Refresh PolicyBench references and rankings

[MaxGhenis]

Summary

• refresh US and UK snapshot reference outputs with current PolicyEngine packages while preserving the frozen 2026 scenario manifests
• update the default Claude Opus benchmark alias from 4.7 to 4.8, with a legacy 4.7 app fallback for pre-rerun result files
• regenerate analysis summaries, dashboard data, and manifest hashes on top of current main
• refresh row/case audit annotations against the new wrong-row set
• pin @policyengine/ui-kit to the merged GitHub commit with the light-mode text contrast fix, so this PR does not depend on an npm publish
• make the removed US SPM input compatibility adjustment reproducible in reference-outputs metadata

PolicyEngine refs

• policyengine.py 4.10.0
• policyengine-us 1.705.1 at 7a7791f7a71e53629ff7b682a6960f3ab3a9e594
• policyengine-uk 2.88.22 at 7445869cfed59248be53778588856c2d688b34be

Verification

• uv run --no-sync pytest -q tests/test_scenarios.py tests/test_snapshot_artifacts.py tests/test_analysis.py
• uv run --no-sync pytest -q tests/test_eval_no_tools.py tests/test_chunked_eval.py tests/test_cli_helpers.py
• uv run --no-sync ruff check policybench/cli.py policybench/policyengine_runtime.py policybench/scenarios.py tests/test_scenarios.py
• uv run --no-sync ruff check policybench/config.py policybench/eval_no_tools.py tests/test_eval_no_tools.py tests/test_chunked_eval.py
• bun run lint
• bun test tests
• bun run build
• GitHub CI: lint, test, and app passed
• Vercel preview passed

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
policybench-site	Ready	Preview, Comment	Jun 7, 2026 7:26pm