surgeval 0.3.0a0__tar.gz

surgeval-0.3.0a0/.dist-smoke/.gitignore

@@ -0,0 +1 @@
+*

surgeval-0.3.0a0/.dockerignore

@@ -0,0 +1,16 @@
+.git
+.github
+.venv
+.pytest_cache
+.mypy_cache
+.ruff_cache
+.coverage*
+__pycache__
+*.pyc
+build
+dist
+data
+scratch
+tests
+docs
+site

surgeval-0.3.0a0/.github/workflows/ci.yml

@@ -0,0 +1,68 @@
+name: CI
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+  workflow_dispatch:
+
+concurrency:
+  group: ci-${{ github.ref }}
+  cancel-in-progress: true
+
+permissions:
+  contents: read
+
+jobs:
+  lint:
+    name: Lint and type-check
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4
+      - uses: astral-sh/setup-uv@d4b2f3b6ecc6e67c4457f6d3e41ec42d3d0fcb86 # v5
+        with:
+          enable-cache: true
+      - name: Install
+        run: uv sync --all-extras
+      - name: Ruff format check
+        run: uv run ruff format --check .
+      - name: Ruff lint
+        run: uv run ruff check --output-format=github .
+      - name: Mypy
+        run: uv run mypy
+
+  test:
+    name: Test (py${{ matrix.python-version }})
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        python-version: ["3.11", "3.12", "3.13"]
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4
+      - uses: astral-sh/setup-uv@d4b2f3b6ecc6e67c4457f6d3e41ec42d3d0fcb86 # v5
+        with:
+          enable-cache: true
+      - name: Install
+        run: uv sync --all-extras --python ${{ matrix.python-version }}
+      - name: Pytest
+        run: uv run --python ${{ matrix.python-version }} pytest --cov --cov-report=term-missing --cov-fail-under=90
+
+  packaging:
+    name: OSS packaging gates
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4
+      - uses: astral-sh/setup-uv@d4b2f3b6ecc6e67c4457f6d3e41ec42d3d0fcb86 # v5
+        with:
+          enable-cache: true
+      - name: Runtime license allowlist
+        run: uv run python scripts/check_license_allowlist.py
+      - name: Identity consolidation + wheel contents
+        run: uv run python scripts/check_identity.py
+      - name: SBOM generation + structural SPDX-2.3 validation
+        run: |
+          uv run python scripts/gen_sbom.py --out sbom.spdx.json
+          uv run python scripts/validate_sbom.py --sbom sbom.spdx.json
+      - name: Validate neutral example taskset
+        run: uv run surgeval tasksets validate docs/examples/tasksets/consequence-rank-v1

surgeval-0.3.0a0/.github/workflows/droid-review.yml

@@ -0,0 +1,31 @@
+name: Droid Auto Review
+
+on:
+  pull_request:
+    types: [opened, ready_for_review, reopened]
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  droid-review:
+    if: github.event.pull_request.draft == false
+    runs-on: ubuntu-latest
+    permissions:
+      contents: write
+      pull-requests: write
+      issues: write
+      id-token: write
+      actions: read
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@fbc6f3992d24b796d5a048ff273f7fcc4a7b6c09 # v5
+        with:
+          fetch-depth: 1
+      - name: Run Droid Auto Review
+        uses: Factory-AI/droid-action@c6951275d02429bb172eadc7302352d21b45a694 # main
+        with:
+          factory_api_key: ${{ secrets.FACTORY_API_KEY }}
+          automatic_review: true
+          automatic_security_review: true

