grade-a-pipelinelisted

# Grade-A Pipeline An opinionated software-engineering workflow that composes `agent-orchestra` for the build case. It maps the repository, plans a dependency graph, isolates implementers in worktrees, tests every integration wave, and subjects accepted work to independent review. The rubric is evidence about the run, not a guarantee of software quality. > This is the build-side sibling of `live-test` (which exercises a running system through the browser) and `saturating-review-engine` (which hardens a rendered artifact). Use this one when the deliverable is *code in a repo*. ## What it does (the test sandwich) ``` BOOTSTRAP open integration branch + worktree IN-REPO (no /tmp) + detect commands CARTOGRAPHY N readers over file shards → synth → MAP.md (committed + at repo root) ← the MAP PLAN 2 isolated planners → arbiter → task DAG (disjoint single-owner files) BASELINE run the existing full suite (read-only worktree) ← sandwich ▽ BUILD waves per wave: file-disjoint batches; each task = 2 TDD impl → verify → adversary ↳ each implementer works in ITS OWN worktree+branch (gap/<run>/task-<id>-<variant>), commits INTEGRATE commit accepted diffs onto the integration branch, re-run FULL suite,← sandwich ▽ (every wave) tag gap/<run>/wave-N, push checkpoint to origin = a checkpoint you can revert to BATTERY unit · integration · e2e · property · lint/type · security in parallel ← sandwich △ REVIEW code-re