Quickstart — Validating Coord Write-Placement Closure & Birth-Cutover

How a reviewer confirms the mission's outcomes (maps to SC-001…006). Run from repo root.

1. Corpus green + stays green (SC-001)

PWHEADLESS=1 uv run pytest tests/specify_cli/migration/test_dogfood_corpus_backfilled.py -q
  • After IC-01 front-load: passes. After IC-09 re-key: still passes, and would red if a mission merged un-reconciled.

2. Whole-tree write enforcement (SC-002)

PWHEADLESS=1 uv run pytest tests/architectural/test_no_write_side_rederivation.py tests/architectural/test_safe_commit_import_boundary.py -q
  • Inject a synthetic non-seam-derived safe_commit/CommitTarget/write_meta into a module outside the former 17-module allowlist → the gate reds and names the site.

3. Read-side partition safety (SC-003)

PWHEADLESS=1 uv run pytest tests/architectural/test_read_surface_placement_guard.py -q
  • A coord-homed read against a primary substitute raises a typed partition-mismatch error (no silent fallback).

4. Repair a pre-existing split-brain (SC-004)

uv run spec-kitty agent mission repair --mission <handle>          # forward-only under strict-ancestor + clean worktree
  • On genuine divergence: refuses with a unified diff, zero mutation.

5. Born-reconciled mission (SC-005)

  • Create a mission, implement a WP (claim + tick a subtask), merge it, then assert it lands status_phase>=1 + verify_backfill().ok + non-empty snapshot with no migrate backfill-runtime-state invocation.

6. Classification + fork closure (SC-006)

uv run pytest tests/architectural/ -q -k "write_side or placement or safe_commit"
  • decisions.events.jsonl / traces/ route to COORD via the port; the emit HEAD-derived fallback is gone.

Gate discipline

  • Every FR ships a red-first test through the real entry point (DIRECTIVE_041).
  • ruff + mypy zero-warning; the three existing write-side gates stay green (NFR-004).