surgeval-0.3.0a0/.github/workflows/droid.yml

@@ -0,0 +1,38 @@
+name: Droid Tag
+
+on:
+  issue_comment:
+    types: [created]
+  pull_request_review_comment:
+    types: [created]
+  issues:
+    types: [opened, assigned]
+  pull_request_review:
+    types: [submitted]
+  pull_request:
+    types: [opened, edited]
+
+jobs:
+  droid:
+    if: |
+      (github.event_name == 'issue_comment' && contains(github.event.comment.body, '@droid')) ||
+      (github.event_name == 'pull_request_review_comment' && contains(github.event.comment.body, '@droid')) ||
+      (github.event_name == 'pull_request_review' && contains(github.event.review.body, '@droid')) ||
+      (github.event_name == 'issues' && (contains(github.event.issue.body, '@droid') || contains(github.event.issue.title, '@droid'))) ||
+      (github.event_name == 'pull_request' && (contains(github.event.pull_request.body, '@droid') || contains(github.event.pull_request.title, '@droid')))
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      pull-requests: write
+      issues: write
+      id-token: write
+      actions: read
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@fbc6f3992d24b796d5a048ff273f7fcc4a7b6c09 # v5
+        with:
+          fetch-depth: 1
+      - name: Run Droid Exec
+        uses: Factory-AI/droid-action@c6951275d02429bb172eadc7302352d21b45a694 # main
+        with:
+          factory_api_key: ${{ secrets.FACTORY_API_KEY }}

