Usage: agentfence runs compare [OPTIONS] {baseline} {candidate}

  Compare two saved runs scenario by scenario. Read-only.

  Both runs are verified first. Pairs whose scenario version, evaluator version,
  fixture (seed), isolation tier or agent differ are INCOMPARABLE, never
  silently compared. Exits 1 when anything REGRESSED, otherwise 0.

Arguments:
  baseline   Baseline run ID.  [required]
  candidate  Candidate run ID.  [required]

Options:
  --json  Emit exactly one JSON document on stdout.
  --help  Show this message and exit.
