SoundMindsAI · SoundMindsAI · Jun 16, 2026 · Jun 16, 2026 · Jun 16, 2026 · Jun 16, 2026
@@ -110,15 +110,28 @@ jobs:
   # for the rationale. The new workflow also bundles a content-scanning
   # gitleaks job; both are parallel + ≤30s each.
 
-  backend-unit-fast:
-    # Fail-fast lane for the most common AI/human iteration failure mode:
-    # broken unit tests. Runs in parallel with the heavy `backend` job and
-    # surfaces unit-test regressions in ~50s vs the full job's ~6m. Unit
-    # tests are pure (no DB, Redis, ES, or OpenSearch needed) — 710 cases
-    # complete locally in ~5s. The full `backend` job below still runs the
-    # same unit tests as part of its coverage-gated run, so coverage logic
-    # is untouched; this lane is an early signal, not a replacement.
-    name: backend (unit tests — fast lane)
+  backend-unit:
+    # Lane 1 of 2 that feed the coverage gate (Win 2′ of
+    # infra_pr_yml_split_backend_test_lanes). Replaces the previous
+    # `backend-unit-fast` lane, doing double duty:
+    #
+    # 1. Fast signal — unit tests run in ~30-60s (vs the heavy lane's ~7-8m),
+    #    so a broken unit test still surfaces quickly. Service containers
+    #    are NOT needed; unit tests are pure (no DB / Redis / ES / OS).
+    # 2. Coverage feed — runs with `--cov=backend` + `COVERAGE_FILE=.coverage.unit`
+    #    so the data file lands at a unique name (NOT in coverage.py's
+    #    parallel mode, which would litter every local-dev `pytest --cov`
+    #    run with `.coverage.<host>.<pid>.<rand>` clutter). The single
+    #    `.coverage.unit` file feeds the `backend-cov-gate` job that
+    #    combines this lane's coverage with the heavy lane's and gates
+    #    against the 80% fail_under threshold.
+    #
+    # `-n auto` parallelism is safe at the unit layer (the layer that
+    # broke under -n auto was integration, due to FK-teardown collisions —
+    # see the `backend-heavy` comment). The combined parallel + coverage
+    # cost is approximately a wash with the previous --no-cov serial run.
+    name: backend (unit + cov, parallel)
+    if: ${{ vars.SKIP_HEAVY_CI != 'true' }}
     runs-on: ubuntu-latest
     timeout-minutes: 5
     steps:
@@ -138,8 +151,31 @@ jobs:
       - name: Install project deps (frozen)
         run: uv sync --frozen
 
-      - name: pytest backend/tests/unit/
-        run: uv run pytest backend/tests/unit/ --no-cov -q
+      - name: pytest backend/tests/unit/ (-n auto, with coverage)
+        # `--cov-report=` (empty) suppresses the immediate terminal report —
+        # the report is generated by `backend-cov-gate` after the combine.
+        # `--cov-fail-under=0` disables pytest-cov's per-lane fail_under
+        # check (which would otherwise reject the unit lane at ~64% — unit
+        # tests alone don't exercise the full app; the 80% gate fires in
+        # `backend-cov-gate` against the COMBINED unit + heavy coverage).
+        env:
+          # Override the default `.coverage` base name so this lane's
+          # data file lands at `.coverage.unit`. `coverage combine` in
+          # `backend-cov-gate` matches any file starting with `.coverage.`
+          # so `.coverage.unit` is picked up alongside the heavy lane's
+          # `.coverage.heavy`.
+          COVERAGE_FILE: .coverage.unit
+        run: uv run pytest backend/tests/unit/ -n auto --cov=backend --cov-report= --cov-fail-under=0
+
+      - name: Upload partial coverage data (unit lane)
+        if: always()
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7
+        with:
+          name: coverage-data-unit
+          path: .coverage.unit*
+          if-no-files-found: error
+          include-hidden-files: true
+          retention-days: 1
 
   # -------------------------------------------------------------------------
   # License headers — REUSE (https://reuse.software/) SPDX compliance.
@@ -358,8 +394,18 @@ jobs:
       - name: vitest
         run: pnpm --dir ui test
 
-  backend:
-    name: backend (tests + coverage)
+  backend-heavy:
+    # Lane 2 of 2 that feed the coverage gate (Win 2′ of
+    # infra_pr_yml_split_backend_test_lanes). Runs contract + integration
+    # tests SERIALLY (no `-n auto`) because the integration layer hit
+    # FK-teardown collisions under parallel workers (PR #291 first run —
+    # `query_sets_cluster_id_fkey` violation when one worker held a FK
+    # reference to a cluster being deleted in another worker's teardown).
+    # Contract tests bundle here too — several boot the FastAPI app via
+    # `LifespanManager` (touches Postgres / Redis at startup), so they
+    # need the service containers this lane provides anyway. `pytest-xdist`
+    # remains in dev deps for local opt-in on the unit layer.
+    name: backend (contract + integration + cov)
     # TEMP CI-budget kill-switch — see the SKIP_HEAVY_CI note at the top of
     # this file. When the repo variable SKIP_HEAVY_CI == 'true', skip this
     # >1-min job to conserve GitHub Actions minutes. Default (var unset) → runs.
@@ -472,7 +518,18 @@ jobs:
       # and GHA does not auto-cancel sibling jobs). See
       # chore_pr_yml_parallelize_backend_job/idea.md (Win-1 residual).
 
