Evaluator already writes and grades the test. The education edition hands you the controls.

A hiring funnel needs none of this. A classroom needs all four: how hard the questions are, which sections are on the test, what a student sees of their own grade, and how much the AI is willing to help.

Start freeJoin a class
10 free cycles to try itNo card requiredWhat a cycle costs
AssignmentPeriod 4, Intro CS
12 questions

Loops, and the off-by-one that eats them

Difficulty
High school, level 3 of 5Sets the vocabulary and how hard the problem is for that audience.
Sections
Code reading, debugging, code writingCommunication, tradeoffs and AI collaboration are off for this one.
Students see
Correctness onlyScore hidden, written feedback hidden. You still see all three.
AI help
Hints onlyPoints at where to look. Names the line or the idea to reconsider, and does not explain why or write code.

Every prompt this class sends the assistant is recorded and lands on your desk with their work.

The four controls.

Set per assignment, before it is generated. Nothing here is a preference buried in an account setting.

  1. 01

    How hard it is.

    Two axes. A grade band (middle school, high school, intro college, advanced) sets the vocabulary and the background the questions may assume. A 1 to 5 level sets how hard the problem is for that audience.

    One axis without the other is the reason a difficulty slider never works. Level 4 means something different to a ninth grader than it does to a sophomore, and the band is what tells the generator which one it is writing for.

    High school, level 2

    Trace three calls of this factorial by hand and write down what each one returns.

    Intro college, level 4

    This recursive flatten blows the stack on deeply nested input. Find the base case that never fires and fix it.

  2. 02

    Which sections are on it.

    Six sections, six checkboxes: code reading, code writing, debugging, communication, tradeoffs, and AI collaboration. Each one is generated separately, so an excluded section is absent rather than thinned out.

    You spent the week on debugging, so turn the other five off. What comes back is all debugging, at the level you set, with none of it surveying things you have not taught yet.

    Friday debugging drill

    • Code readingOff
    • Code writingOff
    • DebuggingOn
    • CommunicationOff
    • TradeoffsOff
    • AI collaborationOff

    One section on. What comes back is a debugging assignment, not a shortened version of everything.

  3. 03

    What the student sees.

    Three independent switches: the score, per question correctness, and the written feedback. The grader always runs, so you have the class picture whichever way they are set.

    Correctness on with the score and the feedback off gives a student a list of what they got wrong to work from, and no number to argue about at the end of class.

    What this class gets back

    • ScoreThe number out of 100.Hidden
    • CorrectnessRight or wrong, per question.Shown
    • FeedbackWhat the grader wrote.Hidden

    All three off is the ungraded practice setting. They submit, you see everything, they see nothing but a confirmation.

04

How helpful the AI is.

This is the control the hiring product would never ship. Its assistant is told not to refuse and to write complete solutions if asked, because collaborating with an AI is the thing a hiring assessment measures. A classroom often wants the opposite, and wants to pick where on that scale each assignment sits.

Same class, same week, two settings. Monday's exercise runs at hints only, so a stuck student is told which line to re-read and nothing more. Friday's runs at full assistant, because working with the thing is what you are teaching that day.

The level lives on the assignment and is read server-side on every request, so a student who edits what their browser sends gets the level you set.

Off

No assistant. Students work without AI help of any kind.

Hints only

Points at where to look. Names the line or the idea to reconsider, and does not explain why or write code.

Socratic

Explains the underlying concept and asks questions back. Does not apply the concept to this task or give the answer.

Full assistant

Behaves like a normal coding assistant. It will write the answer if asked. Use this when working with AI is the skill you are teaching.

From a sentence to a graded class, in one sitting.

  1. 01

    Make a class. Share the join code.

    Every class gets an eight character code with no lookalike letters in it, because it gets read aloud. Students type it once and they are on your roster.

  2. 02

    Build an assignment.

    Say what you are teaching in a sentence and set the four controls. The questions come back as a draft, so you read them before your students do.

  3. 03

    Read the results.

    Submissions are graded per question and per section. You get the class view and the individual view, plus the full assistant transcript for anyone who used it.

The AI level is an instruction, not a wall.

Each level is a strong, specific instruction to the model about what it may and may not do. It is the same kind of instruction every AI product runs on, and it has the same weakness. A student who is determined to argue with a hints-only assistant can push it past hints.

So the level is not the accountability. The transcript is. Every message a student sends the assistant and every reply they get is recorded on every level, stamped with the level that was in force, and shown to you next to their work. If someone talked their way into an answer, it is in the log with their name on it.

Off is the one level that works differently. There is no assistant panel and no request is ever sent, so there is nothing there to argue with.

Set the four controls once and see what your class actually does.

Ten free cycles to try it, with every control switched on. Make a class, write one assignment, and read what comes back.

Start freeSee pricing