Creating a Posting

To create a new posting, click New Posting from your dashboard. A posting defines everything about how candidates will be evaluated for a role.

Quick Setup

Assessment Types

Competencies and Weights

Competencies are the skills and behaviors you're evaluating. Each competency has a weight slider that controls how much it contributes to the overall score.

You can build your competency set in any combination:

• Select from the platform library (behavioral, technical, situational)
• Apply a competency profile to auto-fill a standard set (e.g., "General Behavioral")
• Create custom competencies specific to your organization
• Generate with AI based on the job description and optional context

Viewing Behavioral Anchors

When you pick a competency from the in-house library, click View anchors on that selected competency to see its 1–5 behavioral anchor descriptions — exactly what a score of 1, 2, 3, 4, or 5 means for that competency. This is the same rubric the AI scorer uses, so it's worth a quick read before you commit to a competency: if the level descriptions don't match what you actually want to measure, pick a different competency or create a custom one.

Interview Configuration

For Conversational or Both assessment types, configure how the interview runs:

Practical Assessment Setup

For Practical or Both assessment types:

Each problem opens in the workbench that fits its task:

When you review a candidate, the answer view shows the work as they left it: code with its test results, the spreadsheet, the emails sent, or the drawing.

Work that is run against tests is never scored without its results. If the tests could not be run, the candidate's page shows Scoring Failed with the reason, and Retry Scoring scores the work once they can run again.

Practicals in Steps

A practical can continue in steps that build on each other, so you see how a candidate carries their own work forward. Under the practical's task, add steps in the Steps section (when your platform administrator has switched chained practicals on):

Draft with AI (beside Steps) writes the steps for you from the job title, the job description, the stage's skills and the practical's task: what each step asks for, its workbench and its time. They replace the steps already there, and you can edit every one. For a library practical, each new problem and hand-off is a task from the same library (the TSAI library or My Library), picked to continue the work.

Each step has its own time limit; the platform limits how many steps and how many minutes a practical may have. Candidates hand in each step and can see their earlier steps as they go. On a hand-off, their work arrives inside the step's workbench in its own form, labelled with its step: a spreadsheet becomes a CSV file they can attach to an email, a file in a code project, a new sheet, a section of a document, or notes on a diagram. Any workbench takes in the work of any workbench, its own kind included. The whole chain is one assessment with one score: the scorer reads every step's work together, including how they changed their own earlier work.

The Interviewer on a Practical

A practical stage can have the AI interviewer talk with candidates about their work, inside the workspace, in text or voice (when your platform administrator has switched it on). In the stage's Interviewer section:

Once a conversation is on, What the interviewer can see chooses how much of the candidate's work it reads, up to the most your platform allows:

Wherever the interviewer sees the work, it also sees their last run (on a workbench that runs tests), and with questions while they work a run can prompt one. Seeing nothing, it never learns how a run went, so a run never prompts a question. A debugging task's broken work is always shown to the interviewer: it is the task. Candidates are told what the interviewer can see in its panel beside their work.

Below the options, the editor estimates what the conversations you chose use per candidate, in interview units and voice minutes or text messages, from your pack's rates.

What candidates say is scored together with their work, in the practical's one score. Questions while they work come from each candidate's own work, so candidates are asked different things: the scorer reads only what the answers show, never how many questions there were, and an unanswered or dropped question is not held against anyone. Each skill gets a score for the work and, when the conversations showed something about it, one for the candidate's reasoning; the scorecard shows both (Work 3.0, Reasoning 5.0) beside the combined score. The work counts most, and the conversation can move a score only a little, so talk never makes up for work that is missing. On a candidate's Scorecard tab, the practical's Assessment button shows how many steps and conversations it has; it opens every step's task and answer with the conversations about it (while they worked, after the step, after the practical), what prompted each question, and how each ended (answered, skipped, put off, a dropped call, or not finished), and, for a question while they worked, Their work when this was asked. With a conversation after the practical, the practical is scored when it ends; if a candidate leaves without finishing it, the work is scored on its own a little later. The interviewer never sees the scoring criteria or the answers.

Debugging Tasks

A practical can be a debugging task instead of a build task (when your platform administrator allows them): the candidate is shown something already broken in a workbench (a workbook whose total is wrong, an email that went to the wrong person, code that double counts, a query that drops customers, a diagram with a single point of failure, a document that contradicts itself), read only, and talks the AI interviewer through it: what they see, what is wrong and why, how they would fix it, and how they would know the fix worked. The conversation is the work, so it is always on, in text or voice.

• From the library: filter the task picker by Debugging task. Each comes with its planted problems.
• AI create: choose Task: Debugging task, then how many Problems to plant (up to your platform's limit), How subtle, and whether to Include a red herring (something correct that looks suspicious). Each candidate gets their own. For code and SQL, a generated task is used only if its broken version fails enough of its own tests; every generated task is reviewed so nothing gives a problem away.
• My Library: write your own: choose Debugging task in the task editor, put the broken work in what the candidate starts with, and list The problems you planted: the kind of problem, where it is (named the way a candidate would select it, for example "E7 on Summary" or "line 12 of orders.js"), what goes wrong, why, the right fix, and what a good explanation shows.

In the stage's Interviewer section, a debugging task adds: text or voice, and whether to Tell candidates how many problems there are. The conversation starts as soon as the task opens, beside the work. Candidates show the interviewer what they mean by selecting it in the work (a cell, a line, an email, a shape, a passage, or in a robotics run a stretch of a trace or the moment a paused run is at). In text, the selection goes with their next message, and the transcript you review shows it over their words; by voice, the interviewer looks at what they last selected.

The planted problems are never shown to the candidate or the interviewer. On the candidate's Scorecard, the practical's Assessment button shows, for each planted problem, whether they found it, explained why, proposed the fix and said how to check it, with their own words, beside what was planted; then what they selected by voice.

A "now fix it" step can follow a debugging task: add an Update what they made step, and it opens on the same work, editable.

An AI Assistant on a Practical

When your platform administrator allows it, a practical stage's Interviewer section has Give the candidate an AI assistant. Candidates can then chat with an AI assistant about their task while they work, on any workbench, the way they would use one on the job: it explains, checks their work, and writes formulas, code, queries or drafts when asked. It sees only the task and their own work as it is right now (and their earlier steps, in a practical in steps), helps only with this task, and the number of messages a candidate can send over the practical is set by your platform. You don't choose or see which AI model it runs on.

What they ask it, and what it writes, is scored with the work: the scorer judges how they used it in each workbench (for example, whether they checked what it gave them). On the candidate's Scorecard, the practical's Assessment button shows the conversation with the assistant for each step, marked AI assistant. The editor's estimate counts the assistant's messages against your pack balance.

Application Form

Attach a custom application form to your posting so candidates can apply at a public URL before being interviewed.

When a candidate submits the application form, they are added to your candidate list automatically and receive a confirmation email.

Hiring Manager Notes & Document Upload (AI drafting)

When you first create a posting, the setup modal accepts extra context the AI uses to draft your job description and suggest competencies:

• Hiring Manager Notes — free-text insights about the team, role expectations, or what matters most for this role
• Document Upload — PDFs, DOCs, or TXT files with additional context (a role description, team charter, or existing competency framework)

The job description editor itself is a plain text area — paste or edit your JD directly, or generate a draft with AI and refine it.

Snapshotting: Editing a Skill Later Won't Change This Posting

When you save a posting, each selected competency's anchors, category, and description are frozen onto the posting at that moment. If you later edit that competency in Settings > Competencies, the change applies to new postings — it does not alter a posting that already used the competency.

Publishing vs. Drafts