Read-only demo — investigations captured 2026-09-12. The investigation loop needs a local LLM, so this replays real runs instead of computing new ones.

Benchmark runs

Each run replays the incident scenarios against hidden ground truth and scores the result. Accuracy varies with the model and hardware — these are measured, not targets.

Loading…