Online Judge

01What does the user see the instant they hit Submit?

The API accepts the code, returns a submission id immediately, and judging happens asynchronously — the verdict streams back seconds later, never inline in the request.

02How many languages, and who guarantees the toolchain?

A fixed language matrix (say C++, Java, Python, Go) with pinned compiler versions baked into sandbox images — the platform owns the toolchain, not the worker host.

03What exactly comes back as a verdict?

Per-test results rolled into one verdict — Accepted, Wrong Answer, Time Limit, Memory Limit, Runtime Error, Compile Error — plus measured CPU time and peak memory.

04Who sets the time and memory limits?

The problem does: each problem declares per-language CPU-time and memory limits, and the judge enforces them as hard caps, not suggestions.

05Is output always compared byte-for-byte?

No — each problem picks a checker: exact match with whitespace normalization, floating-point tolerance, or a special-judge program for problems with many valid answers.

06Can users ever see the hidden test data?

Never: users get the verdict and which test failed, but test inputs and expected outputs stay server-side — leaking them destroys the problem for everyone.

01Assume the code is actively malicious — what must hold?

Total containment: no network egress, no reading outside its own scratch directory, dangerous syscalls filtered — a hostile submission can kill its sandbox, never the host or another run.

02Same code submitted twice — same verdict, guaranteed?

Deterministic verdicts: one pinned core per run, CPU-time (not wall-clock) as the limit, no oversubscription — the same code must never get Time Limit on one run and Accepted on the next.

03How fast must verdicts land during a contest spike?

P95 verdict under about 30 seconds even when 10,000 submissions arrive in minute one — the queue absorbs the burst and the fleet drains it.

04One user fires 50 submissions — do others wait?

Scheduling is fair per user — round-robin across users, not FIFO across submissions — so one flooder queues behind their own work, not behind everyone else’s.

05What state may survive from one run to the next?

Nothing survives between runs: every submission gets a fresh sandbox from a clean image and a full teardown after — leftover processes or files corrupt both security and verdicts.