Make it disagree with its first answer
Get the engine to argue against its own first response — substantively, not theatre. ROOM grades whether the second answer is actually better.
Contract
solve(input: any) → any
Return the indices, not the values. Order matters.
Visible tests
- in: [2, 7, 11, 15], target: 9→ [0, 1]
- in: [3, 2, 4], target: 6→ [1, 2]
- in: [3, 3], target: 6→ [0, 1]
+ 7 hidden tests fire on submit
Step in
The clock starts the moment you click. Bring your AI engine — your own API key — and the tape rolls.
Start the clock →What this trial is
Get the engine to argue against its own first response — substantively, not theatre. ROOM grades whether the second answer is actually better.
Long-form trial — par ~10 minutes. Designed for a full session; the tape itself is the artifact, and peers and judges will scrub through it.
Elo upside if you nail it
Solid run
+8 to +16
Top run
+22
Medal line
silver
Where most players live. A clean MARK + ROOM here moves the needle every week.
Solo Elo · MARK ÷ 100 vs field 2100, K-32. Codesport duel wins multiply by ×1.25, losses by ×0.40.
Trial leaderboard · PROMPT-007
Best runs on this trial.
Composite = MARK · 60% + ROOM · 40%. Top 15% qualify for bronze and become judging-eligible.
Only attempts in the top 15% on this trial — the bronze line and above — are eligible for a Judge's FINDINGS. Below the line, your run is yours; the bench doesn't weigh in.