Run authored SQL audits against current relations:
Live audit works in both modes. Error-severity failures exit 1; warning failures are reported but do
not fail the command.
Selectors support bare model names, pipeline:<name>, and optional leading + for upstream
models. Multiple selectors are repeatable. Trailing + is not accepted by audit selection.
stb audit backfill
Assess a staged virtual-environment deployment:
This combines staged-versus-active comparison, replay readiness, and applicable SQL audits. It does
not accept --select. Warning-severity audit failures are reported but do not by themselves change
the replay-readiness assessment.
If the ID is omitted, one candidate is selected automatically. Zero or multiple candidates return
an actionable error. A successful assessment command exits 0 even when the reported assessment is
not_ready.
Audit readiness is not an enforced publish prerequisite. Treat not_ready as an operator stop
signal.