-      - name: pytest (unit + integration + contract) with coverage
+      - name: pytest (contract + integration) with coverage
+        # Unit tests moved to the `backend-unit` lane (which runs in parallel
+        # with this job under `-n auto`). `--cov-report=` (empty) suppresses
+        # the immediate report — the final combined report + 80% gate run in
+        # `backend-cov-gate` after both lanes upload their data files.
+        # `--cov-fail-under=0` disables pytest-cov's per-lane fail_under
+        # check (the contract+integration lane alone measures at ~61% without
+        # the unit lane's contribution; the 80% gate fires in
+        # `backend-cov-gate` against the combined coverage). `COVERAGE_FILE`
+        # writes to `.coverage.heavy` (vs the unit lane's `.coverage.unit`)
+        # so the two artifacts coexist in `backend-cov-gate`'s cwd at
+        # combine time without colliding.
         env:
           DATABASE_URL_FILE: ${{ runner.temp }}/database_url
           POSTGRES_PASSWORD_FILE: ${{ runner.temp }}/postgres_password
@@ -487,30 +544,22 @@ jobs:
           # The health integration test will skip without a running API; that
           # behavior is acceptable until the deploy workflow exists.
           RELYLOOP_API_URL: http://localhost:8000
+          COVERAGE_FILE: .coverage.heavy
         run: |
-          # CI-perf #3 (pytest-xdist `-n auto`) was attempted on the first
-          # PR #291 CI run and reverted: the integration test layer hit FK
-          # collisions (query_sets_cluster_id_fkey violation when parallel
-          # tests held a FK reference to a cluster being deleted in another
-          # worker's teardown). pytest-xdist remains in dev deps for local
-          # opt-in (`pytest -n auto` works fine on the unit-test layer).
-          # A follow-up may split this job into a parallel-safe
-          # "unit + contract" lane (`-n auto`) + a serial "integration" lane
-          # (with a `coverage combine` merge/gate job) to recover the
-          # unit+contract portion's runtime while keeping integration serial.
-          # See planned_features/02_mvp2/infra_pr_yml_split_backend_test_lanes/idea.md.
-          uv run pytest backend/tests/ \
+          uv run pytest backend/tests/contract backend/tests/integration \
             --cov=backend \
-            --cov-report=xml \
-            --cov-report=term-missing
+            --cov-report= \
+            --cov-fail-under=0
 
-      - name: Upload coverage XML
+      - name: Upload partial coverage data (heavy lane)
         if: always()
         uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7
         with:
-          name: coverage-xml
-          path: coverage.xml
-          if-no-files-found: warn
+          name: coverage-data-heavy
+          path: .coverage.heavy*
+          if-no-files-found: error
+          include-hidden-files: true
+          retention-days: 1
 
       - name: Verify source-of-truth enum comments (feat_studies_ui Story 4.2)
         # Fails when a backend Literal / frozenset gains or loses a value but
@@ -537,6 +586,87 @@ jobs:
         # smoke-test job runs on a fresh runner so it never sees that path).
         run: bash scripts/ci/verify_install_builds_all_services.sh
 
