Send us a dump.
We're testing whether a deterministic rule engine can match a senior CUDA engineer on real workloads. If you operate LLM inference at scale, send a redacted Nsight or DCGM dump — we'll run it through strided and share the diagnosis with you.