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

Add new component files and align app to EvalEval design system

Replace shadcn-styled UI elements with design system primitives

Remove unused public/peer-ranks.json

Add plain-language captions and mode-aware framing for policy readers

Align user-facing labels with paper terminology

Merge corpus dashboard into home as paper-aligned landing

Bake DuckDB envs into runner stage

Restore DuckDB-aware build cache logic

Point Dockerfile at production card_backend dataset

Fix Fibble Arena (and similar) suite link routing

Sort eval-detail filenames by codepoint for parity

Set LOCAL_PIPELINE_OUTPUT/HF_DATA_OFFLINE at Docker build time

Bake DuckDB build-time defaults into Dockerfile

Deploy DuckDB-backed frontend to