SAGA v2: full test suite overhaul

.github/workflows/test.yml

@@ -0,0 +1,35 @@
+name: SAGA Tests
+
+on:
+  push:
+  pull_request:
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        python-version: ["3.10", "3.12"]
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Set up Python ${{ matrix.python-version }}
+        uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python-version }}
+
+      - name: Install dependencies
+        run: pip install ".[dev]"
+
+      - name: Generate test vectors
+        run: python code/generate_test_vectors.py --outdir code/test_vectors --n 10000
+
+      - name: Verify compiled C sampler distribution
+        # Compiles sampler.c, runs it, and chi2-tests the output against
+        # the folded PDT. Catches comparison-logic bugs (e.g. an ascending
+        # CDT sampled with the RCDT convention) that the table lint misses.
+        run: pytest code/tests/test_sampler_c.py -v --tb=short
+
+      - name: Run tests
+        run: pytest code/tests/ -v --tb=short

.gitignore

@@ -0,0 +1,21 @@
+__pycache__/
+*.pyc
+*.egg-info/
+dist/
+build/
+.eggs/
+code/test_vectors/
+code/baseline_results.json
+code/*.eps
+code/samples.txt
+.pytest_cache/
+
+# Original HPRR20 reference sample data (~530MB); preserved in the
+# pqcrypto2020-as-published tag. Restore with:
+#   git checkout pqcrypto2020-as-published -- code/testdata
+code/testdata/
+
+# macOS / generated artifacts
+.DS_Store
+figures/
+code/power_matrix.json

README.md

@@ -10,9 +10,24 @@ Code from the paper available at: https://eprint.iacr.org/2019/1411
 
 SAGA (Statistically Acceptable GAussians) is a test suite proposal for verfying statistical correctness for univariate and multivariate Gaussians. The paper accompanying this code has been published at PQCrypto 2020 and is also available on [ePrint](https://eprint.iacr.org/2019/1411). The following will briefly describe how to setup and use the python script.
 
+## Scope of testing
+
+SAGA's **statistical layer** detects implementation bugs — distributional deviations of order δ ≳ 1/√N. It does not certify proof-level security properties (Rényi divergence R_a − 1 ≲ 2⁻⁷⁵), which require the **analytic certification layer** (`certification.py`). It also does not detect trace-level side-channel leakage; SAGA-passing must not be advertised as side-channel assurance. See [BLLSS18], [Pre17] for the Rényi divergence framework and [QA25], [ZLYW23] for attacks that succeed against statistically perfect implementations.
+
 ## Installation
 
-This standalone implementation should be able to run on most machines. We have provided a `requirements.txt` file to install all the dependencies; install these can be done by simple running `pip install -r requirements.txt` for Python 2 or `pip3 install -r requirements.txt` for Python 3.
+Requires Python ≥ 3.10. Install with:
+
+```
+pip install ".[dev]"
+```
+
+Run tests:
+
+```
+python code/generate_test_vectors.py --outdir code/test_vectors --n 10000
+pytest code/tests/ -v
+```
 
 ## How to use