How grading works

A Placed report is written by an AI model that reads your whole session against a fixed rubric. This page says what the numbers mean, what the grader is told to do, and what happens to your session afterwards.

The scale

Each criterion gets a whole number from 0 to 10. The overall score is the average of the criteria, to one decimal place.

The rules the grader follows

The grader is told to be accurate in both directions: a real mistake costs points, a small one costs little. These rules are part of its instructions on every session:

• Each criterion is scored from what actually happened in your session, not from a fixed starting point.
• Deductions match how much an issue matters. A minor one, like a weak variable name or one unclear sentence, costs at most a point, and all the minor issues on a criterion together cost at most a point.
• A real gap, like a missed edge case or never stating the complexity, costs 2 to 3 points. A wrong approach, code that does not work, or not being able to explain your own work costs more.
• The grader scores what happened in the session, not how well you described what you would do. An answer that only sounds good scores 5 at most.
• A 7 or higher needs something it can point to in the transcript or the code. Something the interviewer had to pull out of you counts for less than something you offered.
• Every explanation has to quote the moment it is based on. Those quotes are the "receipts" on a Pro report.

A clean session scores 8 or 9 on most criteria, and a 10 is possible. Skipping a step, like testing your code or asking about the input, costs more than a small slip in how you said something.

The four bands

The band on your report comes from the overall score. Placed works it out from the numbers, so the model cannot give you a kinder label than your scores support.

The line about your level

If you have said what you are interviewing for, the report also reads your score against a bar for that kind of loop. The rubric scores do not change with level; only this comparison does.

These bars are Placed's own estimates. No company publishes a cut-off, and a real loop is decided by people who weigh more than one number.

What each round is graded on

The full rubric has what a strong and a weak answer look like for every criterion, and the sample reports show a finished one for each round.

Which model does the grading

The interviewer and the grader are Claude models from Anthropic. The grader reads the full conversation and, in coding rounds, your final code. Reports on the Pro plan are graded by a larger Claude model than reports on the free plan; the rubric and the rules are the same for both.

It is still a model, and it can misread a session. If a score looks wrong, the receipts on a Pro report show what it was based on: each one is a quote from your transcript. Free reports show the scores and the top two things to fix.

What is stored

• Audio from your microphone is sent to ElevenLabs to be turned into text. Placed keeps the text, not the recording.
• Signed in, your scores, reports, code and chat transcripts are saved to your account so they show up in History.
• As a guest, the session stays in your browser until you create an account.
• Your conversation and code are sent to Anthropic to run the interviewer and the grader, and your code is sent to Judge0 to run it.
• You can ask for your account and everything in it to be deleted from Settings, or by emailing support@placed.dev.

The privacy policy has the full list of services that handle your data.