Skip to content
Crossing Hurdles · Pre-Assessment Guide

Know every step before you open the assessment.

The live qualification assessment has five guided steps and a pass bar of 75 out of 100 points. You may consult an LLM for reference in another window. You may not paste into the studio: every answer is typed by you, in your own words.

  • 5 Guided Steps
  • Pass Bar: 75 / 100 Points
  • LLM Reference Allowed · Copy-Paste Disabled
  • Instant Workspace & 7-Step Stripe Onboarding

Authorship rules

LLM Reference Policy & Strict No-Copy-Paste Rule

Use any reference you like while you think. What you submit has to come from your own hands. These rules apply to every candidate and every field in the studio.

RULE 01 · PERMITTED

Allowed: External LLM Reference

You may use an LLM in a separate window or device to:

  • brainstorm task scenarios,
  • check mathematical or domain formulas,
  • review library or CLI syntax,
  • pressure-test why automated agents fail.

Reference use in a separate window is welcome while you design your task.

RULE 02 · ENFORCED

Prohibited & Disabled: Zero Copy-Paste

Copy-paste is disabled across the assessment studio editors. That covers Ctrl+V / Cmd+V, right-click paste, and text drag-and-drop.

Every sentence and every code block you write in the studio must be typed directly, in your own words.

Attempting to bypass clipboard blocking or submitting unedited AI boilerplate results in permanent rejection of your email address.

RULE 03 · STEP 05

Local Task File Upload in Step 05

In Step 05 (Workspace Files) you may either type task.toml, tests/grade.py, and custom task files in the studio editor, or upload UTF-8 source, config, and data files from your own machine.

instruction.md and README.md are never uploaded. Type them in Steps 03 and 04.

Supports single UTF-8 files and full task folders up to 256 KiB total.

Step by step

What each of the 5 steps asks for.

The studio checks each minimum live and marks a step Ready when it is met. You can move between steps in any order and submit once from the readiness panel.

  1. Step 01 of 05

    Step 01 · Account & Primary Domain

    Create the login you will use after you pass and pick the engineering track your task belongs to.

    Typed fields. Nothing here is scored.
    Full legal name
    As it should appear on agreements and payout records.
    Email address
    Your sign-in address. A rejected email cannot re-apply.
    GitHub username
    Used for automated branch and pull-request provisioning on Terminal Benchmark 4.0.
    Workspace password
    10 to 72 characters.
    Primary engineering domain
    The specialty that best matches the task you author in Steps 02 to 05.
  2. Step 02 of 05

    Step 02 · Engineering Task Proposal

    Describe one realistic task and explain why it is hard for automated agents.

    Typed in the studio. Paste is disabled.
    Task title
    10+ characters.
    Scenario
    A realistic production or scientific scenario, 200+ characters.
    Why frontier agents fail
    A mechanistic explanation of the traps, 160+ characters. Name the step where a shortcut breaks.
    Viable approaches
    At least three distinct engineering approaches, 160+ characters in total.
  3. Step 03 of 05

    Step 03 · Candidate Prompt (instruction.md)

    Write the exact prompt an agent receives. It must stand on its own.

    Typed in the studio. Paste is disabled. Upload is not accepted for this file.
    Length
    A self-contained prompt of 240+ characters.
    No headings
    Plain paragraphs with zero # markdown headings.
    Output paths
    Explicit /app/... input and output paths.
    Tolerances
    Numerical or structural tolerances the verifier will enforce.
    Container handoff
    State that only the declared /app/... output transfers to the isolated grading container and other files are discarded.
  4. Step 04 of 05

    Step 04 · Maintainer Documentation (README.md)

    Tell maintainers why the task is hard, what a correct result looks like, and how it is checked.

    Typed in the studio. Paste is disabled. Upload is not accepted for this file.
    ## Difficulty explanation
    Required heading.
    ## What success looks like
    Required heading.
    ## Verification explanation
    Required heading.
    Reference versus generator
    Explain how the reference solution works without reading generator parameters.
    Verifier limitations
    State plainly what the verifier does not catch.
  5. Step 05 of 05

    Step 05 · Workspace Files (task.toml, tests/grade.py & custom file)

    Supply the task configuration, the check script, and at least one supporting file.

    Type in the studio editor or upload UTF-8 files and folders from your machine.
    task.toml
    schema_version = "2.0" (TB 4.0; schema_version = "1.4" also accepted), environment_mode = "separate", verifier network_mode = "no-network", and declared /app/... artifacts.
    tests/grade.py
    An executable check script of 120+ characters that inspects /app/... and writes a binary reward or raises assertions.
    Custom task file
    At least one file of 60+ characters, such as solution/solve.py, solution/solve.sh, data/sample.csv, or environment/Dockerfile.

Scoring

5 scoring gates, 100 points, 75/100 to pass.

Your submission is scored once across five gates. Reach 75 or more to pass and your expert account is created immediately. Below 75, the result explains which gates need work.

Assessment scoring gates
GateWhat it checksPoints
G1Authorship Authenticity & Domain Originality20 pts
G2task.toml Schema 2.0 / 1.4 & Isolation Contract20 pts
G3tests/grade.py AST Rigor & Custom Workspace File25 pts
G4instruction.md Prompt-Verifier Alignment20 pts
G5Proposal & README.md Engineering Depth15 pts
Total · Pass bar 75/100100 pts

Before you start

Confirm you are ready

The studio opens on Step 01 and keeps your answers in this browser tab until you submit. Set aside an uninterrupted session, keep your references open in another window, and type each answer yourself.

Tick the confirmation above to unlock the live assessment.