HackerRank's Chakra Scores a Silent Fix as 0. Narrate Every AI Prompt

If you're heading into a HackerRank Chakra interview, the good decisions you make silently in the editor probably won't count. Chakra grades the interview transcript, not your keystrokes, so a smart prompt or a sharp rejection of AI output that you never say out loud leaves no evidence, and the grader can't score it.

Chakra became generally available on Oct. 5, 2026, after a six-month beta that ran more than 500,000 interviews, with Snowflake, Snorkel and Capgemini as early customers. HackerRank CEO Vivek Ravisankar says it collapses the screen, the take-home and the engineer round into a single sitting. You're expected to use the built-in AI assistant. The fix for the transcript problem is mechanical: narrate what you do with it.

How does the HackerRank Chakra AI interview score you?

Chakra scores each expectation in a structured interview plan on a raw 0 to 3 scale, and every score has to trace back to a specific, verbatim moment in the interview transcript. Raw scores are then normalized to a 0 to 5 scale for the report the recruiter reads.

HackerRank laid out the mechanics in a July 24, 2026 post by Danielle Bechtel, "How Chakra Scores an Interview." Three agents split the work:

1. The planning agent works with the recruiter before you ever join, turning the job description into a structured interview plan.
2. The interviewer agent runs the live voice interview with you.
3. The Reporter Agent reads the full transcript against the plan and scores each expectation independently.

Here is the raw scale the Reporter Agent uses:

The key line in HackerRank's post: nothing in the report is inferred beyond what's in the transcript. The editor is where you work. The transcript is where you get graded.

Because the plan is built from the job description, the posting is your study guide. Every requirement listed there is a candidate for an expectation you'll be scored on. Before the session, I'd run the posting through Refolk's fit score. It compares the posting against your own history and shows which requirements you can back with real work and which you can't. The second list is what you need to rehearse explaining out loud.

What does the HackerRank AI fluency score measure?

AI fluency is how well you frame a problem for an AI assistant, judge what it gives back, and steer it toward a working solution. That's TechCrunch's summary from the launch, and HackerRank's own candidate support page breaks it into three observable behaviors.

According to HackerRank's "Evaluation in Chakra" page, the interviewer looks at:

• Framing: how clearly you state the task and its constraints.
• Guidance: how deliberately you direct the AI assistant.
• Review: how carefully you check, test and correct its output.

The same page includes the sentence that should change how you prepare: a strong response in one area does not compensate for a weaker response in another. Clean, correct code doesn't cover a blank on review. A great opening prompt doesn't cover a missing test.

The interviewer also pushes. Per TechCrunch, typical follow-ups ask why you chose one approach over another, or how your solution would change if a new constraint came in. Those questions pull your reasoning into the transcript. If you've been narrating all along, they're easy. If you've been silent, they're the first time the grader hears you think.

Why a silent good decision scores the same as a missed one

A decision you make silently produces the same transcript as a decision you never made. Since the Reporter Agent only reads the transcript, it can't tell the two apart.

Take two candidates. The assistant hands both of them a function that breaks on an empty input.

• Candidate A spots it, quietly adds a guard clause, and moves on.
• Candidate B never notices. A later test happens to cover it and they patch it without comment.

The code ends up the same. Neither transcript says anything about the empty input. On the "review and correct AI output" expectation, neither has a verbatim moment to point to. Candidate A did the exact thing the rubric rewards and got no credit for it.

Now take Candidate C, who says: "The assistant's version assumes the list is non-empty. That will throw on the first call in our case, so I'm adding a guard and a test for it." That's one sentence, and it's a citable moment for framing, review and correction all at once.

This is why narration isn't a soft skill here. Given how the scorer is built, it's how you put evidence on the record.

Is "Not Assessed" a penalty in Chakra?

Assume it is. HackerRank's raw scale puts Not Assessed at 0, below Not Met at 1, and the published docs don't say how a 0 is handled when scores are normalized to the 0 to 5 report.

The sources disagree. A write-up on byteiota describes skills you don't demonstrate as marked "Not Assessed" rather than penalized. That may be how a recruiter reads the label, but the raw number sits below an outright miss, and HackerRank's documentation doesn't settle it.

Two things hold either way:

• Criteria are scored independently. A blank on AI fluency stays blank. Your algorithm score can't fill it.
• A recruiter skimming the report sees a gap. Whether it's labeled "0" or "Not Assessed," it doesn't read as a strength next to candidates who put that expectation on the record.

So don't treat "Not Assessed" as neutral. Your expectation went unscored, and the next candidate's didn't.

Only 0.47% of US engineers list prompt engineering. Chakra checks anyway

Almost no software engineers claim prompt engineering on their profiles, and Chakra doesn't read that claim anyway. It scores what you do in the session, which makes the gap an opening for anyone who can show the skill out loud.

