16 — Regression Tests

v_rp_invariant_regression_guard — 6 computed tests, 6/6 PASS (dual-path verified).

#	test	expected	actual	result
1	v1 invariant still shows old fails	2	2	PASS
2	v2 invariant zero real fails	0	0	PASS
3	proof v2 ALL_AXES not FAIL	NOT_FAIL	PASS_WITH_EXPECTED_BLOCKER	PASS
4	dashboard v3 verdict consistent with fails	CONSISTENT	CONSISTENT	PASS
5	static literal v2 zero stale	0	0	PASS
6	reintroduced stale literal would FAIL v2 rule	FAIL_COUNT_SUBSTRATE_MISMATCH	FAIL_COUNT_SUBSTRATE_MISMATCH	PASS

Interpretation

• Test 1 keeps the old failure visible in v1 (sentinel: the fix is in v2, not by hiding v1).
• Test 6 is the teeth-test: it computes what the v2 rule would say if residual count were reintroduced as the stale literal 8 against a scoped reconcile_target of 2 → FAIL_COUNT_SUBSTRATE_MISMATCH. Proves the rule still has teeth and cannot be silently regressed.
• Test 4 enforces the acceptance dashboard cannot show green while real_fails>0. These guards make the green state falsifiable, not asserted.