[trissim]

P440 follow-up: delete tests of removed automatic enforcement

Authorized by commander message 16806; follows merged #2555 (4e6038d).

Head: e6b17a4, remote equality verified. Base: ac57d6a.

Deleted exactly two obsolete tests from tests/test_todos.py: test_blocked_enforcement_rejects_a_recovered_goal and test_blocked_enforcement_preserves_reassigned_orders. They exercise automatic blocked-order rewriting and GoalCannotRun.goal, both deleted by #2564. Keeping them would pin removed behavior or force the implementation back in.

54 test lines deleted; production source change 0. No compatibility alias, fake guard, monkeypatch, duplicated fixture or production behavior added. #2555's typed actor/assigner and fork competition repair remains intact on main.

The existing owner family remains in tests/test_holder_standing_rules.py: failed/unresolved attempts are reported without goal/order writes or UNKNOWN replay; explicit reissue still uses its declaration-owned factory; the real managed SIGTERM case remains, though its native setup is blocked below.

Source evidence

Existing refactor-audit Package parser: before/after 440 production and 473 test modules, zero parse omissions. Read GoalCannotRun and BlockedExecution semantically: neither owns the deleted automatic write loop. Both obsolete test consumers were identified before editing and are absent after; replacement report-only family tests remain. Async SIGTERM declaration was read directly and collected in the real run. Dynamic resolution is outside this syntactic result; AST is not runtime proof. git diff --check passed.

Actual normal pytest/conftest comparison

Both commands use PYTHONPATH=src:tests and the shared /home/ts/code/projects/agent-comms/.venv/bin/python:

timeout 300 python -m pytest -q -n 0 -o faulthandler_timeout=60 --basetemp=<fresh owned path> --junitxml=<owned report> <the same 14 files>

No --noconftest, fixture plugin, native rebuild, venv creation, uv sync, local ratchet or full suite. The exact-pin b6969b1e905f06ca SDK was copied unchanged from the existing P447 checkout, not renamed from an older artifact. That addresses the missing directory, not the prefix mismatch below.

• Head e6b17a4: 114 passed, 5 failed, 2 setup errors, 121 collected, 139.02s, exit 1.
• Main ac57d6a: 114 passed, 7 failed, 2 setup errors, 123 collected, 113.94s, exit 1.
• JUnit comparison: HEAD_ONLY_BAD is empty; MAIN_ONLY_BAD is exactly the two deleted tests. Remaining failing/setup case identities are identical. Zero skipped cases. All 9 test_commander_forks cases passed on both snapshots, including the four requested copied-bus cases and superseded verdict.

Per-file executed counts, passed / failed / setup-error:

tests file	head	main
test_bus_files.py	4/0/0	4/0/0
test_commander_forks.py	9/0/0	9/0/0
test_commander_run_rules.py	2/0/0	2/0/0
test_declared_schema_carry.py	19/0/0	19/0/0
test_held_problem_orders.py	4/0/0	4/0/0
test_holder_standing_rules.py	21/1/1	21/1/1
test_order_progress.py	8/0/0	8/0/0
test_order_scopes.py	8/1/0	8/1/0
test_order_story.py	1/0/0	1/0/0
test_owned_turn_goal_replacement.py	2/0/1	2/0/1
test_proof_build.py	13/0/0	13/0/0
test_storage_violations.py	5/0/0	5/0/0
test_todos.py	16/1/0	16/3/0
test_unowned_checkout_work.py	2/2/0	2/2/0

Remaining failures, not marked passed

• P447/native qualification, owner unit-mi-1d: both setup errors say Native Pi package is outside the running runtime prefix. conftest's native_runtime_package says it selects the running interpreter's artifact but pinned_native_package resolves a source checkout's stack; acquire correctly refuses it. Exact reproducer/owner trace sent as 16843. No production guard was weakened. A matching SDK copy alone does not resolve this.
• Current-main carry/schema cases: test_install_discovers_the_decision_cutover_from_its_dry_run_and_runs_it_once encounters undeclared table goal_human_decision; test_carry_scopes_current_nested_and_prior_orders_on_a_copy calls a tuple. These fail identically on main and head. They need the current carry/schema integration owner (claude-ai), not a competing P440 schema implementation.
• Earlier baseline fixtures: watcher fixture has no configured repository holding agent_comms.plan_phases; two unowned-checkout fixtures expect missing config/global paths. All fail on both snapshots; commander triage remains needed.

The source-only cleanup is complete. Broader P440 installed/native acceptance is not claimed; prefix qualification and existing baseline failures remain named open ends. Merger owns fresh-environment installation and actual affected installed acceptance. The earlier schema carry failure is not excused by these test results.

Hosted checks and Core delivery

All six exact-head hosted jobs passed: smoke, native-scripts, install-rehearsal, review-verdict, debt-src, debt-tools (run 39688966135). PR is ready; gh pr merge 2576 --auto returned success.

Core linked #2576 to P440, but CommsOrderDeliverTool refused the holder's claim: Only a pending order can be claimed by its holder. The existing P440 order already merged with #2555. A pending follow-up order from claude-code is required to record this delivery; the refusal was not retried or bypassed. This is not an accepted delivery or installed result.

Assumed: no installed success, no clean full suite, no full NRA or clean typing claim.

Evidence owner: unit-mi-1b. The two disposable b696 SDK copies made for this run were under /home/ts/wt/unit-mi-1b--P440/stack/ and /home/ts/wt/unit-mi-1b--P428/stack/. Both test processes finished, the baseline checkout was restored and process-argument borrower checks found no users of either copy; these copies are retired. P447's original artifact and all session/input evidence are preserved. Normal pytest logs/XML: p440-cleanup-head-tests.log, p440-cleanup-head.xml, p440-cleanup-main-tests.log, p440-cleanup-main.xml. Source traces: p440-cleanup-before.log, p440-cleanup-after.log. All under /home/ts/.cache/agent-scratch/unit-mi-1b/. P428 baseline checkout was restored to fix/p428-resolved-entry; no original session/UNKNOWN input was replayed or removed.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

An organization owner can choose the repository organization as the billing source for reviews requested by members with a paid Copilot license. Learn about Copilot code review billing.