the problem but let the AI choose the approach. 3-4 = Handed the raw problem over and worked from what came back. 0-2 = No independent problem-solving visible. (Hard cap: submitted AI code verbatim → max 5.)

02

Code Quality

0–10

"Would I approve this PR, and can they defend it?" Ownership is part of this score — code you can't account for isn't quality code, however well it reads.

Good signal

  • Readable and idiomatic

    Clear names, sensible structure, no dead code.

  • Adapted, not accepted

    Reshaped AI output to fit the problem rather than pasting it.

  • Can explain every line

    Engaged with each part enough to defend it.

Bad signal

  • Mystery blocks

    Whole sections the candidate never examined.

  • Verbatim submission

    The assistant's output, untouched, shipped as the answer.

  • "The AI wrote it"

    Can't account for a section when asked.

Common pitfalls

  • ·Mistaking "it runs" for "I understand it".
  • ·Letting structure drift because you accepted whatever came back each turn.
›What the AI grader actually checks

9-10 = Readable, idiomatic, well-structured; every part is something they engaged with, adapted, or interrogated. 7-8 = Solid code with minor style issues; clear ownership. 5-6 = Works but messy, or partly accepted without scrutiny. 3-4 = Poor structure, or long stretches never examined. 0-2 = Barely functional, or wholesale AI output they show no understanding of. (Hard cap: submitted AI code verbatim → max 4. Floor: modified AI code → min 5.)

03

Verification

0–10

The rhythm interviewers watch for is prompt, review, run, confirm, move on. Building on unverified output is the failure mode.

Good signal

  • Runs constantly

    Executes after each meaningful change, not once at the end.

  • Tests the edges

    Empty input, single element, duplicates, large input.

  • Challenges output

    Catches an AI mistake and says how they caught it.

Bad signal

  • Never ran it

    Submitted code that was never executed.

  • Happy path only

    One example input, no edges.

  • Builds on unverified code

    Stacks the next step on an unchecked answer.

Common pitfalls

  • ·Trusting confident-sounding output because it is well formatted.
  • ·Leaving all testing to the last two minutes.
›What the AI grader actually checks

9-10 = Ran code frequently, tested edge cases, challenged suspicious output, caught an AI mistake. 7-8 = Ran the code and made at least one meaningful check before proceeding. 5-6 = Ran the happy path only. 3-4 = Took output at face value. 0-2 = Never ran or verified anything. (Hard caps: never ran the code → max 3; no edge case probed → max 6.)

04

Communication

0–10

In a live round you narrate to a human. Here the proxy is your transcript: can a reader follow your reasoning, your constraints, and why you kept or rejected each answer?

Good signal

  • States intent

    Says what they are trying to do before asking for it.

  • Reasons out loud

    Explains why an answer was accepted or rejected.

  • Specific context

    Names the constraint, the shape, the expected behaviour.

Bad signal

  • Bare commands

    "fix this", "make it faster" — no framing at all.

  • Invisible reasoning

    Decisions appear with no stated why.

  • Silent acceptance

    Output taken without a word either way.

Common pitfalls

  • ·Going quiet once the AI starts producing working code.
  • ·Assuming the reader can infer intent you never stated.
›What the AI grader actually checks

9-10 = Intent explicit throughout; context specific; reasoning about AI output visible and easy to follow. 7-8 = Generally clear; a reader can reconstruct the thinking. 5-6 = Terse or under-specified. 3-4 = Bare commands with no framing; reasoning invisible. 0-2 = Nothing communicated beyond dumping the problem. (Hard cap: fewer than 3 prompts → max 5.)

Overall rating thresholds

Insufficient

overall < 4.0

Developing

overall 4.0–6.4

Proficient

overall 6.5–8.4

Exceptional

overall ≥ 8.5

Your final rating is the mean of your per-criterion scores (0–10), rounded to one decimal.

Start a GenAI Coding Session
Mode 03Read a broken PR — graded on ranking, not volume

Code Review

You are handed a pull request with real defects in it and asked to review it the way you would on a Tuesday. Airbnb built a rubric-scored round around this; Google has handed candidates a 200-line program to comment on. The thing that decides it is SEVERITY RANKING, not comment count — a review that finds two nits and misses a data-loss bug fails no matter how long it is. Every finding you leave carries a severity, and you should be able to say what it costs in production.

15–45 min · 25 default0–10 scale4 criteria
01

Problem Solving

0–10

"Did they find the thing that would have taken us down?" Finding the serious defect is worth more than finding six small ones.

Good signal

  • Catches the criticals

    The security hole and the data loss are in the review, near the top.

  • Explains the mechanism

    Says how the failure actually happens, not just that it looks wrong.

  • Reads what was deleted

    Notices the check that was removed and never replaced.

Bad signal

  • Skims the additions

    Reviews the new lines and ignores what the diff took away.

  • Right line, wrong reason

    Flags the correct spot with an explanation that does not hold.

  • Surface only

    Naming, formatting, and nothing that would page anyone.

