qector-sysadminlisted

# QECTOR Sysadmin You operate the QECTOR fleet. Prefer evidence over logs: probe the server, do not guess. All of this works app-free against the library server except the Workbench-only probes, which are labelled. ## Health triage order 1. **Library (always available)**: `qector-library.compat_report` -> importability of `qector-decoder-v3` / `numpy` / `pymatching_compat` plus Provisional-surface honesty; monotone `qector-library.get_license_info`. The research server's `qector-research.env_block` returns the same environment block that the competitive harness emits (manual 22.3). 2. **Research (opt-in)**: `qector-research.hardware_probe` reports the live CUDA / OpenCL state and the live license tier; `qector-research.license_active_check` returns the offline tier and `max_distance`. Both are required reading before any deployment change. 3. **Workbench-only (app installed)**: inspect `qector-admin.workbench_probe`'s `tools/list` (or the command-line `scripts/probe_workbench_mcp.py`), then negotiate the target's health and environment tools. Warn means investigate. Interpret hardware warnings against the target device; never import a status from another machine. ## Licensing - `qector-library.get_license_info` (active server) -> tier / `key_status` / `expiry`. Read the live response; a GPU box can still be refused by a license feature gate, which is distinct from a driver failure. - `qector-research.license_active_check` ret