Publish registry data (seed @ cebc9be9)

aliases.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8e01b9ec5038d7e37d19b8936b341c7f3f922f2799a9cc3d3b825f6030519a39
-size 1178567
+oid sha256:9154cdc3236de3e7085d5ead071c77fee1f653577a071fb5b48bc9967702d572
+size 1196313

canonical_benchmarks.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:858316db3acd8a5b5cd98e5b096efc20f7e4dc7ce7eedb999a557bfd50e86bb8
-size 131689
+oid sha256:a59efb6d4891819a462a04fea946eff2e9e943a0cdb7d9188fff50a4ba14e92f
+size 135586

canonical_composites.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a213267c9241679352d2c88e421a8951c3766ee4e3a4c9ff95361fe1a04c6249
-size 8262
+oid sha256:409dde753f3f7b6293df804fdbfeeae1119679fc7d308c26700aa613da0c9aa2
+size 8671

canonical_families.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5520abae73056acfc5ddd943314aab237754a0e01e77af5ae28c4cada24bb8c8
-size 9025
+oid sha256:d68884df101637befed2e189855c0f7f6ee52673ed07464ea5d2623c3c9a01aa
+size 9384

canonical_metrics.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1539221023006422373973032c4bd7a7177fc9a39ab2dfc05232c3d82282b7b0
-size 7269
+oid sha256:c5e6307a7b93c6a302fb75e381f3329925ef8679b27ea2974a29d7e29fbceeb2
+size 7692

canonical_models.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1341e7ded05b89d58f645f4466364da80cdc0566796a5b3cc51da5cad452828a
-size 117949
+oid sha256:bf922613562c669dfa24c65caecfac8189cbb442b03ea33b6c44d70dc1324a4e
+size 117460

canonical_orgs.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f50edf47d333642a8bfee3bd5fbf95f985a2f327c5e1800f8b376a2ae4ead51e
-size 9139
+oid sha256:4f57770faa71bccc0f6bb496a8bc77a1a8b35b166ea6730feafc5519b5e12fec
+size 9155

display_overrides.yaml

@@ -1,8 +1,5 @@
 # Display-polish overrides: slugs that should render uppercase in human-
-# facing display rather than title-cased. Ports `ACRONYMS` from the
-# reference hierarchy builder (ref-build_hierarchy.py:152-158) and
-# replaces the SUITE_DISPLAY_NAMES hardcoded map in
-# general-eval-card/components/benchmark-detail.tsx.
+# facing display rather than title-cased.
 #
 # Producer / frontend behaviour: when a slug appears in this list, the
 # display name is `slug.upper()` rather than the default prettifier
@@ -58,3 +55,34 @@ acronyms:
   - scicode
   - hal
   - usaco
+
+# Benchmark display names: curated human-readable names for benchmarks
+# whose slug doesn't title-case cleanly. Keyed by the benchmark slug
+# (hyphens, not underscores). Applied during hierarchy generation when
+# the canonical entity doesn't already carry a curated display_name.
+benchmark_display_names:
+  hf-open-llm-v2: "HF Open LLM v2"
+  helm-lite: "HELM Lite"
+  helm-capabilities: "HELM Capabilities"
+  helm-classic: "HELM Classic"
+  helm-instruct: "HELM Instruct"
+  helm-mmlu: "HELM MMLU"
+  reward-bench: "RewardBench"
+  reward-bench-2: "RewardBench 2"
+  bfcl: "BFCL"
+  global-mmlu-lite: "Global MMLU Lite"
+  swe-bench: "SWE-bench"
+  arc-agi: "ARC-AGI"
+  tau-bench-2: "TAU-Bench 2"
+  ace: "ACE"
+  apex-agents: "APEX Agents"
+  apex-v1: "APEX v1"
+  appworld: "AppWorld"
+  browsecompplus: "BrowseComp+"
+  livecodebenchpro: "LiveCodeBench Pro"
+  sciarena: "SciArena"
+  terminal-bench-2-0: "Terminal Bench 2.0"
+  la-leaderboard: "LA Leaderboard"
+  theory-of-mind: "Theory of Mind"
+  fibble-arena: "Fibble Arena"
+  wordle-arena: "Wordle Arena"

eval_harnesses.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c7e07e7f7916214b1835be45d123ef322fd87ae6bafd3cc73284f2b438c4d4ab
-size 5557
+oid sha256:b6bccb92d059677d0e09c9f6f58c6236229460d16c48f01f47b4d351420477d3
+size 5563

manifest.json

@@ -1,13 +1,13 @@
 {
-  "content_hash": "5960c05b823f43f1d98dfc0a1e5195815c108342f00abe95603e5c6e55eb5128",
-  "generated_at": "2026-05-05T11:29:25+00:00",
+  "content_hash": "74e5f90c408a7c493a403961cc909ddf654688eeab26e6bf8d00bc75e20297f7",
+  "generated_at": "2026-05-28T21:51:58+00:00",
   "row_counts": {
-    "aliases": 16559,
-    "canonical_benchmarks": 2548,
-    "canonical_composites": 30,
-    "canonical_families": 14,
-    "canonical_metrics": 22,
-    "canonical_models": 1434,
+    "aliases": 16724,
+    "canonical_benchmarks": 2597,
+    "canonical_composites": 35,
+    "canonical_families": 21,
+    "canonical_metrics": 27,
+    "canonical_models": 1428,
     "canonical_orgs": 35,
     "eval_harnesses": 11,
     "eval_results": 0,
@@ -15,5 +15,5 @@
     "sync_runs": 0
   },
   "schema_version": "registry.2.0",
-  "seed_git_sha": "0d5013ed37c78a006cfcf4f4ee1a79e3b69a9f8c"
+  "seed_git_sha": "cebc9be99abd168e0d562cee9aec206065e760a7"
 }

slice_overrides.yaml

@@ -1,15 +1,8 @@
-# Slice grouping overrides. The default heuristic groups siblings by
-# stripping recognised suffixes (`-vN`, `-level-N`, `-vs-X`, …) and
-# bucketing benchmarks that share a stem. When the heuristic over-groups,
-# add the benchmark id to `promote_to_benchmark` to leave its
-# parent_benchmark_id NULL — promoting it to a sibling benchmark in the
-# family rather than a slice of the stem.
+# Slice grouping overrides. With suffix stripping disabled,
+# this list only needs entries that override a registry-authored
+# parent_benchmark_id — i.e. cases where benchmarks.yaml sets a parent
+# but the benchmark should actually be standalone.
 
 promote_to_benchmark:
-  - bfcl-live
-  - bfcl-multi-turn
-  - bfcl-non-live
-  - bfcl-web-search
   - gpqa-diamond
   - mmlu-pro
-  - swe-bench-verified