troupe-standup/wizard-work/forge/Q93_REFACTOR_VERIFICATION.md

Q93 selftest_sweep refactor -- verification (forge: Refactor lane)

Work

Commit 9878f12 ("forge: Q93 contract-grading + unified run_verdict in selftest_sweep (44 cells, 0 fatal)") refactored tools/selftest_sweep.py:

• judge_malformed(rc, out, err, expect_rc, need) -- the 5-arg CONTRACT form. A refusal (expect_rc == REFUSAL_RC == 2) is now graded on its contract (documented rc, no traceback, one non-empty line on stdout OR stderr), not on a pinned prose phrase. need is consulted only where a tool GRADES a malformed input (rc 0/1), where the verdict word IS the verdict.
• run_verdict(tool_fatal, read_fatal, malformed_fatal, pin_rc, pin_fatal) -> (line, exit_code) from ONE function, so the printed summary and the exit code cannot disagree (the Q93 core fix: the old sweep printed "sweep: 22 tools, 22 OK, 0 fatal" while exiting 1).
• count pin moved 39 -> 44; the stale 1 if (rc or rrc or pin_rc) else 0 main() arm is removed.

Evidence

The committed verifier reproduces the green run against the on-disk tree:

$ python3 wizard-work/forge/verify_selftest_sweep_q93_refactor.py
  [ok  ] py_compile
  [ok  ] static marker: def run_verdict(...)
  ... (18 checks) ...
  [ok  ] judge_malformed has 5-arg (rc,out,err,expect_rc,need) signature
verify_selftest_sweep_q93_refactor: ALL GREEN (0 failed, all checks)        rc=0

Full captured stdout/stderr is in Q93_REFACTOR_VERIFICATION_EVIDENCE.txt.

Gate

tools/selftest_sweep.py --selftest prints selftest: 44 cells, 44 passed, 0 failed and exits 0; the read-refusal (Q53) and malformed (Q53) axes both grade against the Q93 contract; run_verdict returns the rc its verdict line states. No regressions vs. the pre-Q93 Counter-census refactor (27c9869), whose verifier this file supersedes.