RU Benchmark Review Dashboard

Static GitHub Pages dashboard for reviewing a 250-row RU-only benchmark sample stratified by:

• project
• source
• time
• chunk word count

What This Repo Contains

• docs/index.html
  • Interactive dashboard with 3 tabs:
    • Distribution: project/source/time/word-bucket visualizations
    • Expert Review: inspect a row and save human judgments
    • Project Assets: prompt/result source manifest
• docs/data/sample.json
  • Benchmark sample rows (latest snapshot)
• docs/data/sample_latest_summary.json
  • Sample distribution summary
• docs/data/project_assets_manifest.json
  • Prompt + result source references per project

Included Project Strata

• rnla-deterrence
• rnla-grey-zone
• wacko
• ruw-core-taxonomy
• ruw-transparent-battlefield
• ruw-red-lines

Project/source references are listed in the Project Assets tab and in docs/data/project_assets_manifest.json.

Expert Review Behavior

• Judgments are stored in browser localStorage (client-side, no backend).
• You can export all judgments to CSV via Download CSV.

Why Static

GitHub Pages serves static sites only. This implementation keeps the review workflow interactive in-browser without requiring a server runtime.

Local Preview

From repo root:

python3 -m http.server 8080

Open: http://localhost:8080/docs/

Deployment

This repo uses a GitHub Actions Pages workflow (.github/workflows/pages.yml) that deploys the docs/ folder.

Data Refresh Workflow

If you regenerate benchmark data upstream, refresh these files:

• docs/data/sample.json
• docs/data/sample_latest_summary.json
• docs/data/project_assets_manifest.json

Then commit + push to update Pages.