Launch Gloss as an open PowerPoint artifact benchmark

.gitattributes

@@ -0,0 +1,10 @@
+# Binary benchmark artifacts
+*.pptx binary
+*.png binary
+*.ttf binary
+*.ttc binary
+*.mp4 binary
+
+# Preserve licensed text and prompt-significant whitespace verbatim.
+acidslide-v1/benchmark/fonts/licenses/** -whitespace
+acidslide-v1/benchmark/prompts/variants/** -whitespace

.github/ISSUE_TEMPLATE/benchmark-case.yml

@@ -0,0 +1,50 @@
+name: Propose a benchmark case
+description: Add a slide behavior that generated PowerPoint artifacts should prove.
+title: "[case]: "
+labels: ["benchmark-case"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Thanks for making the corpus harder. Requirements must come from a public prompt, reference, or allowed asset — never from reverse-engineering the gold deck.
+  - type: textarea
+    id: failure
+    attributes:
+      label: What fails today?
+      description: Describe the generated deck behavior a screenshot or current check misses.
+      placeholder: A chart looks correct but is stored as a picture, so its data cannot be edited.
+    validations:
+      required: true
+  - type: textarea
+    id: requirement
+    attributes:
+      label: Proposed requirement
+      description: State the observable visual and/or native artifact requirement.
+    validations:
+      required: true
+  - type: dropdown
+    id: layer
+    attributes:
+      label: Primary layer
+      options:
+        - Native OOXML / package structure
+        - Rendered visual fidelity
+        - Both
+    validations:
+      required: true
+  - type: textarea
+    id: evidence
+    attributes:
+      label: Public source and evidence plan
+      description: Link the prompt/reference/asset source and describe a positive control plus a single-fault negative fixture.
+    validations:
+      required: true
+  - type: checkboxes
+    id: terms
+    attributes:
+      label: Contribution checks
+      options:
+        - label: I can share every attached artifact under the repository license or its stated compatible license.
+          required: true
+        - label: This proposal does not contain private decks, credentials, customer data, or proprietary fonts.
+          required: true

.github/ISSUE_TEMPLATE/config.yml

@@ -0,0 +1,8 @@
+blank_issues_enabled: true
+contact_links:
+  - name: Security vulnerability
+    url: https://github.com/aronchick/gloss/security/advisories/new
+    about: Report security issues privately, never in a public issue.
+  - name: Contribution guide
+    url: https://github.com/aronchick/gloss/blob/main/CONTRIBUTING.md
+    about: Read the evidence and fixture requirements before opening a pull request.

.github/ISSUE_TEMPLATE/evidence-challenge.yml

@@ -0,0 +1,37 @@
+name: Challenge evidence or a claim
+description: Question provenance, a mutation operator, a scoring assumption, or release wording.
+title: "[evidence]: "
+labels: ["evidence"]
+body:
+  - type: input
+    id: claim
+    attributes:
+      label: Claim or evidence ID
+      placeholder: gloss-technical-preview-2026-07-22 or a repository path
+    validations:
+      required: true
+  - type: textarea
+    id: challenge
+    attributes:
+      label: What does the current evidence fail to establish?
+      description: Separate observed facts from your interpretation.
+    validations:
+      required: true
+  - type: textarea
+    id: falsify
+    attributes:
+      label: How can we test or falsify it?
+      description: Propose a fixture, review, independent reproduction, or wording change.
+    validations:
+      required: true
+  - type: dropdown
+    id: urgency
+    attributes:
+      label: Release impact
+      options:
+        - Blocks any public claim
+        - Blocks an official benchmark freeze
+        - Improves the technical preview
+        - Documentation only
+    validations:
+      required: true

.github/ISSUE_TEMPLATE/grader-bug.yml

@@ -0,0 +1,46 @@
+name: Report a grader gap
+description: Reproduce a false positive, false negative, crash, or nondeterministic result.
+title: "[grader]: "
+labels: ["grader"]
+body:
+  - type: textarea
+    id: behavior
+    attributes:
+      label: Observable behavior
+      description: What did Gloss report, and what should it have reported?
+    validations:
+      required: true
+  - type: textarea
+    id: reproduce
+    attributes:
+      label: Minimal reproduction
+      description: Include exact `uv` commands, tier, platform, and a minimal safe fixture or repository path.
+      render: shell
+    validations:
+      required: true
+  - type: input
+    id: commit
+    attributes:
+      label: Gloss commit
+      placeholder: 40-character Git SHA
+    validations:
+      required: true
+  - type: dropdown
+    id: impact
+    attributes:
+      label: Result
+      options:
+        - False positive
+        - False negative
+        - Crash or incomplete verification
+        - Nondeterministic result
+        - Other
+    validations:
+      required: true
+  - type: checkboxes
+    id: safety
+    attributes:
+      label: Safety
+      options:
+        - label: This is not a security vulnerability. I will use the private process in SECURITY.md for vulnerabilities.
+          required: true

.github/pull_request_template.md

@@ -0,0 +1,23 @@
+## What changed
+
+<!-- Name the observable gap and the smallest change that closes it. -->
+
+## Evidence
+
+<!-- Link fixtures, reports, prompt requirements, screenshots, or source records. Do not invent model results or generation provenance. -->
+
+## Verification
+
+```text
+Exact commands run:
+
+Results:
+```
+
+## Release truth
+
+- [ ] Public claims in this pull request are traceable to committed evidence.
+- [ ] I did not imply that Draft candidate checks or local scores are official leaderboard results.
+- [ ] Grader behavior changes include focused tests or fixtures.
+- [ ] I checked for private decks, credentials, customer data, and proprietary fonts.
+- [ ] Documentation reflects any new command, contract, or contribution path.

