Arena FAQ

Arena FAQ

Short answers to what people ask in the first ten minutes of the L.A.B. simulator. Edit freely.

"Where do I start?"

python scripts/lab.py next. It lists what you can start right now, challenges first. C01 and C03 need nothing; everything else unlocks from something.

In a Codespace this runs for you on attach.

"What is the difference between a challenge and a lab?"

	Challenge C	Lab L
Time	5–15 min	15–50 min
Asks for	one mechanism	a decision, then the code
Shapes	implement, fill, fix, predict	implement
Checks	public + hidden	public + hidden
Ends with	what you can now do, and the lab it unlocks	a debrief, and the next lab with the reason

Challenges are the on-ramp. They are not prerequisites for labs — nobody partway through the labs is walled off by them.

"I filled the blanks and it says expected 4 sentences, got 1. What?"

That is the check working. A fill starter imports with its blanks unfilled — the harness binds ____ to placeholder text — so every check gets to say in the mechanism's own words what the blank costs. Read the message, not the traceback; there is no traceback.

If a message is unhelpful — you read it and still do not know what to change — that is a bug in the lab. Open a lab feedback issue.

"The public checks pass and --hidden fails."

Normal. Expected. The gap between them is the lesson. Hidden checks cover what the brief deliberately did not mention — empty input, a duplicate id, a degenerate config. The reply names the exact check and why.

"What does the sandbox actually see when I post in a thread?"

Your comment, and nothing else. The job that runs your code has no credentials (a zero-scope token), runs it in a container with no network, a read-only filesystem, no capabilities, as an unprivileged user, with a 30-second kill. The job that posts the reply never sees your code — it reads a JSON report. Your comment never reaches a shell command line either; it is passed by environment variable into a file.

So: a hostile submission finds nothing to reach. That is why it is safe to have a public thread accept code.

"Does a predict challenge need a code block?"

No. Post the word or number — negative, packing. The workflow wraps it as ANSWER = ... for you.

"I passed. Why does re-posting still count an attempt?"

Because the board answers how many tries did it take, and a pass never turns back into Retrying just because you were curious. Attempts go up; Outcome stays Passed.

"Where is my progress?"

• Locally: python scripts/lab.py status. Derived from the checks — nothing stored, nothing to drift.
• For the cohort: the Hands-on Tracker — one row per person per item, moved by the workflows when you submit in a thread or open a PR.

"Can I see the reference solution?"

reference.py in every lab directory. Read it after you pass. The debrief in the brief is worth more than the reference and is where the actual teaching is.

"The workflow did not reply."

Check the thread title starts with an [arena · Cnn] or [submit · Lnn] prefix; the workflow only listens there. If it does and nothing came back in ~2 minutes, the Actions tab shows the run — a timeout means something in your solution never returned.