Copilot/apmlifecycle phase2 interactions

.github/workflows/ci.yml

@@ -198,6 +198,7 @@ jobs:
           tests/quality/test_test_taxonomy.py \
           tests/quality/test_quality_baselines.py \
           tests/quality/test_lifecycle_bug_ledger.py \
+          tests/quality/test_lifecycle_interaction_coverage.py \
           tests/quality/test_ci_topology.py
 
     - name: Run bounded architecture scaling gate

docs/src/content/docs/contributing/integration-testing.md

@@ -46,7 +46,9 @@ APM uses a tiered approach to integration testing:
 - **Fixture controls**: lifecycle helpers set `APM_TEST_LOOPBACK_PORTS` for a port-scoped local registry and `APM_TEST_FAIL_LOCK_REPLACE=1` for atomic-write fault injection. These are internal test controls, not user-facing APM settings.
 - **Learning ledger**: `tests/fixtures/lifecycle_bug_ledger.json` records representative escaped defects, their generalized laws, oracle tiers, implementation phases, and executable regression node IDs. It is deliberately not a bug-count census; `tests/quality/test_lifecycle_bug_ledger.py` validates its taxonomy and links in the required ratchet job.
 - **Generated lifecycle model**: `test_generated_lifecycle_state_machine.py` uses Hypothesis to generate guarded install, dry-run, audit, tamper, repair, declaration, and prune sequences against the real CLI. The model tracks declaration, materialization, integrity, and lock state independently of the product lockfile. Every transition captures complete project and user roots, and mutating commands must stay inside reviewed write sets. It stays in the merge-group family until hosted runtime supports promotion to the bounded PR-time smoke set.
-- **Known gap**: a late lockfile replacement failure can leave target files on the newly declared target while retaining the prior lockfile. The required lifecycle suite bounds that blast radius and proves the next install converges; expanding the install transaction is a separate design decision recorded in the ledger.
+- **Routing-cell coverage**: `test_primitive_target_covering_array.py` derives rows from `KNOWN_TARGETS` and exercises every valid primitive, target, and project/user-scope cell. Each install and uninstall must stay within reviewed project and user trees, and each reinstall must be byte-stable unless the cell is an explicit known gap. `tests/quality/test_lifecycle_interaction_coverage.py` is the required, subprocess-free ratchet: it checks complete catalog coverage, unique row IDs, and a 100-cell runtime ceiling. The executable rows remain merge-group-only.
+- **Known gap (lockfile replacement)**: a late lockfile replacement failure can leave target files on the newly declared target while retaining the prior lockfile. The required lifecycle suite bounds that blast radius and proves the next install converges; expanding the install transaction is a separate design decision.
+- **Known gap (Copilot user instructions)**: a repeated global Copilot instruction install adds a duplicate block under the lock identity before a third install converges byte-stable. The covering array restricts the second-pass changes to the lockfile and instruction file, asserts the duplicate exactly, and requires third-pass convergence. Unifying remote and lock identities is a separate production fix. Both known gaps are recorded in `tests/fixtures/lifecycle_bug_ledger.json`.
 - **Run it locally** (the exact command CI runs):
   ```bash
   APM_E2E_TESTS=1 uv run --extra dev pytest -p no:cacheprovider -q --strict-markers \

tests/fixtures/lifecycle_bug_ledger.json

@@ -186,6 +186,13 @@
       "properties": ["transaction.failed_command_preserves_state"],
       "bounded_by": "tests/integration/test_required_lifecycle_state_machine.py::test_required_failed_lock_write_bounds_partial_state_and_recovers",
       "next_decision": "Full install transaction expansion remains deferred with phases 2-4."
+    },
+    {
+      "id": "copilot-user-instruction-second-pass-convergence",
+      "summary": "The first repeated global Copilot instruction install adds a duplicate block under the lock identity before a third install converges.",
+      "properties": ["idempotency.byte_stable", "ownership.preserve_unowned"],
+      "bounded_by": "tests/integration/test_primitive_target_covering_array.py::test_primitive_target_covering_array",
+      "next_decision": "Unify remote and lock identities for user-root instruction ownership in a focused production fix."
     }
   ]
 }