ci.yml

name: CI

on:
  pull_request:
  push:
    branches: [main]

jobs:
  test:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-python@v5
        with:
          python-version: '3.11'

      - name: Install
        run: pip install -e ".[dev]"

      - name: Lint changed verifier files
        run: |
          ruff check \
            src/openamp_foundry/reproducibility.py \
            tests/release/test_run_manifest_verifier.py

      - name: Unit + integration tests
        run: pytest -q --tb=short

      - name: Demo pipeline
        run: make demo

      - name: Validate evidence certificates
        run: |
          for cert in outputs/evidence/*.json; do
            PYTHONPATH=src python -m openamp_foundry.cli validate \
              --certificate "$cert" \
              --schema schemas/candidate.schema.json
          done

      - name: Leakage check (informational)
        run: make bench-leakage

      - name: Doc link check (tolerates current 114 broken, fails on regressions)
        run: python scripts/check_doc_links.py --max-errors 10

      - name: Benchmark regression gate (AUROC drop <= 0.02 vs committed baseline)
        run: make bench-gate

      - name: Multi-negative-set benchmark (composition-independent sets > 0.70)
        run: make bench-multi-negatives

      - name: Hidden-active benchmark (require EF >= 1.5 at k=5)
        run: |
          PYTHONPATH=src python -c "
          import json, subprocess, sys, os
          result = subprocess.run(
              ['python', '-m', 'openamp_foundry.cli', 'bench', 'baseline',
               '--candidates', 'examples/benchmark/mixed_candidates.csv',
               '--positives', 'examples/benchmark/active_labels.csv',
               '--k', '5'],
              capture_output=True, text=True, check=True,
              env={**os.environ, 'PYTHONPATH': 'src'}
          )
          data = json.loads(result.stdout)
          ef = next(r['enrichment_factor'] for r in data['results'] if r['k'] == 5)
          print(f'Enrichment factor at k=5: {ef}')
          if ef < 1.5:
              print(f'FAIL: EF={ef} below minimum 1.5')
              sys.exit(1)
          print('PASS: Pipeline outperforms random ranker on hidden-active benchmark')
          "

      - name: Expanded benchmark (500 AMP + 500 decoy, AUROC > 0.70)
        run: |
          PYTHONPATH=src python -m openamp_foundry.cli validate-scoring \
            --amp-csv examples/validation/known_amps_500.csv \
            --decoy-csv examples/validation/random_background_500.csv \
            --out outputs/validate_scoring_500_ci.json
          PYTHONPATH=src python -c "
          import json
          data = json.load(open('outputs/validate_scoring_500_ci.json'))
          auroc = data['auroc']
          print(f'Expanded benchmark AUROC: {auroc:.4f}')
          if auroc < 0.70:
              print(f'FAIL: Expanded benchmark AUROC {auroc:.4f} < 0.70')
              exit(1)
          print(f'PASS: Expanded benchmark AUROC {auroc:.4f} >= 0.70')
          "

      - name: Easy baseline benchmark (informational — compares pipeline to trivial features)
        run: |
          make bench-easy-baseline
          echo "Easy baseline is informational — does not gate CI."

      - name: Order-dependent features benchmark (informational — analyzes which features survive scrambling)
        run: |
          make bench-order-dependent
          echo "Order-dependent benchmark is informational — does not gate CI."

      - name: Per-family benchmark (informational — stratifies 500 AMPs by structural class, per-class AUROC)
        run: |
          make bench-per-family
          echo "Per-family benchmark is informational — does not gate CI."

      - name: Policy version validation guard (fails if recalibration_policy.yaml changes without a version-bump decision log)
        run: |
          git fetch origin main
          if ! git diff --quiet origin/main -- configs/recalibration_policy.yaml; then
            echo "configs/recalibration_policy.yaml has changed — validating version bump..."
            git show origin/main:configs/recalibration_policy.yaml > /tmp/prev_policy.yaml
            PYTHONPATH=src python -m openamp_foundry.cli validate-policy-version \
              --current-policy configs/recalibration_policy.yaml \
              --previous-policy /tmp/prev_policy.yaml \
              --decision-log-dir decision_logs/
            echo "Policy version validation PASSED."
          else
            echo "configs/recalibration_policy.yaml unchanged — skipping policy version validation."
          fi