India has roughly 1.9 times as many Software Engineers listing the skill as the US, with Bengaluru the largest cluster in the profiles Refolk looked at. In the US, Google was the most common employer among those engineers and the Bay Area the top region. Those employer and region splits come from just 25 sampled profiles, so treat them as directional.

Why it matters for Chakra: the skill is barely claimed, and the interview ignores the claim. If you've used AI assistants at work but never wrote "prompt engineering" on your resume, you aren't behind. Your spoken reasoning in the session is the evidence.

The resume still has to get you the Chakra link, though. If your AI-assisted work is real, it should appear as something you did ("cut review time on a migration by having an assistant draft tests, then rewrote the ones that missed edge cases"), not as a keyword. That's the kind of bullet Refolk writes from your own history and then tailors to each posting, so the recruiter's plan and your resume use the same language.

How to prepare for an AI interviewer coding session

Prepare by practicing one loop until it's automatic: say what you're about to ask the assistant and why, then say whether you're keeping its answer and why. Each pass through that loop gives the Reporter Agent a citable moment for framing, guidance or review.

The narration loop

1. Frame before you prompt. Restate the problem and its constraints in your own words. "Input can be up to a million rows, memory matters more than speed, and the IDs aren't guaranteed unique."
2. Announce the prompt and its purpose. "I'm asking the assistant for a streaming version, not a full load, because of the memory constraint I just mentioned."
3. Read the output aloud in summary. "It's using a dict keyed on ID. That breaks on duplicates."
4. Give a verdict with a reason. Accept, reject or modify, and say which and why. "Rejecting the dict. Switching to a counter so duplicates survive."
5. Test, and say what the test proves. "This test covers the duplicate case the first version would have dropped."
6. Absorb the follow-up out loud. When the interviewer adds a constraint, say what changes and what stays. "If the data arrives sorted, I can drop the counter and do a single pass. I'll keep the duplicate test either way."

Phrases that cost you

A secondary write-up of a sample Chakra report (from Progressive Robot; the original report isn't public) describes a candidate marked down for "passive vibe-coding" and for saying "I trusted the AI." Vibe-coding means accepting AI output without reviewing or steering it. Avoid:

• "I'll just go with what it gave me."
• "Looks fine." (Fine how? Against which case?)
• Long silent stretches while code appears in the editor.

A drill that actually works

Do two or three practice problems with any coding assistant while recording yourself. Then read the transcript, not the code. For every place you changed or kept AI output, look for a sentence that explains it. Each gap is a moment Chakra would have scored as Not Assessed. Keep drilling until a stranger reading only the transcript could rebuild your decisions.

Why the session is your only chance to put evidence on the record

The Chakra session is likely your only real contact with the company before a decision, and the report it produces goes to the employer, not to you. Whatever you don't say during the session doesn't reach anyone.

Greenhouse's Oct. 21 report found that 81% of US candidates had been rejected from a job in the past year without speaking to anyone at the company, human or AI. The same research found 48% had taken an AI interview in that time, and 70% of those said it let them show their skills well. An AI interview is at least a conversation, which beats a silent reject.

What happens afterward is less encouraging. Greenhouse's May 2026 data on candidates who completed an AI interview:

• 28% moved forward
• 13% were formally rejected
• 51% never heard back

A PracHub prep guide adds that the "candidate report" is a report about you for the employer's review, with no guarantee that you get a copy. So you probably won't see your score, can't dispute a blank, and may not hear anything. Snowflake candidates are already reporting Chakra screens for 2027 internships, so early-career pipelines are being screened this way too. Treat the session as the full record, because it probably is.

FAQ

Can I use AI in a HackerRank Chakra interview?

Yes. Chakra includes a built-in AI assistant and grades AI fluency, so skipping it to look "pure" leaves those expectations with nothing to score. Stick to the built-in assistant, since it's the one that shows up in the transcript. In the beta, suspicious-activity flags ran 70% to 80% lower than in comparable traditional HackerRank assessments, varying by geography and seniority, on the reasoning that open AI use removes the incentive to secretly run external tools. Analysts note that figure measures flags, not proven reductions in cheating.

Does Chakra watch what I type in the editor?

HackerRank says every score traces to a verbatim moment in the interview transcript, and nothing in the report is inferred beyond it. Your code matters because you'll explain it and the interviewer will ask about it. A decision you make silently in the editor, like quietly fixing a bug the AI introduced, has no transcript moment, so the scorer can't credit it.

Will I get to see my Chakra score?

Don't count on it. The report is written for the employer's review, and prep guides note there's no guarantee candidates receive a copy. Raw scores run from 0 (Not Assessed) to 3 (Met) and are normalized to 0 to 5 for the recruiter. Since you likely can't review or contest it, the only lever you have is what you say during the session.

What if I need quiet time to think during the interview?

Short pauses are fine. Just bookend them. Say what you're about to think through ("I want to check whether this handles the sorted case"), then say what you concluded. That turns a silent stretch into two citable moments. Long, unexplained silence while code fills the editor is how candidates end up looking like passive vibe-coders.