Common pitfalls

  • ·Treating a diff as a list of added lines rather than a change in behaviour.
  • ·Reviewing the code that is there instead of the code path that runs.
›What the AI grader actually checks

9-10 = Found every critical defect and the majors, and explained the mechanism correctly. 7-8 = Found the criticals; missed or under-explained a major. 5-6 = Found some real defects but missed a critical, or explained one incorrectly. 3-4 = Mostly surface observations; the serious defects went through. 0-2 = Nothing of substance found. (Hard cap: a critical defect was missed = max 4.)

02

Code Quality

0–10

"If the author did exactly what this review says, is the code right afterwards?" A correct diagnosis with a broken prescription is half a review.

Good signal

  • Actionable fixes

    Says what to do, specifically enough to act on without a follow-up.

  • The right standard

    Holds the code to the bar its blast radius deserves.

  • Fixes that work

    The suggested change actually resolves the defect.

Bad signal

  • Vague direction

    "This should be handled better" with no handling described.

  • Cosmetic only

    Every suggestion is about how the code reads.

  • Harmful advice

    A fix that would introduce a new problem.

Common pitfalls

  • ·Describing the problem and stopping short of the remedy.
  • ·Applying a style guide where a correctness argument was needed.
›What the AI grader actually checks

9-10 = Every suggestion is correct and actionable; the resulting code would be right. 7-8 = Sound suggestions with a gap or an imprecision. 5-6 = Identifies problems but is vague about the fix. 3-4 = Suggestions are wrong, harmful, or purely cosmetic. 0-2 = No usable suggestion at all. (Hard cap: no findings at all = max 2.)

03

Verification

0–10

"Did they work out what breaks, or did they recognise a shape?" Naming the input or the condition is the difference between a review and a lint rule.

Good signal

  • Traces the failure

    Follows the path: this throws, so this returns, so this is authenticated.

  • Names the trigger

    The specific input, state, or outage that sets the defect off.

  • Reasons about time

    What the cache holds an hour from now, not just what it holds today.

Bad signal

  • Pattern matching

    "Bare except is bad practice" with no consequence attached.

  • Assertion without a path

    Calls something a bug without saying when it fires.

  • Ignores the failure branch

    Reviews the happy path only — the same gap the PR author had.

Common pitfalls

  • ·Reviewing against a checklist instead of against the running code.
  • ·Assuming the error handler is fine because an error handler exists.
›What the AI grader actually checks

9-10 = Named concrete failure scenarios (what happens when rotation throws, what the cache holds after an hour) and reasoned from the code, not from appearances. 7-8 = Reasoned about consequences for the main defects. 5-6 = Asserted problems without working through what actually goes wrong. 3-4 = Pattern-matching only ("bare except is bad practice"). 0-2 = No reasoning about behaviour at all. (Hard cap: no findings at all = max 2.)

04

Communication

0–10

A review is read by someone who has to act on it. Is the important thing unmissable, and is a nit visibly a nit?

Good signal

  • Honest severities

    The critical is marked critical; the preference is marked nit.

  • Ordered by weight

    The thing that matters is not buried under four style notes.

  • Each note acts

    The author can work straight from the review.

Bad signal

  • Flat severities

    Everything filed at the same level, so nothing stands out.

  • Inflated nits

    Naming preferences marked major — the fastest way to lose a reviewer their credibility.

  • Buried lede

    The security hole is comment nine.

Common pitfalls

  • ·Spending the round on naming while something serious goes through — the single most common way this round is failed.
  • ·Withholding small observations entirely; noticing a nit AS a nit is correct ranking, not noise.
›What the AI grader actually checks

9-10 = Severity assignments match reality, the important things are unmissable, each note says what to do. 7-8 = Clear and mostly well-ranked. 5-6 = Understandable but flat — everything at one severity, or the important note buried. 3-4 = Mis-ranked (nits marked major, or a critical marked minor), or too vague to act on. 0-2 = Unusable as a review. (Hard cap: mostly nits while a critical was missed = max 4.)

Overall rating thresholds

Insufficient

overall < 4.0

Developing

overall 4.0–6.4

Proficient

overall 6.5–8.4

Exceptional

overall ≥ 8.5

Your final rating is the mean of your per-criterion scores (0–10), rounded to one decimal.

Start a Code Review Round
Mode 04Behavioral assessment — how you think about AI

GenAI Fluency

Seven behavioral criteria scored 0–10 each. This isn't about code — it's about the judgment, ethics, and communication skills hiring managers probe when deciding if you can be trusted with AI tools at work. Hypothetical "I would" answers are capped hard.

Open-ended0–10 per criterion7 criteria
01

Specificity & Concreteness

0–10

"Tell me about a time you…" — the signal is a specific tool, task, project, and outcome. Vagueness reads as inexperience.

Good signal

  • Names the tool

    "Copilot in VS Code", "Claude via the API" — not "an AI".

  • Names the project

    Concrete artifact, timeframe, outcome.

