Skip to content
ManualMode

Review an AI patch. Decide if it ships.

Review two synthetic patches against their contracts. Run one simulated check, decide whether to ship, and leave with concrete checks for your next AI-generated PR. No account or code upload.

Review workspace

2 patches · 1 simulated check each · no account

01 / Inspect proposed patch

Compare the diff with the review contract.

JavaScript / TypeScript
proposed.diff
AI patch · read only
export function findAccount(accounts: Account[], id?: number | null) {unchanged line
if (id == null) return null;removed line
if (!id) return null;added line
return accounts.find((account) => account.id === id) ?? null;unchanged line
}unchanged line
Review the change against the contract before you run a check. Scroll sideways for long lines.

Existing CI signal

current
  • PASSpatch typechecks; runtime boundaries are untested.
Run one check

Pick the check most likely to settle the contract. Its fixed result appears before you decide.

What this review measures

Code reading

Trace ownership, precedence, and state transitions before editing.

Debugging

Use failing evidence to isolate the smallest incorrect condition.

Test discipline

Choose a boundary check and interpret its simulated result.

AI-output verification

Approve a safe patch unchanged or correct a plausible defect.

Take the next step in the app

After a JavaScript/TypeScript review, continue into 3 executable JavaScript browser tasks. Any selected language can continue into 1 verified repository task using your local runtime.

  1. 01Complete 3 baseline tasksCode reading, async debugging, and judging a patch that may already be safe.
  2. 02Get a provisional profileYour tests, edits, completion, and optional Coach use shape the first recommendation.
  3. 03Complete one Project repConnect your coding agent and reserve one useful task already needed by your repository.

The short audit is review-only and does not count as a verified rep. Browser Gym executes JavaScript. Project Mode uses your repository language and local runtime through a connected coding agent. All 4 signed-in calibration reps remain free.

Before you start

Is this an AI coding assessment or a coding interview?
It is a short review exercise for developers who use AI coding agents. Two synthetic patches sample your test choice, diagnosis, and release judgment; they cannot measure your overall engineering ability. It is not an interview or hiring score.
What does the assessment measure?
The public audit samples test choice, diagnosis, and release judgment across two synthetic patches. The checks show fixed results; no code runs. Optional TypeSafe AI feedback reads only a review note you agree to send. The signed-in JavaScript baseline adds executable work. Neither result is a claim about overall engineering ability.
Can I use AI during the assessment?
Answer the public patch reviews yourself. Afterward, optional AI feedback can comment on a short note you choose to send. In the signed-in baseline, an optional bounded Coach can question your reasoning and point to current evidence without taking over the task.
Which programming languages are supported?
The no-account patch review supports JavaScript/TypeScript, Python, Go, Java, and C++ without executing code. Browser Gym currently executes JavaScript. Project Mode uses your repository language and local runtime through a connected coding agent.