Add module governance engine parity test

[b-macker]

Summary

• Verifies CLAUDE.md claim: VM skip_main_ guards match tree-walker module_loading_depth_ == 0 guards for function-body governance during module loading
• 14 assertions across 5 groups test that both engines agree AND produce the correct verdict
• Follows the pattern established by Make the two taint engines agree #115 (taint parity) and Make govern.json sandbox settings reach every execution path #116 (sandbox parity)

What was traced

VM (compiler.cpp): skip_main_ at lines 89, 97, 399, 1129, 1267

• Line 399: skip main block when skip_main_
• Lines 1129/1267: two function-body sites — !skip_main_ → checkNaabFunctionBody() (full), skip_main_ → checkFunctionBehavioralContract() (lightweight)

Tree-walker (interpreter.cpp): module_loading_depth_ == 0 at lines 975, 982, 995, 1027, 1032, 1037, 1067, 1151

• Line 1027: skip main block
• Line 1151: module_loading_depth_ == 0 → checkNaabFunctionBody(), else → checkFunctionBehavioralContract()
• Lines 975-1067: lifecycle guards (hooks, scope validation, reports, scanner) — no VM equivalent needed (handled in main.cpp)

Result: The guards match. The 3-vs-8 site count asymmetry is different granularity, not different behavior.

Test groups

Group	Property	Assertion
A (3)	Module placeholder skip	Both engines allow placeholder in imported module
B (3)	Main file enforcement	Both engines block placeholder in main file (exit 3)
C (3)	Behavioral contracts	must_call fires during module load on both engines
D (3)	Main block skip	Module main{} not executed on import
E (2)	Controls	Placeholder check is live; disabling it lets code through

Degraded-case validation (guard removal, not expectation inversion)

Broke the property in the source: replaced both if (!skip_main_) guards in compiler.cpp (lines 1129, 1267) with if (true), removing the VM's module-loading skip while leaving the tree-walker's intact. Rebuilt and ran the test.

Assertion	Broken result	Message
MOD-A1	FAIL	engines disagree — VM matches=3 tree-walk matches=0
MOD-A2	FAIL	engines disagree — VM matches=0 tree-walk matches=1
MOD-A3	FAIL	exit code divergence — VM=3 tree-walk=0
MOD-C1	FAIL	engines disagree — VM matches=0 tree-walk matches=2
MOD-C2	FAIL	engines disagree — VM matches=1 tree-walk matches=0
B1-B3, D1-D3, E1-E2	PASS	Unaffected (main-file checks and main-block skip use separate guard)

All 5 broken assertions reported "engines disagree", confirming the test measures parity, not just correctness. Restored compiler.cpp, rebuilt, confirmed 14/14 pass.

Suite results (clean tree, run to completion)

Branch (claude/module-governance-parity-test): 441 tests, 4 unexpected failures
Master (origin/master): 441 tests, 4 unexpected failures (identical set)

Pre-existing failures on this platform (Termux/ARM64):

• test_consequence_proof.sh
• test_sandbox_engine_parity.sh
• test_semantic_signals.sh
• test_d1_reconciliation.sh

The branch introduces 0 new failures. Line 3308 of the suite log: test_module_governance_parity.sh: ALL PASSED.

Committed diff is exactly 2 files:

 run-all-tests.sh                                   |  15 +
 tests/governance_v4/test_module_governance_parity.sh | 436 +
 2 files changed, 451 insertions(+)

🤖 Generated with Claude Code

[github-actions]

NAAb Governance Report

Metric	Count
Files checked	16
Passed	16
Failed	0

All governance checks passed!

_{Generated by NAAb Governance Engine v4.0}