  • Verifiable detail

    The story could be checked against a commit history or PR.

Bad signal

  • Hypothetical framing

    "I would use AI to…" instead of "I used AI to…"

  • Generic "we used AI"

    No tool, no task, no outcome.

  • Could apply to anyone

    No personal fingerprint on the story.

Common pitfalls

  • ·Hiding behind "we" instead of saying what you did.
  • ·Picking a tiny example when a bigger one exists.
›What the AI grader actually checks

9-10 = Names specific tool, task, project, verifiable outcome. 7-8 = Mostly specific, one key detail missing. 5-6 = Generic but names at least a tool or context. 3-4 = Vague; no named artifacts. 0-2 = Entirely hypothetical. (Hard cap: no tool named → max 3. No project/artifact → max 4.)

02

GenAI Literacy

0–10

Can you explain what the tool actually does, and where it fails? Understanding hallucination, prompt sensitivity, and variability is table stakes.

Good signal

  • Knows the failure modes

    Hallucination, prompt sensitivity, stale training, nondeterminism.

  • Uses correct terminology

    Distinguishes completion from chat, RAG from fine-tuning.

  • Maps tools to tasks

    Knows which tools suit which jobs and why.

Bad signal

  • Treats it as magic

    No mental model for why it fails.

  • Conflates AI with search

    Expects deterministic, factual output.

  • One tool for everything

    Never evaluated alternatives.

Common pitfalls

  • ·Parroting marketing copy instead of speaking from experience.
  • ·Overclaiming deep knowledge of training dynamics.
›What the AI grader actually checks

9-10 = Clear understanding, aware of hallucinations, prompt sensitivity, variability. 7-8 = Practical understanding, no deeper technical awareness. 5-6 = Uses GenAI as black boxes. 3-4 = Holds misconceptions. 0-2 = Conflates GenAI with deterministic software. (Hard cap: no failure mode mentioned → max 5.)

03

Critical Thinking & Evaluation

0–10

Do you have a review process? Proactive verification separates someone who ships AI-assisted work from someone who ships AI slop.

Good signal

  • Has a review ritual

    Runs, tests, reads — every time, not occasionally.

  • Caught real errors

    Specific time they caught a hallucination before it shipped.

  • Iterates on prompts based on quality

    Treats low-quality output as feedback.

Bad signal

  • Gut-feel review

    "If it looks right, it probably is."

  • Only reviews on complaint

    Reactive, not proactive.

  • Takes at face value

    No mental check between read and ship.

Common pitfalls

  • ·Claiming to "always check" without describing how.
  • ·Over-trusting output from a familiar tool.
›What the AI grader actually checks

9-10 = Clear review process, caught errors proactively, iterates based on quality. 7-8 = Reviews but process informal. 5-6 = Occasional review, mostly gut-feel. 3-4 = Accepts with minimal scrutiny. 0-2 = Takes AI content at face value. (Hard cap: no failure mode mentioned → max 5.)

04

Judgment & Risk Awareness

0–10

When NOT to use AI is as important as when to. Teams need people who weigh downstream risk, not enthusiasts who reach for AI reflexively.

Good signal

  • Explicit decision criteria

    "I use AI for X and Y, but not for Z because…"

  • Has said no

    Specific time they chose not to use AI for a task.

  • Thinks downstream

    Considers who else is affected by the output.

Bad signal

  • AI as default

    Starts every task by opening the AI tool.

  • Reactive risk awareness

    Only thinks about risk after a problem.

  • No holding back

    Has never declined to use AI.

Common pitfalls

  • ·Equating "no AI" with "against AI" — interviewers hear the nuance.
  • ·Conflating risk awareness with fear.
›What the AI grader actually checks

9-10 = Articulates decision criteria, has examples of choosing NOT to use GenAI. 7-8 = Aware of risks, applies judgment reactively. 5-6 = Uses GenAI broadly without strong filtering. 3-4 = Weak risk awareness. 0-2 = Views GenAI as a default solution. (Hard cap: no "chose NOT to" example → max 6.)

05

Responsibility & Ethics

0–10

Data privacy, bias, IP, compliance, human accountability — hiring managers want people whose ethical instincts are proactive, not policy-driven.

Good signal

  • Concrete responsible decisions

    Redacted data before a prompt, or flagged a bias issue.

  • Multiple dimensions

    Privacy AND IP AND bias — not just one.

  • Owns accountability

    Doesn't hide behind "the AI decided".

Bad signal

  • Ethics only when prompted

    Doesn't bring it up unless asked.

  • Delegates to policy

    "Compliance handles that" — no personal framework.

  • Unaware of risks

    No mention of privacy, bias, or IP.

Common pitfalls

  • ·Reciting buzzwords without concrete examples.
  • ·Treating ethics as a blocker instead of a design constraint.
›What the AI grader actually checks

9-10 = Proactively considered privacy, bias, IP, compliance, or accountability with concrete examples. 7-8 = Aware, relies on org policy. 5-6 = Mentions ethics only when prompted. 3-4 = No concrete examples. 0-2 = No awareness.

