What is the Interview Guide?

Overview

The Interview Guide is a feature that prepares the questions to ask in a deep-dive interview, organized by evaluation criterion, based on a candidate's AI Collaboration format submission.

Before a deep-dive interview, the interviewer needs to read through the submitted code and the candidate's exchanges with the AI agent and decide where to dig in. This preparation tends to depend on each interviewer's experience, so the quality of the interview changes depending on who conducts it.

With the Interview Guide, you provide an Evaluation Criteria Sheet that describes "what you want to measure," and for each criterion, questions are generated that are tied to the code the candidate actually wrote and to the candidate's exchanges with the AI. Interviewers choose which of the prepared questions to use, take notes during the interview, and can export what they prepared as Markdown into their own documents.

• The Interview Guide is available for the AI Collaboration format only. It cannot be used with the Coding Format, Practical Format, Quiz Format, or Live Coding.
• It can be generated after the submission has been scored.
• Questions, answer guidelines, model answers, and notes are shown to your company only and do not appear on any screen shared with the candidate.

Available Plans

Whether the Interview Guide is available to you depends on your contract. For availability, please contact your Customer Success representative.

How to Use It

1. Create an Evaluation Criteria Sheet (once, at the start)

In Company Settings, under "Evaluation Criteria Sheets," write in Markdown what your company wants to measure. You can create as many sheets as you need, for example one per role or hiring track.

An Interview Guide cannot be generated without an Evaluation Criteria Sheet. Please set this up first.

2. Generate the Interview Guide

Generate it from the "Interview Guide" section on the Playback tab in the candidate details. In the dialog that appears, choose which criteria to provide from the following two options.

Generation takes a few minutes. When it is complete, a link to the Interview Guide appears in the same section.

3. Narrow down questions and take notes in the Interview Guide

The generated Interview Guide opens in a separate tab. The list of evaluation criteria is on the left, and the questions are listed by criterion on the right.

After the interview, the notes for each question are saved linked to the evaluation criterion. People who did not attend the interview can also read the "evaluation criterion + what you wanted to confirm + notes" together as a set.

How to Write an Evaluation Perspective Sheet

Sheets are written in Markdown, using the following format.

• A heading (##) becomes an evaluation criterion
• The body text under the heading becomes the description of that criterion
• Bullet lines that start like - Excellent: become the Answer Guideline (how to read the answers)

# Evaluation Criteria for Mid-Career Backend Hiring

## Recovery design for failures
Check whether the candidate can explain the reasoning behind their retry and timeout design.
- Excellent: Works backward from the target convergence time and explains even the initial interval and multiplier.
- Meets expectations: Can explain the intent, but admits the numbers themselves are rules of thumb.
- Below expectations: Stops at "these are commonly used values."
- Insufficient: Answers that they used the values the AI produced as is.

About the number of levels

You can freely decide the number of levels to match your company's grading definitions. Four levels are recommended (Excellent / Meets expectations / Below expectations / Insufficient), but if you write five levels they are treated as five, and the level names are displayed exactly as you wrote them (A / B / C or Lv4 to Lv1 are also fine).

• Levels are handled in the range of 2 to 6 levels. A single level is treated as "no levels," and if there are 7 or more levels, the top 6 are used
• You do not need to write every level. Levels you do not write are filled in by the AI from the submission at generation time
• You can also include criteria that have only a description and no levels at all

Limitations

• If there is not a single heading (##), no criteria are created, so the guide cannot be generated. Please refer to the sample on the sheet editing screen in Company Settings
• The sheet body can be up to 65,536 bytes

What the Generated Questions Contain

Each question has the following elements.

The three sources

"Constraint change" questions check whether constraints and design decisions are connected in the candidate's mind, and whether the candidate can make the same kind of judgment when conditions change. Because they cannot be answered with a prepared answer, they let you measure the candidate's understanding on the spot.

"Not confirmed" questions ask about criteria for which no evidence was found in the submission. The fact that a relevant behavior was not observed is not treated as evidence that the candidate "cannot" do it. It is treated as something to confirm in the interview.

Criteria for Which No Questions Are Created

For criteria that cannot be confirmed from the submission, such as personality or past management experience, questions are not forced. In this case, the criterion remains in the guide with the message "This criterion cannot be probed in a follow-up interview."

The criterion remains instead of disappearing because it is a clue for reviewing how the Evaluation Criteria Sheet is written.

Selecting Questions and Taking Notes

• "Use in interview" is on by default. You remove the questions you don't need with "Do not use in interview"
• Removed questions are only hidden, not deleted. You can bring them back with "Show all" at the top of the screen
• Notes are saved linked to the criterion and are included as is in the Markdown export
• The generated text (question text, what to confirm, answer guidelines, and model answers) cannot be edited. What the app keeps is your "Use in interview" selections, their order, and your notes

Viewing Alongside the Submission

Click "View with Submission" in the toolbar to show the submission on the left and the Interview Guide on the right of the same screen. Clicking an evidence link attached to a question switches to the same view. Click "Show Interview Guide Only" to return.

Exporting as Markdown

Click "Copy Markdown" in the toolbar to copy to the clipboard a Markdown version of the questions kept under "Use in interview," grouped by evaluation criterion. Paste it into the document you use to run the interview.

Sharing the Submission with the Candidate

If there are several days between submission and the interview, the candidate may not remember the code they wrote or the instructions they gave the AI. Click "Share Submission" in the toolbar to copy the URL of a screen where the candidate can reread their own submission. Give it to the candidate at any time you choose.

Only the submission is shared. The Interview Guide's questions, answer guidelines, model answers, and notes are not included.

Regenerating the Interview Guide

You can regenerate it from the same section. When you regenerate it, your previous "Use in interview" selections and notes are lost. Export anything you need as Markdown beforehand.

Regenerating does not change the submission's score.

Related Pages

• Understanding the AI Collaboration Format
• AI Collaboration Format Reports
• September 2026 | Added "Editor Mode" to AI Collaboration Format Playback to Replay the Candidate's Editor Activity