conductor(plan): add hard rules #11 (no-op ban) and #12 (metric revert) after Tier 2 failure

conductor/tracks/metadata_promotion_20260624/plan.md

@@ -1872,6 +1872,26 @@ EOF
 8. **Do NOT exceed 5 nesting levels** — extract to functions if you hit the limit.
 9. **Do NOT modify `src/code_path_audit*.py`** — the audit infrastructure is correct.
 10. **Do NOT promote `Metadata: TypeAlias = dict[str, Any]` itself** — it's preserved as the catch-all.
+11. **NO-OP CLASSIFICATION BAN (added 2026-06-25 after a Tier 2 failure)** — Do NOT classify any planned task as "no-op" without explicit user authorization in the same session. If a planned migration site exists in the plan's Task table, MIGRATE it. The plan is the contract; FR2 is a rule in the spec; the plan's tasks are what you execute. If FR2 and the plan conflict at a specific site:
+    - Read the site's variable type via `git grep -B 5 -A 5 <site>` to determine if it's a dict or a dataclass instance.
+    - If it's a dataclass instance: migrate to direct field access per the plan.
+    - If it's a dict at a genuinely collapsed codepath (TOML config, JSON wire, handler-map): migrate anyway by inserting `entry = <Aggregate>.from_dict(entry)` (or equivalent) BEFORE the access site, OR classify the SPECIFIC site as collapsed-codepath and document in the per-site commit message (do NOT classify the whole phase).
+    - If you can't decide: STOP. Write a 1-sentence question. Wait for the user. Do NOT invent a reconciliation.
+12. **METRIC REGRESSION REVERT (added 2026-06-25)** — After every consumer-migration phase (Phases 1-10), run:
+    ```bash
+    uv run python -c "
+    import sys
+    sys.path.insert(0, 'scripts/code_path_audit')
+    sys.path.insert(0, 'src')
+    from code_path_audit import build_pcg
+    from code_path_audit_ssdl import count_branches_in_function
+    pcg = build_pcg('src').data
+    metadata_consumers = pcg.consumers.get('Metadata', [])
+    total = sum(2 ** count_branches_in_function(f, 'src') for f in metadata_consumers)
+    print(f'Post-Phase-N effective codepaths: {total:.3e}')
+    "
+    ```
+    If the metric did NOT decrease vs. the previous measurement, `git revert <phase_commit_sha>` IMMEDIATELY. Do NOT add a followup task to "investigate". Do NOT rationalize in the commit message or TRACK_COMPLETION. The 4.014e+22 → < 1e+20 drop is the success criterion; if it's not happening, the implementation is wrong; revert and report.
 
 ## Per-phase Tier 2 review checklist
 
@@ -1882,7 +1902,7 @@ Before approving each phase, Tier 2 verifies:
 3. The pre-phase git grep counts decreased by the expected amount (e.g., Phase 1 should remove ~50 `.get('id', default)` sites).
 4. The audit gates (`audit_weak_types.py --strict`, `audit_main_thread_imports.py`, etc.) still pass.
 5. The batched test suite (`scripts/run_tests_batched.py`) still passes 10/11 tiers.
-6. The effective codepaths metric decreased (or held steady for design-only phases).
+6. The effective codepaths metric decreased (or held steady for design-only phases). **If metric did NOT decrease after a consumer-migration phase (1-10), REVERT that phase commit immediately per Hard Rule #12. Do NOT approve.**
 
 If any check fails, Tier 2 REVERTS the phase commit and reports to the user.