surgeval-0.3.0a0/.github/workflows/install-smoke.yml

@@ -0,0 +1,238 @@
+name: install-smoke
+
+# Deliberately a separate workflow from kernel CI (§6.5 pattern).
+#
+# Kernel CI must stay fast, hermetic, and green on every PR: it proves the
+# harness contract. This workflow proves something different and structurally
+# flakier — that the *install ladder* still works against the outside world.
+# Its jobs depend on registries, vendor accounts, EULA gates, and GPU hosts we
+# do not control, so a broken upstream must never be able to red-flag a kernel
+# change. Hence: no `push`/`pull_request` trigger, manual dispatch plus a
+# weekly schedule, and a plan-only default for anything that would touch a
+# third-party registry.
+#
+# One job per install strategy *the catalog actually uses*, discovered from the
+# data rather than hardcoded here. The list moved once already — the audit
+# deleted both `pip-extra` rows (neither package is on PyPI) and added
+# `source-build` — and a hardcoded matrix answered that by failing on an empty
+# strategy instead of by testing the new one. Plus a real `surgeval quickstart`
+# job, the only one that runs a world end to end (CPU-only, no sim deps, no GPU).
+#
+# Until a strategy's entries are license-audited and pin-recorded, its plan
+# step *refuses*. That is a pass here: the ladder's job is to name the missing
+# artifact, and this workflow asserts it names one rather than guessing.
+
+on:
+  workflow_dispatch:
+  schedule:
+    # Mondays 05:17 UTC. Off-hour so a registry hiccup does not land in
+    # the middle of a review cycle.
+    - cron: "17 5 * * 1"
+
+permissions:
+  contents: read
+
+concurrency:
+  group: install-smoke-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  signals:
+    name: audited signal citations resolve at their pins
+    runs-on: ubuntu-latest
+    # Lives here, not in kernel CI, for the same reason as the ladder: it
+    # fetches third-party source trees over the network (git, against each
+    # catalog pin), so it must not sit in front of a PR. It is also the only
+    # check that can catch the class of defect it was written
+    # for - a citation pointing at code the pinned revision does not contain -
+    # because a prose evidence field cannot be checked by a machine.
+    #
+    # It fails closed, and being off the PR path is exactly what buys that: an
+    # unreachable upstream exits 2 and turns this job red. The alternative was
+    # tried and was worse - the first version printed SKIP and still exited 0
+    # as long as some other world checked, so a registry outage could have
+    # reported success while verifying none of the seven wrap targets. A green
+    # tick here means all of them were fetched and checked, or it means
+    # nothing at all.
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4
+      - uses: astral-sh/setup-uv@d4b2f3b6ecc6e67c4457f6d3e41ec42d3d0fcb86 # v5
+        with:
+          enable-cache: true
+      - name: Verify every audited env path and signal line at its world pin
+        # Exit 1 = a citation does not resolve. Exit 2 = something could not be
+        # checked. Neither is success; do not add continue-on-error here.
+        run: uv run python scripts/check_world_signals.py
+
+  quickstart:
+    name: quickstart (CPU-only, no sim deps)
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4
+      - uses: astral-sh/setup-uv@d4b2f3b6ecc6e67c4457f6d3e41ec42d3d0fcb86 # v5
+        with:
+          enable-cache: true
+
+      # The N10 contract is "first vector in minutes on any machine" in <=3
+      # commands, so this installs the distribution the way a user would rather
+      # than syncing the dev environment. The wheel force-includes the
+      # reference pair under `or_audit/_examples`, so every step below runs
+      # from a scratch directory with no checkout in sight: if quickstart only
+      # worked inside a clone, the ladder's first rung would be a lie.
+      - name: Install the distribution the way a user would
+        run: |
+          uv tool install .
+          echo "$HOME/.local/bin" >> "$GITHUB_PATH"
+
+      - name: Leave the checkout
+        run: mkdir -p "$RUNNER_TEMP/elsewhere"
+
+      - name: Doctor
+        working-directory: ${{ runner.temp }}/elsewhere
+        run: surgeval doctor
+
+      - name: Quickstart and record time-to-first-vector
+        working-directory: ${{ runner.temp }}/elsewhere
+        run: |
+          surgeval quickstart --out "$RUNNER_TEMP/quickstart" --json \
+            | tee "$RUNNER_TEMP/ttfv.json"
+
+      - name: Assert the metric is real and the vector replays
+        working-directory: ${{ runner.temp }}/elsewhere
+        run: |
+          python3 - "$RUNNER_TEMP/ttfv.json" > "$RUNNER_TEMP/replay.args" <<'PY'
+          import json
+          import sys
+
+          with open(sys.argv[1], encoding="utf-8") as handle:
+              record = json.load(handle)
+          assert record["time_to_first_vector_sec"] > 0, record
+          assert record["task_id"], record
