First official grading pass (before the grader-rule correction described in POSTRUN_CHANGES.md), 2026-08-26:

  score_stage_a.py --label official
  A5_mitigation_control    mitigated Z: expected CONSISTENT                 got INCONSISTENT               MISMATCH
  STAGE A ALL AS FROZEN: False
  build_reports: verdict: FAIL/INCOMPLETE | stage A as frozen: False | reproduction identical: True

Cause: exact zero-probability-cell rejection applied to the mitigated pseudo-count distribution (confusion-matrix inversion, clipped and
renormalised), whose residual mass on ideal-zero Z outcomes is ~1e-4 (processing artefact). All other classifications were identical to the
final pass. SaC states, hardware data and criteria were unchanged between passes.
