add composite eval page

fix joins

fix data URL

minor fix of the matrix script

Merge origin/main and integrate research joins

Bump clean-hierarchy cache version to v13 to drop stale blob

Merge cross-source benchmark families; tidy leaderboard panel + table chrome

Drop alias-only single-bench families without merging them

Restore curated benchmark families; polish frontier panel UX

Live snapshot date, hide empty Updated col, clean slice contamination

Humanize family names whose display matches the key under different separators

Make /models tables column-sortable; rebalance /evals + /models toolbars

Clean up Source column and per-row dataset label noise

Hide subtask-scope metrics from chips by default in matrix view

Render score-distribution metric picker as chips, not a dropdown

Treat single-root-metric subtask evals as slice-pickable, not matrix

Precompute eval matrices for multi-metric + per-slice leaderboards

Restore HF Open LLM v2 composite and dedup vals.ai aliases

Move split selector below the reporting comparison heading

Add local parquet read support

Fix sort toggle direction and remove categories as sortable column

Sort evals list by family name; add sortable columns; use cleaned display names

Dedup logic to counts

Fix ranks-high/low-in using only sidecar ordinal data

Wire search bar to overlaps table and hide chips in overlaps view

Compute and apply cleaned benchmark counts per model

Remove raw-hierarchy fallback — only ever serve cleaned hierarchy

Harden cleanHierarchy fallback and add family-name filter chips

Bump clean-hierarchy cache version to v10 to bust stale HF Space cache

Restructure model details + extend cleanHierarchy for split families and aggregator dedup

Add list-view toggle to consolidate cross-family duplicate benchmarks

Square off deep-dive theme and surface cross-family duplicates

Add option to purge cache

stats change

Prefer /data persistent bucket for sidecar cache when available

Disk-cache snapshot sidecars to skip cold-start re-downloads

Switch family/model views to curated category tags

Route peer-ranks fetch through SNAPSHOT_URL sidecar

Hotfix: categories

Group model/eval-detail benchmarks by hierarchy.json families

Drop latest_timestamp fallback for release_date display

Guard summaryText against null in PolicyOverview

Wrap /evals page in Suspense for useSearchParams

Refactor to align on benchmark hierarchy

Remove unnecessary distinct() when reporting total results

Update with datafix v2

Consolidate hierarchy terminology + handle v2 hierarchy shape

Reconcile UI with v2 backend payload + drop redundant signal cards

Swap backend data (#3)

Tighten eval cards UI and clean up stale local data