Quickstart — Validating Single-Authority Resolution Gates
Each scenario maps to a Success Criterion and is the acceptance proof for its IC.
SC-001 — The implement/review loop is unblocked (#2154, IC-04)
1. On a mission under coordination topology, mark a WP's subtasks done: spec-kitty agent tasks mark-status T001 T002 … --status done. 2. Advance it: spec-kitty agent tasks move-task WP## --to for_review. 3. Expect: the WP advances. Before the fix it fails with phantom "unchecked subtasks" (the write landed in coord, the validator read primary).
SC-002 — Mixed-bundle auto-commit, no silent drop (#2155, IC-04b)
1. Under coordination topology + an unprotected target branch, run move_task WP## --to <lane> (and the implement/claim path) so it auto-commits a primary WP file bundled with coord-owned status artifacts. 2. Expect: coord status lands on the coord surface, the WP file on primary, no swallowed SafeCommitPathPolicyError, tree clean. Before the fix the commit is silently skipped ("Auto-commit skipped" warning) and the activity-log update is dropped (dirty tree). A deliberately wrong-surface .worktrees/ write staged from primary is still refused by the unchanged guard (git/commit_helpers.py:983-991).
SC-003 — A future bypass fails CI (IC-02 / IC-03 self-tests)
1. Run the gate suite: pytest tests/architectural/test_resolution_authority_gates.py -q. 2. The in-test self-mutation proofs inject a bypass and assert the gate goes RED, then revert and assert GREEN — for both discriminators (un-canonicalized handle → blind primitive; kind-blind mandated write). 3. Expect: green (the self-tests prove the gates are non-vacuous).
SC-004 — Zero un-sanctioned bypassing sites (routed by default)
1. Run the gates against src/. 2. Expect: green — every one of the 38 primary_feature_dir_for_mission sites and the coord-authority write sites is routed (default) or carries a rationale'd already-canonical allowlist entry; ≥ the bare-handle census is routed, not allowlisted; the allowlist count is ≤ the pre-sweep baseline (shrink-only, no inflate-then-freeze).
SC-005 — Convergence across handle forms (FR-006, IC-05)
1. Run the convergence test: pytest tests/missions/test_*_convergence.py -q (uv-run if events 6.1.0 matters). 2. Expect: for every handle form (full slug, <slug>-<mid8>, bare mid8, ULID, numeric) the read-seam dir equals every write/placement-seam dir; the stub returns distinguishable per-form outputs (a constant-return stub is rejected); ambiguous handles raise MissionSelectorAmbiguous; cold-miss fails closed; a negative-control form (divergent under pre-fix code) proves the test is red-first. Stub-driven — no live kitty-specs/ fixtures.
SC-006 — Test-hygiene folds (FR-007/008, IC-06)
1. pytest tests/architectural/test_no_tmp_paths_in_tests.py -q → green; injecting a new /tmp/ literal in a test file turns it RED (frozen-baseline ratchet, ~82 baseline; the full litter sweep is out of scope). 2. FR-008: a before/after pytest --collect-only -m <shard> diff confirms the mission-owned target files are collected. If the originally-named files already run (audit indicates they do), this is satisfied-by-verification — no redundant marker added.
Full gate sanity (pre-merge)
PYTHONPATH=$PWD/src TERM=dumb NO_COLOR=1 pytest tests/architectural/ -q -p no:cacheproviderThe two new gates + the existing surface-resolution gates all green; the canonicalizer + coord-authority allowlists carry only live, rationale'd entries.