+  backend-cov-gate:
+    # Coverage combine + 80% gate. Downloads partial `.coverage.*` data
+    # files from `backend-unit` (parallel unit lane) and `backend-heavy`
+    # (serial contract + integration lane), merges them via
+    # `coverage combine`, and fails the build if combined coverage is
+    # below the 80% threshold set in pyproject.toml's
+    # [tool.coverage.report] fail_under = 80.
+    #
+    # Win 2′ of infra_pr_yml_split_backend_test_lanes. The single tradeoff
+    # vs the previous all-in-one `backend (tests + coverage)` job: the gate
+    # is reported in its own job rather than inline with the test output,
+    # so a developer hunting a coverage regression in PR-checks UI clicks
+    # one more level to see the missing-line report. Acceptable for the
+    # ~1.5 min wall-clock recovery.
+    name: backend (coverage gate)
+    if: ${{ vars.SKIP_HEAVY_CI != 'true' }}
+    needs: [backend-unit, backend-heavy]
+    runs-on: ubuntu-latest
+    timeout-minutes: 5
+    steps:
+      - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@37802adc94f370d6bfd71619e3f0bf239e1f3b78 # v7
+        with:
+          enable-cache: true
+          version: "0.5.7"
+
+      - name: Set up Python
+        uses: actions/setup-python@a309ff8b426b58ec0e2a45f0f869d46889d02405 # v6
+        with:
+          python-version: "3.13"
+
+      - name: Install project deps (frozen)
+        # Need coverage.py + the project source tree (for the source path
+        # mapping in [tool.coverage.paths]). `uv sync` is the simplest way
+        # to materialize both.
+        run: uv sync --frozen
+
+      - name: Download partial coverage data (unit lane)
+        uses: actions/download-artifact@018cc2cf5baa6db3ef3c5f8a56943fffe632ef53 # v5
+        with:
+          name: coverage-data-unit
+          # All `.coverage.*` files land in the workspace root so
+          # `coverage combine` picks them up.
+          path: .
+
+      - name: Download partial coverage data (heavy lane)
+        uses: actions/download-artifact@018cc2cf5baa6db3ef3c5f8a56943fffe632ef53 # v5
+        with:
+          name: coverage-data-heavy
+          path: .
+
+      - name: Combine coverage data files
+        # `coverage combine` merges every file starting with `.coverage.`
+        # in the cwd into a single `.coverage` database. In our setup
+        # that's `.coverage.unit` (from backend-unit) and `.coverage.heavy`
+        # (from backend-heavy), downloaded by the two preceding steps.
+        # The [tool.coverage.paths] mapping in pyproject.toml maps per-runner
+        # absolute paths (e.g. /home/runner/work/relyloop/relyloop/backend)
+        # back to the canonical `backend/` source so the combined report
+        # deduplicates correctly across runners.
+        run: uv run coverage combine
+
+      - name: Coverage report (--fail-under=80)
+        # `fail_under = 80` is set in pyproject.toml's [tool.coverage.report]
+        # — `coverage report` honors it and exits non-zero if combined
+        # coverage drops below.
+        run: uv run coverage report --show-missing
+
+      - name: Generate coverage XML (for downstream tooling)
+        run: uv run coverage xml
+
+      - name: Upload combined coverage XML
+        if: always()
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7
+        with:
+          name: coverage-xml
+          path: coverage.xml
+          if-no-files-found: warn
+
   frontend:
     name: frontend (lint + typecheck + tests + build)
     # TEMP CI-budget kill-switch — see the SKIP_HEAVY_CI note at top of file.

@@ -2,7 +2,7 @@
 
 # RelyLoop BACKLOG Dashboard
 
-_Reflects feature-folder state as of **2026-06-09** (latest mtime of any planned/implemented feature `.md` file). Regenerated by `make dashboard` and the `mvp1-dashboard-regen` pre-commit hook. For the rich local view (filter chips, type colors), open [`backlog_dashboard.html`](backlog_dashboard.html) in a browser._
+_Reflects feature-folder state as of **2026-06-16** (latest mtime of any planned/implemented feature `.md` file). Regenerated by `make dashboard` and the `mvp1-dashboard-regen` pre-commit hook. For the rich local view (filter chips, type colors), open [`backlog_dashboard.html`](backlog_dashboard.html) in a browser._
 
 ## Next up
 

@@ -1,6 +1,6 @@
 # RelyLoop — Release Roadmap
 
-_Top-level index across MVP1 → GA v1+ as of **2026-06-09**. Click a release name to drill into the per-release dashboard. Theme labels sourced from [`docs/01_architecture/tech-stack.md` §"Canonical release matrix"](../01_architecture/tech-stack.md). For the rich local view, open [`dashboard.html`](dashboard.html) in a browser._
+_Top-level index across MVP1 → GA v1+ as of **2026-06-16**. Click a release name to drill into the per-release dashboard. Theme labels sourced from [`docs/01_architecture/tech-stack.md` §"Canonical release matrix"](../01_architecture/tech-stack.md). For the rich local view, open [`dashboard.html`](dashboard.html) in a browser._
 
 ## Releases
 

@@ -2,7 +2,7 @@
 
 # RelyLoop GA Dashboard
 
-_Reflects feature-folder state as of **2026-06-09** (latest mtime of any planned/implemented feature `.md` file). Regenerated by `make dashboard` and the `mvp1-dashboard-regen` pre-commit hook. For the rich local view (filter chips, type colors), open [`ga_dashboard.html`](ga_dashboard.html) in a browser._
+_Reflects feature-folder state as of **2026-06-16** (latest mtime of any planned/implemented feature `.md` file). Regenerated by `make dashboard` and the `mvp1-dashboard-regen` pre-commit hook. For the rich local view (filter chips, type colors), open [`ga_dashboard.html`](ga_dashboard.html) in a browser._
 
 ## Next up
 

@@ -2,7 +2,7 @@
 
 # RelyLoop MVP1 Dashboard
 
-_Reflects feature-folder state as of **2026-06-09** (latest mtime of any planned/implemented feature `.md` file). Regenerated by `make dashboard` and the `mvp1-dashboard-regen` pre-commit hook. For the rich local view (filter chips, type colors), open [`mvp1_dashboard.html`](mvp1_dashboard.html) in a browser._
+_Reflects feature-folder state as of **2026-06-16** (latest mtime of any planned/implemented feature `.md` file). Regenerated by `make dashboard` and the `mvp1-dashboard-regen` pre-commit hook. For the rich local view (filter chips, type colors), open [`mvp1_dashboard.html`](mvp1_dashboard.html) in a browser._
 
 ## Next up