06

Learning Agility

0–10

AI tooling changes monthly. Candidates who experiment and iterate beat candidates who plateaued on the first tool that worked.

Good signal

  • Deliberate experimentation

    Tries new tools, new techniques, with clear hypotheses.

  • Learns from failures

    Specific story of a failure that changed how they work.

  • Skills visibly evolving

    Can describe what they do now that they didn't six months ago.

Bad signal

  • Same workflow as day one

    No evolution in approach.

  • Passive learning

    Only improves when mistakes force it.

  • No curiosity

    Hasn't explored beyond defaults.

Common pitfalls

  • ·Conflating tool-switching with learning.
  • ·Learning trivia (model names) without practical change.
›What the AI grader actually checks

9-10 = Deliberate experimentation, iteration, learning from failures; skills evolving. 7-8 = Has improved, mostly reactively. 5-6 = Same way they started. 3-4 = Little evidence of growth. 0-2 = No curiosity.

07

Communication & Influence

0–10

Can you translate AI capabilities to non-technical stakeholders, and navigate skeptics? Influence is the multiplier on every other skill here.

Good signal

  • Audience calibration

    Explains AI differently to execs, engineers, end users.

  • Has shifted minds

    Brought a skeptic or zealot to a balanced view.

  • Clear, jargon-free writing

    Describes what AI did without impenetrable vocabulary.

Bad signal

  • One-mode communicator

    Same jargon for every audience.

  • Can't say why

    Describes what but not the reasoning.

  • Never navigated resistance

    No examples with skeptical colleagues.

Common pitfalls

  • ·Overloading technical detail to sound credible.
  • ·Dismissing skeptics instead of engaging.
›What the AI grader actually checks

9-10 = Calibrates message to different audiences; has influenced others. 7-8 = Communicates clearly, no significant resistance navigated. 5-6 = Can explain what, struggles with why. 3-4 = Jargon-heavy. 0-2 = Communication unclear throughout.

Automatic red flags

Hit any of these during the assessment and they’re flagged in your report regardless of score:

  • RF1Candidate has never used a GenAI tool in a real work or personal context
  • RF2Candidate shows no awareness that GenAI can produce inaccurate or misleading output
  • RF3Candidate has never considered data privacy when using GenAI tools
  • RF4Candidate describes using GenAI to generate work submitted as their own without any review
  • RF5Candidate expresses blanket refusal to use GenAI tools without a reasoned explanation

Overall rating thresholds

Insufficient

overall < 4.0

Developing

overall 4.0–6.4

Proficient

overall 6.5–8.4

Exceptional

overall ≥ 8.5

Your final rating is the mean of your per-criterion scores (0–10), rounded to one decimal.

Start a Fluency Assessment
Mode 05Classic SWE "tell me about a time"

Behavioral Interview

Two STAR-format questions drawn from publicly-reported behavioral rounds at Google, Meta, Amazon, Stripe, and others. Eight criteria scored 0–10, grader calibrated to the bar a real hiring manager would set. Hypothetical answers are capped at 3.

~15 min0–10 per criterion8 criteria
01

Specificity & Evidence

0–10

"Is this a story that actually happened?" Specific tool, artifact, person, timeframe, and verifiable outcome — the #1 thing hiring managers listen for. Hypothetical answers get capped at 3.

Good signal

  • Names the artifact

    Specific project, tool, commit, document, meeting.

  • Names the people

    Team composition, stakeholder role — not "someone".

  • Verifiable timeframe

    "Q2 2024" — not "a while back".

Bad signal

  • "I would"

    Hypothetical or aspirational framing.

  • No artifact named

    The story could apply to any project.

  • No timeframe

    "At some point" — impossible to ground.

Common pitfalls

  • ·Over-sanitizing the story until no specifics remain.
  • ·Keeping the story vague "to protect confidentiality" when a general summary would be fine.
›What the AI grader actually checks

9-10 = Names specific tool, artifact, project, person, timeframe; outcome is verifiable. 7-8 = Concrete example, one detail missing. 5-6 = Generic but attempts named specifics. 3-4 = Vague; no named artifacts. 0-2 = Entirely hypothetical.

02

Ownership & Accountability

0–10

"Who actually did the work?" Interviewers listen for clear "I" framing — what you personally did, decided, and owned, not what the team did around you.

Good signal

  • Clear "I" framing

    "I proposed…", "I decided…", "I shipped…" — specific to you.

  • Owns the outcome

    Good or bad — takes responsibility without shifting blame.

  • Names what was at stake

    Explains why the decision mattered and who it affected.

Bad signal

  • Heavy "we" framing

    "We built…", "We decided…" — impossible to tell what you did.

  • Passive voice

    "It was decided that…" — distances you from the call.

  • Deflects blame

    When things went wrong, it was someone else's fault.