.github/workflows/ci.yml

@@ -0,0 +1,127 @@
+name: CI
+
+on:
+  push:
+    branches: [main]
+    tags: ["acidslide-v*"]
+  pull_request:
+
+permissions:
+  contents: read
+
+jobs:
+  grader:
+    runs-on: ubuntu-22.04
+    defaults:
+      run:
+        working-directory: acidslide-v1/grader
+    steps:
+      - uses: actions/checkout@v4
+      - uses: astral-sh/setup-uv@v6
+        with:
+          version: "0.8.4"
+          enable-cache: true
+      - run: uv sync --extra dev --locked
+      - run: uv run ruff check .
+      - run: uv run ruff format --check .
+      - run: uv run mypy acidslide tests
+      - run: uv run pytest --cov=acidslide --cov-report=term-missing --cov-fail-under=85
+      - run: uv build
+      - run: uv run pip-audit
+      - name: Verify self-contained wheel data
+        run: |
+          uv run python -m zipfile -l dist/acidslide-1.0.0a1-py3-none-any.whl | grep 'acidslide/data/benchmark/checklist/deck.yaml'
+          uv run python -m zipfile -l dist/acidslide-1.0.0a1-py3-none-any.whl | grep 'acidslide/data/schemas/ecma-376/xsd-transitional/pml.xsd'
+
+  corpus:
+    runs-on: ubuntu-22.04
+    defaults:
+      run:
+        working-directory: acidslide-v1/grader
+    steps:
+      - uses: actions/checkout@v4
+      - uses: astral-sh/setup-uv@v6
+        with:
+          version: "0.8.4"
+          enable-cache: true
+      - run: uv sync --extra dev --locked
+      - run: uv run python ../benchmark/validate_corpus.py
+      - name: Require the complete frozen release on version tags
+        if: startsWith(github.ref, 'refs/tags/acidslide-v')
+        run: |
+          test "$GITHUB_REF_NAME" = "acidslide-v1.0.0"
+          uv run python ../benchmark/validate_corpus.py --release
+
+  launch:
+    runs-on: ubuntu-22.04
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-node@v4
+        with:
+          node-version: "22"
+      - name: Install launch media inspector
+        run: |
+          sudo apt-get update
+          sudo apt-get install --yes ffmpeg
+      - run: node launch/verify-launch.mjs
+
+  container:
+    runs-on: ubuntu-22.04
+    steps:
+      - uses: actions/checkout@v4
+      - run: docker build --platform linux/amd64 --tag acidslide-ci acidslide-v1
+      - name: Verify frozen runtime and font inventory
+        run: |
+          docker run --rm --platform linux/amd64 --network none acidslide-ci --version
+          docker run --rm --platform linux/amd64 --network none \
+            --entrypoint /bin/sh acidslide-ci -ec '
+              test "$(date -u +%Y-%m-%dT%H:%M:%SZ)" = "2025-01-01T00:00:00Z"
+              libreoffice --version | grep -F "LibreOffice 7.3.7.2"
+              pdftoppm -v 2>&1 | grep -F "22.02.0"
+              test "$(fc-list -f "%{file}\n" | sort -u | wc -l)" -eq 26
+            '
+      - name: Reference deck passes quarantine and ECMA-376 validation
+        run: |
+          output=$(docker run --rm --platform linux/amd64 --network none \
+            --volume "$PWD/acidslide-v1/benchmark/deck/gold/acidslide-v1-gold.pptx:/input.pptx:ro" \
+            acidslide-ci validate /input.pptx 2>&1)
+          grep -F "Quarantine passed" <<<"$output"
+          grep -F "ECMA-376 schema: valid" <<<"$output"
+
+  service:
+    runs-on: ubuntu-22.04
+    defaults:
+      run:
+        working-directory: acidslide-v1/service
+    steps:
+      - uses: actions/checkout@v4
+      - uses: astral-sh/setup-uv@v6
+        with:
+          version: "0.8.4"
+          enable-cache: true
+      - run: uv sync --extra dev --locked
+      - run: uv run ruff check .
+      - run: uv run ruff format --check .
+      - run: uv run mypy acidslide_service tests
+      - run: uv run pytest --cov=acidslide_service --cov-report=term-missing --cov-fail-under=85
+      - run: uv run pip-audit
+
+  service-images:
+    runs-on: ubuntu-22.04
+    steps:
+      - uses: actions/checkout@v4
+      - name: Build canonical grader base
+        run: |
+          docker build --platform linux/amd64 --provenance=false \
+            --tag acidslide/grader:1.0.0 acidslide-v1
+      - name: Build API, worker, and quarantine images
+        run: |
+          docker build --platform linux/amd64 --provenance=false \
+            --build-context schemas=acidslide-v1/schemas \
+            --target api --tag acidslide/service-api:1.0.0 acidslide-v1/service
+          docker build --platform linux/amd64 --provenance=false \
+            --build-context schemas=acidslide-v1/schemas \
+            --target worker --tag acidslide/service-worker:1.0.0 acidslide-v1/service
+          docker build --platform linux/amd64 --provenance=false \
+            --build-context schemas=acidslide-v1/schemas \
+            --target quarantine-job --tag acidslide/quarantine:1.0.0 acidslide-v1/service

.gitignore

@@ -21,3 +21,4 @@ htmlcov/
 
 # Large reference docs (download separately)
 ECMA/
+.gstack/