adversary: AC-mapped test ACTION-path invokes BC §Trigger method — rubric gap allows structurally green-by-design tests

## Summary

The adversary agent's review rubric does not check whether an AC-mapped test's ACTION path invokes the same method that the BC's §Trigger names as the AC's precondition-emission chain. This allows structurally green-by-design tests to survive multi-pass fresh-context review.

## Evidence from ftc-blue W3B2

STORY-5.05.005 `test_BC_5_05_005_ac001_puzzle_state_not_locked_or_failed_after_fm003`:
- Test was named to verify AC-001 post-FM-003 state invariant
- Test body called only `get_required_steps(...)` (pure helper, zero side effects)
- The AC's §Trigger names `enter_puzzle()` (real FM-003 emission path)
- Test docstring itself admitted the coupling gap
- **Survived Passes 2 through 7 of adversarial review before Pass 8 fresh context caught it**

Fix Phase 5f rewrote the test to invoke real `enter_puzzle()` and assert on `puzzle_states[...]["solve_state"]`. Falsifiability then proven via mutation cycle: inserting `solve_state="locked"` in the FM-003 branch made the test fail; reverting made it pass.

## Root cause

Adversaries currently check name/docstring coherence, non-tautological assertion form, and BC-section citation presence. They do NOT walk the test body's method-call graph to verify the SUT method named in the AC's §Trigger is invoked.

## Requested fix

1. Add to adversary rubric: for every AC-mapped test (name pattern `test_BC_X_YY_ZZZ_ac_N_*`), identify the BC's §Trigger method for AC-N and verify the test body ACTION path invokes it. If not, MOD finding regardless of assertion quality.
2. Add a codebase-analyzer heuristic that flags AC-N tests whose bodies don't call the AC's §Trigger method.
3. Pure-helper tests belong under `test_*_invariant_*_is_pure` naming, not AC labels.

## Cross-refs

- L-W3B2-06 codified in ftc-blue `.factory/cycles/cycle-001/lessons.md`
- Related: BC-5.38.001 Red Gate discipline (falsifiability proof)