Common pitfalls

  • ·Overclaiming — taking credit for things a teammate drove.
  • ·Undersharing — hiding behind the team when you actually led.
  • ·Confusing responsibility with authority.
›What the AI grader actually checks

9-10 = Uses "I" correctly, names what they did, owns outcomes. 7-8 = Mostly owns it with occasional drift to "we". 5-6 = Heavy "we" framing. 3-4 = Attributes credit or blame away. 0-2 = Deflects entirely. (Hard cap: heavy "we" → max 4.)

03

Dealing with Ambiguity

0–10

"What did you do before the path was clear?" Senior engineers scope, decide, and adjust; less-experienced ones wait for someone to tell them what to do.

Good signal

  • Scoped the problem

    Broke an unclear task into questions, assumptions, a first bet.

  • Decided with incomplete info

    Picked a path and named the risk.

  • Adjusted on signal

    Changed direction when new info came in.

Bad signal

  • Waited for direction

    "I asked my manager what to do" — no independent thinking.

  • Oversimplified

    Collapsed the ambiguity by ignoring the hard parts.

  • False certainty

    Projected confidence they didn't have.

Common pitfalls

  • ·Framing "I asked for clarity" as the full answer — it's the setup, not the story.
  • ·Analysis paralysis dressed up as "being thorough".
›What the AI grader actually checks

9-10 = Scoped, chose a path, communicated reasoning, adjusted on signal. 7-8 = Competently but waited for clarity. 5-6 = Navigated with visible friction. 3-4 = Stalled or oversimplified. 0-2 = Collapsed under uncertainty.

04

Collaboration & Influence

0–10

"How did you move people who don't report to you?" Senior signal is earned influence — trust, evidence, well-framed arguments — not structural authority.

Good signal

  • Influenced without authority

    Moved a peer team or skeptical stakeholder via reasoning.

  • Named the other perspective

    Can articulate what the other person was worried about.

  • Adapted their approach

    Changed how they communicated based on the audience.

Bad signal

  • Escalated to win

    "I told my manager" — a decision by authority.

  • Parallel play

    Worked alongside others without actually collaborating.

  • Worked around people

    Avoided the hard conversation.

Common pitfalls

  • ·Confusing "got buy-in" with "told everyone the plan".
  • ·Over-crediting yourself for team-driven work.
›What the AI grader actually checks

9-10 = Moved others through clarity/trust/evidence without authority. 7-8 = Earned but narrow influence. 5-6 = Parallel play. 3-4 = Influence via authority/escalation only. 0-2 = No collaboration evidence.

05

Conflict & Feedback

0–10

"Did you engage, or did you avoid?" Strong candidates have had real disagreements, handled them directly, and walked away with intact relationships.

Good signal

  • Engaged directly

    Had the hard conversation.

  • Separated idea from person

    Disagreed strongly without making it personal.

  • Updated when warranted

    Changed their mind when the other side had a point.

Bad signal

  • Avoided

    "We agreed to disagree" with no actual engagement.

  • "Won" at cost

    Got their way, but burned a bridge.

  • No real disagreement

    Story is conflict-adjacent, not contentious.

Common pitfalls

  • ·Describing a disagreement where you were obviously right — unconvincing.
  • ·Confusing "being heard" with "being right".
›What the AI grader actually checks

9-10 = Engaged directly, separated idea from person, updated when warranted, relationship preserved. 7-8 = Slight over-accommodation or grudge. 5-6 = Avoided some of the hard conversation. 3-4 = Avoided or escalated damagingly. 0-2 = No real conflict or handled so poorly it's a red flag.

06

Impact & Outcomes

0–10

"What changed because of you specifically?" Activity ≠ impact. Interviewers want numbers, users, time saved — and a clear line back to your contribution.

Good signal

  • Concrete numbers

    "Cut latency 40%", "saved 12 hours a week".

  • Traced back to you

    Clear line between the work and the outcome.

  • Owned the follow-through

    Stuck around to measure and iterate.

Bad signal

  • Activity without outcome

    "We launched it" — no metric attached.

  • Fuzzy attribution

    Big outcome, unclear what you did.

  • Hypothetical impact

    "It would save X" — never measured.

Common pitfalls

  • ·Vanity metrics (commits) instead of outcome metrics.
  • ·Claiming impact for something you shipped but never measured.
›What the AI grader actually checks

9-10 = Concrete, measurable outcome tied to specific contribution. 7-8 = Clear outcome, mostly clear attribution. 5-6 = Fuzzy metrics or unclear attribution. 3-4 = Activity, not outcome. 0-2 = No outcome, or hypothetical. (Hard cap: no measurable outcome → max 4.)

07

Self-Awareness & Growth

0–10

"What are you actually bad at?" Credible self-awareness is specific, non-performative, and paired with evidence of actual change.

Good signal

  • Specific real weakness

    Genuine blindspot or gap.

  • Named what changed

    Concrete behavior change or habit shift.

  • Evidence it stuck

    Recent situation where the change showed.

