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.
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.
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.
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.
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.
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.
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
Gate
What it checks
Points
G1
Authorship Authenticity & Domain Originality
20 pts
G2
task.toml Schema 2.0 / 1.4 & Isolation Contract
20 pts
G3
tests/grade.py AST Rigor & Custom Workspace File
25 pts
G4
instruction.md Prompt-Verifier Alignment
20 pts
G5
Proposal & README.md Engineering Depth
15 pts
Total · Pass bar 75/100
100 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.