Read an AI review summary
Cet article n'est pas encore traduit : il s'affiche en anglais.
What the summary contains, how to check its evidence, what the model looked at, and when to regenerate it.
The AI summary card sits at the top of a finished attempt's review page, above the challenges. It's marked Advisory: read it before the code, not instead of it. It has no score, no pass or fail and no recommendation. You rate and decide.
Run or regenerate it
If no summary exists yet, choose Summarise. It takes about a minute. Owners and editors can run it; viewers can read it.
When the attempt changes after a summary, for example a late recording or a new save, the card says so. Choose Regenerate to include it. You can pick a different model next to the button first.
What's in it
- Overview: a few plain sentences about the whole attempt.
- Per challenge, Outcome, How they worked and, when the candidate used one, AI agent use.
- Strengths and Concerns, each with evidence.
- Integrity notes: tab switches and pastes stated as facts with times, never as proof of anything.
- Interview questions: three to six questions that probe this attempt.
- What the summary couldn't see: anything that was missing or shortened.
Check the evidence
Each finding ends with chips such as src/cart.ts:42, terminal, agent chat or activity. Click one and the review page opens that file at that line, or that recording. A chip that points at something that wasn't recorded is greyed out instead of clickable.
What the model looked at
Before the call, Kendor runs the challenge's tests on the final code, when the challenge has a test command. The model then gets the brief, the candidate's changes against the starter, the test results, the terminal, the AI agent chat, integrity events and written answers.
To check a claim, the model can list, read and search the starter or the final code, and read a terminal or the agent chat in full. These tools only read. It's one model call, not an agent: it can't run code or reach the network. What the AI checked at the bottom of the card lists every lookup it made, and the ones in the final code or a recording are links.
If an attempt is too large to send whole, parts are shortened and the card says so.
Live coding sessions
A live coding session is summarised the same way. Because the code, terminals and agent were shared with the interviewers, the summary credits work to the candidate only where the evidence supports it, and says when it can't tell who did what.