Bad signal

  • Humblebrag weakness

    "I care too much".

  • Generic growth

    "I learned to communicate better" — no specifics.

  • Defensive posture

    Minimizes or justifies the weakness.

Common pitfalls

  • ·Picking a weakness that's obviously a strength.
  • ·Naming a weakness but showing no evidence it's changed.
›What the AI grader actually checks

9-10 = Names specific weakness, what they changed, how they know it stuck. 7-8 = Soft evidence. 5-6 = Surface-level or mildly defensive. 3-4 = Humblebrag or deflected. 0-2 = No self-awareness visible.

08

Judgment in Context

0–10

"How do you decide when it's hard?" Interviewers want to see you weigh real trade-offs, pick pragmatically, and own the reasoning — not retrofit a good outcome.

Good signal

  • Weighed real alternatives

    Names what they considered and rejected, and why.

  • Pragmatic over pure

    Picked the right call for the situation, not the textbook answer.

  • Would make the same call

    Can defend, not just rationalize.

Bad signal

  • Single rule applied reflexively

    "I always…" — with no weighing.

  • Post-hoc rationalization

    Suspiciously clean in hindsight.

  • Can't articulate why

    Knows what they chose, can't explain the trade-off.

Common pitfalls

  • ·Picking the "elegant" answer when the situation called for the ugly one — and pretending it was fine.
  • ·Describing the good outcome as evidence the decision was good.
›What the AI grader actually checks

9-10 = Weighed trade-offs, named rejected alternatives, chose pragmatically. 7-8 = Thin trade-off reasoning. 5-6 = One rule without weighing alternatives. 3-4 = Poor decision, or cannot articulate why. 0-2 = No judgment visible. (Hard cap: no rejected alternative → max 5.)

Overall rating thresholds

Insufficient

overall < 4.0

Developing

overall 4.0–6.4

Proficient

overall 6.5–8.4

Exceptional

overall ≥ 8.5

Your final rating is the mean of your per-criterion scores (0–10), rounded to one decimal.

Start a Behavioral Interview
Mode 06Open-ended architecture, no code

System Design

One open-ended design problem walked through the five phases of a real system design round: clarify, estimate, design, deep dive, then tradeoffs with cost and operations. You talk while you draw the architecture on a component canvas, and the interviewer sees the diagram and pushes back on it — and, once the core design stands, layers a generative-AI feature onto it the way 2026 rounds do. Ten criteria scored 0–10, including cost and operational maturity, which moved from bonus material to graded criteria this year. Guided mode defines the vocabulary as you go; Real mode gives you an unhelped score.

~35 min0–10 per criterion10 criteria
01

Requirements Gathering

0–10

"Did they ask before they designed?" Skipping this is the single most common way a system design round goes wrong — the constraint you never asked about is usually the one that decides the architecture.

Good signal

  • Asks before drawing

    Questions come first; boxes come after.

  • Functional AND non-functional

    What it does, plus scale, latency, consistency.

  • Finds the driving constraint

    Surfaces the 100:1 read/write ratio that justifies a cache.

Bad signal

  • Straight to architecture

    Starts naming components in the first reply.

  • Only functional questions

    Never asks how big or how fast.

  • Assumes the hard parts

    Invents requirements instead of asking.

Common pitfalls

  • ·Asking questions as a ritual and then designing as if the answers did not matter.
  • ·Never asking whether the system needs to be consistent or merely available.
›What the AI grader actually checks

9-10 = Asked before designing; elicited functional AND non-functional needs, including the constraint that drives the design (read/write ratio, latency, consistency). 7-8 = Asked good clarifying questions but missed one meaningful constraint. 5-6 = Asked a couple of surface questions, then moved on. 3-4 = Started designing almost immediately; requirements were assumed. 0-2 = Never asked anything; designed against an imagined problem. (Hard cap: designed before asking anything -> max 3.)

02

Scale Estimation

0–10

"Did they quantify, or hand-wave?" The numbers are not a ritual — they are what justifies the cache, the shard count, and the storage engine that follow.

Good signal

  • Shows the arithmetic

    users × actions ÷ 86,400 = writes per second.

  • Uses the number

    Ties the figure to a design decision they then make.

  • Sizes storage over time

    Per-record size × volume × years.

Bad signal

  • Qualitative only

    "It will be a lot of traffic."

  • Numbers with no consequence

    Computes QPS, then designs as if it were unknown.

  • False precision

    Six significant figures on a guess.

Common pitfalls

  • ·Treating estimation as a box to tick before the "real" design.
  • ·Refusing to estimate because the exact numbers are unknown — order-of-magnitude is the point.
›What the AI grader actually checks

9-10 = Quantified users, throughput, and storage with defensible arithmetic, and USED those numbers to justify a design decision. 7-8 = Produced sound numbers but only loosely connected them to the design. 5-6 = Rough hand-wave numbers, no derivation. 3-4 = Named quantities without arithmetic. 0-2 = Skipped estimation entirely. (Hard cap: no arithmetic anywhere -> max 3.)

