# FLR-015 local lab

Use only the supplied synthetic fixtures. Implement `evaluate(case)` in `starter.py` so that classification follows foreground latency plus per-workload CPU/memory/I/O pressure evidence rather than raw utilization alone.

Your result must expose eight fields: `classification`, `latency_state`, `dominant_pressure`, `containment`, `closure_state`, `host_responsiveness`, `background_progress_state`, and `next_action`.

The fixture set includes a high-CPU negative control, a changed-context memory/I/O case, an over-constrained policy, and successful closure. Do not run destructive stress tests on a real workstation.