Merge origin/main and integrate research joins

Refactor to align on benchmark hierarchy

Add researcher join analysis to eval detail

Add DuckDB shadow-read backend with source-metadata fix

Separate policy and researcher views

Add interpretive signals, corpus dashboard, and slice browser

improve ux

Differentiate audience modes and tighten eval navigation

Aggregate setup aliases and clarify benchmark variants

Improve eval/model UX, lite data paths, and leaderboard clarity

Add per-benchmark comparison histograms on model detail

Improve eval score displays and summary fallbacks

Refresh eval cards UI and backend data flow

fix bugs

fix: align reporting cues and developer slugs

feat: refine model and benchmark exploration

redesigned