$ reward-lens doctor

  Install
    reward-lens 3.1.0  ·  python 3.11.13  ·  linux x86_64  ·  <captured-install-location>
    extras installed: trace, train          audit on local graders: yes

  Sandbox
    L3: Landlock + seccomp inside bubblewrap --unshare-all     probed in 14 ms, held
    per call: 30 s wall, 2 GB, 64 procs, no network

  What this machine can measure now
    ✔ validity (the static checks and replay), reach (static exposure only)
    ✔ trace, forecast, improve

  What it cannot, and what each would take
    ○ soundness, exploits, framing, signal, cost      not in this build
    ○ compare                                         not in this build
    ○ the seeker, either arm                          not in this build
    ○ calibration and limits of detection             no reference material for this substrate yet

  Network   offline is available and is the default for local graders
  Cost      $0.00 for everything in the first list