03

High-Level Architecture

0–10

"Does the request actually flow?" Every box should earn its place and every box should be connected. The diagram is checked mechanically against the reference design, so a missing cache is a fact, not an opinion.

Good signal

  • End-to-end path

    A read and a write both traced through every box they touch.

  • Components earn their place

    Each box justified by a requirement or a number.

  • Nothing orphaned

    No box sitting on the canvas connected to nothing.

Bad signal

  • Boxes without flows

    Components drawn but never connected.

  • Missing an essential layer

    No cache on a heavily read-skewed workload.

  • Over-engineering

    A queue and workers bolted onto a key-value lookup.

Common pitfalls

  • ·Drawing an impressive diagram nobody can trace a request through.
  • ·Adding every component you know rather than the ones this problem needs.
›What the AI grader actually checks

9-10 = Clean end-to-end request path; every component earns its place; read and write paths both explained. 7-8 = Sound architecture with one gap or one unjustified component. 5-6 = Recognizable design missing a component the workload clearly requires. 3-4 = Components present but the data flow between them is unclear or wrong. 0-2 = No coherent architecture. (Hard cap: missing an essential component -> max 6.) (Hard cap: empty diagram -> max 4.)

04

Data Modeling & Storage

0–10

"Why that database?" SQL vs. NoSQL is an access-pattern decision. Naming a product is not an answer; matching the store to how the data is actually read is.

Good signal

  • Justified against access pattern

    Point lookups by key → key-value store, and says so.

  • Names keys and schema

    Primary key, index, and the query shape.

  • Addresses growth

    How it shards or partitions as data grows.

Bad signal

  • Product without reason

    "We would use Postgres" with no why.

  • Contradicts the access pattern

    Picks a store that fights the queries described.

  • Storage never addressed

    Data has to live somewhere.

Common pitfalls

  • ·Defaulting to whatever database you use at work without checking it fits.
  • ·Storing large blobs in the database instead of object storage.
›What the AI grader actually checks

9-10 = Chose a store and justified it against the access pattern; described the schema/keys; addressed how it grows. 7-8 = Reasonable storage choice with partial justification. 5-6 = Named a database without saying why that one. 3-4 = Storage choice contradicts the access pattern they described. 0-2 = Storage never addressed.

05

Scalability & Bottlenecks

0–10

"What breaks first, and at what number?" Asserting a design scales is worth nothing; naming the component that fails and the fix that addresses it is the whole signal.

Good signal

  • Names the real bottleneck

    The specific component that saturates first.

  • Mitigation fits the bottleneck

    Caching, sharding, or replication chosen for a reason.

  • Answers 10×

    Says concretely what changes an order of magnitude up.

Bad signal

  • "Add more servers"

    Generic horizontal scaling as a universal answer.

  • Asserts it scales

    No evidence, no limit named.

  • Misses the obvious limit

    Ignores the single coordination point in their own design.

Common pitfalls

  • ·Scaling the stateless tier — the easy part — and ignoring the stateful one.
  • ·Treating the cache as infinite.
›What the AI grader actually checks

9-10 = Identified the real bottleneck, proposed caching/sharding/replication that addresses it, and said what breaks at 10x. 7-8 = Named bottlenecks and a credible mitigation. 5-6 = Generic "we'd add more servers" scaling, or mitigations out of proportion to the stated scale. 3-4 = Claimed the design scales without evidence. 0-2 = No awareness of limits.

06

Tradeoff Reasoning

0–10

"Did they name the tension and pick a side?" Listing two options is not tradeoff reasoning. Naming what you give up, and defending it, is.

Good signal

  • Unprompted tension

    Raises the tradeoff before being asked.

  • Takes a side

    Chooses, and says what the choice costs.

  • Defends against the alternative

    Explains why the rejected option is worse here.

Bad signal

  • Options without a choice

    Lists alternatives and never decides.

  • One true answer

    Presents a choice as obviously correct.

  • No tradeoffs at all

    Design presented as free of cost.

Common pitfalls

  • ·Naming a textbook tradeoff that does not apply to this problem.
  • ·Choosing the "correct" option without acknowledging what it costs.
›What the AI grader actually checks

9-10 = Named a genuine tension unprompted, took a side, and defended it against the alternative they rejected. 7-8 = Named tradeoffs when asked and picked deliberately. 5-6 = Acknowledged alternatives without choosing between them. 3-4 = Presented one option as obviously correct. 0-2 = No tradeoff reasoning at all. (Hard cap: never named a tradeoff unprompted -> max 6.)

07

Cost & Right-Sizing

0–10

"Does the infrastructure fit the numbers?" This moved from bonus points to a graded criterion in 2026. Multi-region active-active for a small service no longer reads as ambition — it reads as not having done the arithmetic.

Good signal

  • Sizes to the estimate

    Every box is justified by a number they derived.

  • Names what they left out

    "No queue — there is nothing asynchronous here."