+          print(record["out"], record["head"])
+          PY
+          read -r OUT HEAD < "$RUNNER_TEMP/replay.args"
+          surgeval replay "$OUT" --expect-head "$HEAD"
+
+      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4
+        with:
+          name: time-to-first-vector
+          path: ${{ runner.temp }}/ttfv.json
+
+  strategies:
+    name: discover strategies in use
+    runs-on: ubuntu-latest
+    outputs:
+      list: ${{ steps.read.outputs.list }}
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4
+      - uses: astral-sh/setup-uv@d4b2f3b6ecc6e67c4457f6d3e41ec42d3d0fcb86 # v5
+        with:
+          enable-cache: true
+      - name: Install
+        run: uv sync --all-extras
+      - name: Read the strategies the catalog uses
+        id: read
+        run: |
+          uv run python - <<'PY' >> "$GITHUB_OUTPUT"
+          import json
+
+          from or_audit.install.catalog import load_catalog
+
+          used = sorted({pkg.strategy.value for pkg in load_catalog().worlds})
+          assert used, "catalog has no worlds at all"
+          print(f"list={json.dumps(used)}")
+          PY
+
+  ladder:
+    name: plan ${{ matrix.strategy }}
+    runs-on: ubuntu-latest
+    needs: strategies
+    strategy:
+      fail-fast: false
+      matrix:
+        strategy: ${{ fromJSON(needs.strategies.outputs.list) }}
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4
+      - uses: astral-sh/setup-uv@d4b2f3b6ecc6e67c4457f6d3e41ec42d3d0fcb86 # v5
+        with:
+          enable-cache: true
+      - name: Install
+        run: uv sync --all-extras
+
+      - name: List the worlds using this strategy
+        run: uv run surgeval worlds list --strategy "${{ matrix.strategy }}"
+
+      - name: Plan every install for this strategy (dry run, no network)
+        env:
+          STRATEGY: ${{ matrix.strategy }}
+        run: |
+          uv run python - <<'PY'
+          import os
+          import subprocess
+          import sys
+
+          from or_audit.install.catalog import (
+              InstallStrategy,
+              VendorRuntimeInstall,
+              iter_packages,
+          )
+
+          strategy = InstallStrategy(os.environ["STRATEGY"])
+          packages = iter_packages(strategy=strategy)
+          assert packages, f"catalog has no {strategy.value} entries"
+
+          planned, refused = 0, 0
+          for pkg in packages:
+              # Accepting the vendor EULA in CI is exactly what a user does
+              # interactively, and only a vendor runtime has one to accept. The
+              # plan still pulls nothing without --execute. Derived per package
+              # rather than per matrix row so a mixed strategy cannot silently
+              # skip the acceptance path.
+              extra = (
+                  ["--accept-vendor-eula"]
+                  if isinstance(pkg.install, VendorRuntimeInstall)
+                  else []
+              )
+              argv = ["uv", "run", "surgeval", "worlds", "install", pkg.id, *extra]
+              done = subprocess.run(argv, capture_output=True, text=True)
+              sys.stdout.write(done.stdout)
+              sys.stderr.write(done.stderr)
+              if done.returncode == 0:
+                  planned += 1
+                  # A dry run must never execute anything.
+                  assert "dry run" in done.stdout, pkg.id
+              elif done.returncode == 1:
+                  refused += 1
+                  # A refusal is a pass only when it names the fix.
+                  assert "REFUSED:" in done.stderr, pkg.id
+                  assert "Fix:" in done.stderr or "re-run with" in done.stderr, pkg.id
+              else:
+                  raise SystemExit(f"{pkg.id}: unexpected exit {done.returncode}")
+          print(f"{strategy.value}: {planned} planned, {refused} refused with a named fix")
+          PY
+
+      - name: Doctor for each world using this strategy
+        env:
+          STRATEGY: ${{ matrix.strategy }}
+        run: |
+          uv run python - <<'PY'
+          import os
+          import subprocess
+
+          from or_audit.install.catalog import Disposition, InstallStrategy, iter_packages
+
+          strategy = InstallStrategy(os.environ["STRATEGY"])
+          for pkg in iter_packages(strategy=strategy):
+              if pkg.disposition not in {Disposition.SHIPPED, Disposition.WRAP}:
+                  continue
+              # Exit 1 is expected on a runner with no SOFA/Isaac; the point is
+              # that every reported problem carries a fix, which the report's
+              # own contract enforces.
+              done = subprocess.run(
+                  ["uv", "run", "surgeval", "doctor", "--world", pkg.id, "--json"],
+                  capture_output=True,
+                  text=True,
+              )
+              assert done.returncode in {0, 1}, (pkg.id, done.returncode, done.stderr)
+              print(pkg.id, "->", "ok" if done.returncode == 0 else "needs work")
+          PY

