How CodeGround assesses interviews

A report is practice feedback written the way an interviewer writes up a loop: a recommendation for your target level, a score for each dimension the format tests, and the moments that justify each score.

See the sample report.

During the interview, the room records the evidence the evaluator reads afterwards:

• A timestamped transcript of both speakers.
• Snapshots of your code at every test run and periodically when it changes, plus the final code.
• Every test run, with failures and errors.
• Interviewer actions, such as highlighting lines in your code or annotating the whiteboard.
• A text summary of your whiteboard, your notes, and, in AI-assisted rounds, your conversation with the assistant.

Raw interview audio is not stored by default. See the Privacy Policy for retention.

Each format has its own versioned rubric. Every dimension has written anchors for each score, and a weight that sets how much it contributes to the overall number.

Each dimension is scored from 1 to 4, always relative to the level you chose, from new grad to staff. A hint on a hard problem is normal for a new grad and a significant negative for a senior candidate.

4/4

Above bar

Independent, efficient, and clearly beyond expectations.

3/4

At bar

What a solid candidate at this level does.

2/4

Below bar

Partial, only after prompting, or inconsistent.

1/4

Clear gap

The core of the dimension was missing or wrong for the level.

Every dimension score has to cite moments from the session: what you said, a line of your code, or a message to the AI assistant, each with a timestamp and a note on why it matters.

Quotes are checked on the server before the report is saved. A quote is kept only if it appears word for word in the transcript turn, code, or assistant message it cites; anything that cannot be matched is discarded rather than shown to you.

For coding problems, your final code is re-run on the server against the full test suite, including hidden tests you never saw. That verdict, not what anyone said during the interview, decides correctness. Passing tests is not treated as proof of quality: the actual complexity of your code is compared with the problem's constraints.

Hints are recorded with their significance: a minor nudge toward an area, a moderate hint that narrowed the solution space, or a major hint that supplied the key insight or fix. How much a hint costs depends on your target level.

The evaluator makes a holistic recommendation on a six-step scale, from Strong no hire to Strong hire, with a confidence of low, medium, or high depending on how much signal the session produced.

The 0–100 score is the weighted average of the dimension scores, kept inside the range that matches the recommendation so the number never contradicts the judgment. Scores map to bands:

Excellent

85–100

Strong

70–84

Developing

50–69

Needs work

0–49

If a session ends before there is enough of your work to judge, no report is produced, and you are told why.

Reports are written by an AI evaluator. It can be wrong, and companies use different rubrics, levels, and interview conditions. Read the quoted evidence, and disregard any conclusion it does not support.