  • Can put a rough price on it

    Knows which line item dominates and why.

Bad signal

  • Machinery by habit

    Sharding at forty writes a second because sharding is what you do.

  • Scale for its own sake

    Designs for two orders of magnitude beyond the requirement.

  • Cost never mentioned

    Reasoning is entirely about capability, never about proportion.

Common pitfalls

  • ·Treating "it scales" as the goal rather than "it fits".
  • ·Adding a component to look thorough and being unable to say what it costs or what removes it.
›What the AI grader actually checks

9-10 = Sized every component to the numbers they derived, named what they deliberately left out because the scale did not warrant it, and could say roughly what the design costs to run. 7-8 = Proportionate design with one component that is larger than needed, acknowledged when pushed. 5-6 = Reasonable design but cost never entered the reasoning; sizing was by habit, not by the numbers. 3-4 = Added significant machinery the stated scale does not call for and defended it. 0-2 = Designed for a scale two or more orders of magnitude beyond the requirement without noticing. (Hard cap: infrastructure clearly disproportionate to the stated scale -> max 4.)

08

Operational Maturity

0–10

"How would you know it is broken?" A design nobody can operate is not finished. Monitoring, deploys and rollback are graded now, and the candidate who never says "monitoring" leaves points on the table in a real room.

Good signal

  • Names the paging metric

    The one number that means someone gets woken up.

  • Names the failure to survive

    What happens when the cache is gone, when the queue is lagging.

  • Ships and rolls back

    How a change reaches production and how it comes back out.

Bad signal

  • Done when drawn

    The design ends at the diagram.

  • Generic "we'd monitor it"

    No metric, no threshold, no consequence.

  • Stateful servers, stateless deploys

    Describes rolling restarts for servers that hold connections or windows.

Common pitfalls

  • ·Saying "add logging" and calling that operations.
  • ·Forgetting that the stateful component is the one that makes deploys hard.
›What the AI grader actually checks

9-10 = Unprompted, named the metric that pages someone, the failure the design must survive, and how a change ships and rolls back. 7-8 = Covered monitoring and failure handling well when asked; deploy or rollback was thin. 5-6 = Mentioned monitoring or a failure mode only when asked directly, in general terms. 3-4 = Treated the design as finished once it was drawn; operations came up only because the interviewer forced it. 0-2 = No awareness that the system has to be run. (Hard cap: never mentioned monitoring, alerting, deployment, or failure handling anywhere -> max 3.)

09

Communication & Structure

0–10

"Who is driving?" There is no prompt telling you what to do next in a design round. The candidate is expected to signpost where they are and move the session forward.

Good signal

  • Signposts the phase

    "Requirements are settled — let me size this."

  • Thinks out loud

    Reasoning is audible, not just conclusions.

  • Stays organized

    Holds one layer of abstraction at a time.

Bad signal

  • Waits to be asked

    Answers questions, never drives.

  • Jumps between layers

    Bounces from index design to load balancing and back.

  • Silent thinking

    Long gaps with no narration.

Common pitfalls

  • ·Narrating the design while never saying why any of it was chosen.
  • ·Going so deep on the first component that the round ends before the design exists.
›What the AI grader actually checks

9-10 = Drove the session, thought out loud, signposted where they were, stayed organized under an open-ended prompt. 7-8 = Clear and mostly self-directed; occasionally needed steering. 5-6 = Answered what was asked but never drove. 3-4 = Disorganized, jumped between layers, hard to follow. 0-2 = Could not structure a response.

10

Handling Pushback

0–10

"Does the position move with the argument?" Folding instantly and digging in blindly both score badly. The signal is whether a challenge is actually engaged with.

Good signal

  • Engages the objection

    Restates the challenge, then answers it.

  • Updates when out-argued

    Changes position and says what changed their mind.

  • Holds when right

    Defends a sound choice with a reason.

Bad signal

  • Instant fold

    Abandons a good design at the first question.

  • Digs in without argument

    Repeats the original claim louder.

  • Absorbs without engaging

    "Good point" and then continues unchanged.

Common pitfalls

  • ·Reading every question as a signal you got it wrong — interviewers probe strong answers too.
  • ·Changing the design three times in a row to chase approval.
›What the AI grader actually checks

9-10 = Engaged with challenges directly; changed position when the argument was better, held it when it wasn't, and said why either way. 7-8 = Responded substantively to pushback. 5-6 = Absorbed pushback without really engaging. 3-4 = Folded instantly on any challenge, or dug in without argument. 0-2 = Could not respond to challenges.

Overall rating thresholds

Insufficient

overall < 4.0

Developing

overall 4.0–6.4

Proficient

overall 6.5–8.4

Exceptional

overall ≥ 8.5

Your final rating is the mean of your per-criterion scores (0–10), rounded to one decimal.

Start a System Design Interview

The canonical rubric text on each criterion mirrors the exact anchors used by the AI grader in production. If the grader changes, this page changes.