Update Space to n=5 run (2026-07-19): ranking-led report, /30 scores, 30 runs, appendix at bottom

Invert scale: quality score 1-18, higher = better (report + both charts)

Move pass/fail appendix to the very bottom, after the Runs section

Charts: add footnote that scoring is joint (per-attempt judging is brittle)

Explain why we judge together: single-judge scoring is brittle with current rubric

Charts now use the single-judge ranking (mean rank + spread), not pass/fail

Remove limits section from report

Rewrite report: self-contained, PM- and scientist-readable; precise per-trial framing

Report: lead with single-judge ranking, demote pass/fail to appendix

Single-page site: fold all runs into index.html (fixes secondary-page 401)

Render PDF pages as inline images (fixes PDF viewing on static host)

Convert to static Space (no runtime): report + charts + 18 run PDFs/verdicts

Pin python 3.11 for gradio 5.9.1

Pin gradio 5.9.1 (fix py3.13 audioop / HfFolder crash)

Add Gradio app + deep-run artifacts (report, charts, 18 PDFs + verdicts)

initial commit