surgeval-0.3.0a0/.github/workflows/publish.yml

@@ -0,0 +1,198 @@
+name: Publish
+
+# Publisher-signature verification at the release-artifact layer.
+#
+# Trust model: no dormant in-band signing key. Publisher provenance is
+# established out-of-band and enforced end to end:
+#
+#   1. PyPI trusted publishing (OIDC) — secretless upload; GitHub's OIDC
+#      id-token proves workflow identity to PyPI.
+#   2. Sigstore build attestations, two independent records over the SAME
+#      built bytes:
+#      - a GitHub Artifact Attestation (``actions/attest-build-provenance``),
+#        verified cryptographically by ``gh attestation verify`` against
+#        Sigstore's public-good root (this is the publisher-signature check);
+#      - a PEP 740 attestation stored on PyPI by ``gh-action-pypi-publish``,
+#        whose *presence* and digest-consistency are confirmed by
+#        ``scripts/check_pypi_attestation.py`` (the crypto verification of the
+#        same bytes is the GitHub attestation above).
+#
+# Least privilege: only jobs that publish/attest hold ``id-token``; the build
+# job that executes project code is unprivileged and hands over immutable
+# artifacts via upload/download. Project code is never built or installed in a
+# job that holds OIDC release credentials.
+#
+# Task-registry (taskset/agent) signature verification is a separate trust
+# domain; see docs/OSS_GOVERNANCE.md.
+
+on:
+  push:
+    tags: ["v*"]
+  workflow_dispatch:
+    inputs:
+      version:
+        description: "Release version (e.g. 0.4.0). Must match pyproject/package version."
+        required: true
+        type: string
+
+permissions:
+  contents: read
+
+jobs:
+  build:
+    # Manual releases must originate on main; the tag-triggered path is the
+    # sole other entry (pushed v* tags). Skipping this job cascades to every
+    # dependent job, so a dispatch from a feature branch runs nothing.
+    if: github.event_name != 'workflow_dispatch' || github.ref == 'refs/heads/main'
+    runs-on: ubuntu-latest
+    outputs:
+      version: ${{ steps.version.outputs.value }}
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4
+      - uses: astral-sh/setup-uv@d4b2f3b6ecc6e67c4457f6d3e41ec42d3d0fcb86 # v5
+        with:
+          enable-cache: true
+      - name: Resolve release version
+        id: version
+        env:
+          DISPATCH_VERSION: ${{ github.event.inputs.version }}
+        run: |
+          if [ "${{ github.event_name }}" = "push" ]; then
+            echo "value=${GITHUB_REF_NAME#v}" >> "$GITHUB_OUTPUT"
+          else
+            # The dispatch input is user-controlled; never splice it into a
+            # shell program. Validate it looks like a version, then pass it
+            # through the environment rather than the command string.
+            if ! printf '%s' "$DISPATCH_VERSION" | grep -Eq '^[0-9][A-Za-z0-9._+]*$'; then
+              echo "invalid release version: $DISPATCH_VERSION" >&2
+              exit 1
+            fi
+            echo "value=$DISPATCH_VERSION" >> "$GITHUB_OUTPUT"
+          fi
+      - name: Enforce pre-publication version gate
+        run: uv run python scripts/check_identity.py --expected-version "${{ steps.version.outputs.value }}"
+      - name: Install
+        run: uv sync --all-extras
+      - name: Test
+        run: uv run pytest -q
+      - name: Verify runtime licenses (commercial allowlist)
+        run: uv run python scripts/check_license_allowlist.py
+      - name: Verify identity consolidation
+        run: uv run python scripts/check_identity.py
+      - name: Build sdist + wheel
+        run: uv build
+      - name: Twine metadata check
+        run: uvx twine check dist/*
+      - name: Generate SPDX SBOM
+        run: uv run python scripts/gen_sbom.py --out sbom.spdx.json
+      - name: Validate SBOM against SPDX 2.3 schema
+        run: |
+          curl -fsSL \
+            https://raw.githubusercontent.com/spdx/spdx-spec/v2.3/schemas/spdx-schema.json \
+            -o spdx-schema.json
+          uv pip install jsonschema
+          uv run python scripts/validate_sbom.py --sbom sbom.spdx.json --schema spdx-schema.json
+      - name: Record artifact digests
+        run: sha256sum dist/* sbom.spdx.json > RELEASE_DIGESTS.txt
+      - name: Upload build artifacts
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4
+        with:
+          name: surgeval-dist
+          path: |
+            dist/
+            sbom.spdx.json
+            RELEASE_DIGESTS.txt
+
+  attest:
+    name: Attest + verify (GitHub Sigstore)
+    runs-on: ubuntu-latest
+    needs: build
+    permissions:
+      contents: read
+      id-token: write
+      attestations: write
+    steps:
+      - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4
+        with:
+          name: surgeval-dist
+      - name: Attest build provenance (GitHub artifact attestation)
+        uses: actions/attest-build-provenance@ef244123eb79f2f7a7e75d99086184180e6d0018 # v1
+        with:
+          subject-path: dist/*
+      - name: Verify GitHub attestation (Sigstore public-good root)
+        env:
+          GH_TOKEN: ${{ github.token }}
+        run: |
+          for file in dist/*; do
+            gh attestation verify "$file" \
+              --repo "${{ github.repository }}" \
+              --predicate-type https://slsa.dev/provenance/v1
+          done
+
+  publish:
+    name: Publish to PyPI (minimal, OIDC)
+    runs-on: ubuntu-latest
+    needs: attest
+    environment: release
+    permissions:
+      contents: read
+      id-token: write
+    steps:
+      - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4
+        with:
+          name: surgeval-dist
+      - name: Publish to PyPI (trusted publishing + PEP 740 attestations)
+        uses: pypa/gh-action-pypi-publish@dc37677b2e1c63e2034f94d8a5b11f265b73ba33 # release/v1
+        with:
+          attestations: "true"
+
+  verify-pypi:
+    name: Check PyPI PEP 740 bundle + digest (unprivileged)
+    runs-on: ubuntu-latest
+    needs: [publish, build]
+    permissions:
+      contents: read
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4
+      - uses: astral-sh/setup-uv@d4b2f3b6ecc6e67c4457f6d3e41ec42d3d0fcb86 # v5
+        with:
+          enable-cache: true
+      - name: Sync (for scripts + uv)
+        run: uv sync --no-dev
+      - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4
+        with:
+          name: surgeval-dist
+      - name: Check PyPI serves built digests with PEP 740 bundles present
+        run: |
+          version="${{ needs.build.outputs.version }}"
+          # PEP 740 index entry may lag upload by a few seconds; retry before failing.
+          for attempt in 1 2 3 4 5; do
+            if uv run python scripts/check_pypi_attestation.py \
+              "$version" RELEASE_DIGESTS.txt \
+              --publisher-repo "${{ github.repository }}"; then
+              exit 0
+            fi
+            echo "provenance not yet visible (attempt $attempt); retrying"
+            sleep 10
+          done
+          exit 1
+
+  release:
+    runs-on: ubuntu-latest
+    needs: [attest, verify-pypi, build]
+    permissions:
+      contents: write
+    steps:
+      - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4
+        with:
+          name: surgeval-dist
+      - name: Create GitHub Release
+        uses: softprops/action-gh-release@3bb12739c298aeb8a4eeaf626c5b8d85266b0e65 # v2
+        with:
+          tag_name: ${{ github.event_name == 'push' && github.ref_name || format('v{0}', needs.build.outputs.version) }}
+          files: |
+            dist/surgeval-*.whl
+            dist/surgeval-*.tar.gz
+            sbom.spdx.json
+            RELEASE_DIGESTS.txt
+          generate_release_notes: true

