Clarifications on runtime limit, RAM spec, and GPU driver

Dear Organizing Committee,

We noticed some inconsistencies between the released documents.

1. Per-case runtime limit
The Submission Guidelines (2026/06/16) state "Runtime: No upper limits (runtime measured only during calls to solve())".
However, the provided op_wrapper.py example sets timeout=60 (with check=True) on the subprocess call inside solve(), which would raise a TimeoutExpired exception once a single case exceeds 60 seconds.
- Which behavior applies to the official evaluation: no upper limit, or a 60-second per-case timeout?
- If a hard timeout is enforced, how is a timed-out case scored? Is it treated as infeasible (Cost = M = 10 for that case), or handled some other way?

2. Evaluation machine RAM
The Q&A document (2026/06/18) states 128 GB RAM, while the Submission Guidelines (2026/06/16) state 250 GB RAM. Could you confirm the correct amount of RAM available per submission during evaluation?

3. NVIDIA driver version
The documents specify the A100 80 GB GPU, CUDA 12.5, and PyTorch 2.12.0+cu130, but we could not find the NVIDIA driver version of the evaluation machine. Could you please provide it, so we can ensure our bundled CUDA runtime is compatible?

Thank you very much for your help. We appreciate your clarification so we can align our implementation with the official evaluation setup.

Start by comparing the Submission Guidelines, Q&A document, and provided op_wrapper.py example, especially the solve() subprocess timeout behavior. Confirm the official per-case timeout and scoring, evaluation RAM, and NVIDIA driver version, then update the relevant documents so these specifications agree and are unambiguous.

Written by the indexing model from the issue text.