surgeval-0.3.0a0/.github/workflows/sbom.yml

@@ -0,0 +1,42 @@
+name: SBOM
+
+# On-demand / branch SPDX SBOM generation: resolves uv.lock into an SPDX 2.3
+# JSON document, validates it against the official SPDX 2.3 JSON schema plus
+# the SurgEval structural checks, gates the license allowlist, and uploads the
+# SBOM as an artifact. The release path (Publish.yml) also generates and
+# validates the SBOM; this workflow covers ad-hoc dispatch and branch pushes.
+
+on:
+  workflow_dispatch:
+  push:
+    branches: [main]
+
+permissions:
+  contents: read
+
+jobs:
+  sbom:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4
+      - uses: astral-sh/setup-uv@d4b2f3b6ecc6e67c4457f6d3e41ec42d3d0fcb86 # v5
+        with:
+          enable-cache: true
+      - name: Install
+        run: uv sync --all-extras
+      - name: Generate SPDX SBOM
+        run: uv run python scripts/gen_sbom.py --out sbom.spdx.json
+      - name: Validate against SPDX 2.3 schema
+        run: |
+          curl -fsSL \
+            https://raw.githubusercontent.com/spdx/spdx-spec/v2.3/schemas/spdx-schema.json \
+            -o spdx-schema.json
+          uv pip install jsonschema
+          uv run python scripts/validate_sbom.py --sbom sbom.spdx.json --schema spdx-schema.json
+      - name: License allowlist gate
+        run: uv run python scripts/check_license_allowlist.py
+      - name: Upload SBOM
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4
+        with:
+          name: surgeval-sbom
+          path: sbom.spdx.json

surgeval-0.3.0a0/.gitignore

@@ -0,0 +1,46 @@
+# Python
+__pycache__/
+*.py[cod]
+*.egg-info/
+build/
+dist/
+
+# Environments
+.venv/
+venv/
+
+# Tooling caches
+.pytest_cache/
+.mypy_cache/
+.ruff_cache/
+.coverage
+.coverage.*
+htmlcov/
+coverage.xml
+
+# Editors / OS
+.DS_Store
+.idea/
+.vscode/
+
+# Agent tool state
+.opencode/
+
+# Local data. Surgical media must never enter version control.
+/data/
+/scratch/
+*.mp4
+*.mkv
+*.mov
+*.avi
+*.dcm
+
+.projects/cache
+.projects/vault
+.projects/state.test.json
+.projects/state.local.test.json
+.env
+.env.*
+!.env.example
+.projects/
+.pnpm-store/