feat: add model metadata sidecar + merge for leaderboard

scripts/generate_leaderboard.py

@@ -15,6 +15,9 @@ import os
 import re
 import sys
 from collections import defaultdict
+from datetime import date
+
+import yaml
 
 
 # Known exam names that may contain underscores
@@ -90,6 +93,30 @@ def load_summary_jsonl(filepath: str) -> list[dict]:
     return records
 
 
+def load_model_metadata(path: str) -> dict:
+    """Load the model metadata sidecar (model id -> dict). Returns {} if absent."""
+    if not path or not os.path.exists(path):
+        return {}
+    with open(path, "r") as f:
+        return yaml.safe_load(f) or {}
+
+
+def merge_metadata(entries: list[dict], metadata: dict) -> list[str]:
+    """Attach metadata to each entry under 'meta'. Return model ids lacking metadata."""
+    missing: list[str] = []
+    seen: set[str] = set()
+    for e in entries:
+        meta = metadata.get(e["model"])
+        if meta is None:
+            e["meta"] = {}
+            if e["model"] not in seen:
+                seen.add(e["model"])
+                missing.append(e["model"])
+        else:
+            e["meta"] = meta
+    return missing
+
+
 def extract_max_score_from_md(filepath: str) -> int | None:
     """Extract max possible score from summary.md's Overall Score line.
 

scripts/model_metadata.yaml

@@ -0,0 +1,46 @@
+# Model metadata sidecar for the HTML leaderboard.
+# Keyed by exact OpenRouter model id. All fields optional.
+#   open_weight: true|false   -> open/closed badge
+#   contamination: clean|suspect   -> 'suspect' adds a footnote
+#   quant: <str>   -> footnote noting served precision
+#   lab / license / release_date / provider_pinned -> shown inline or in tooltip
+
+"google/gemini-3.1-pro-preview":
+  lab: Google
+  open_weight: false
+  release_date: 2026-01
+  contamination: clean
+"moonshotai/kimi-k2.6":
+  lab: Moonshot
+  open_weight: true
+  contamination: clean
+"moonshotai/kimi-k2.5":
+  lab: Moonshot
+  open_weight: true
+"qwen/qwen3.7-plus":
+  lab: Alibaba
+  open_weight: false
+  contamination: suspect      # released after exam — not ruled out
+"qwen/qwen3-vl-235b-a22b-thinking":
+  lab: Alibaba
+  open_weight: true
+  license: Apache-2.0
+  release_date: 2025-09-22
+  provider_pinned: alibaba
+  contamination: clean
+"z-ai/glm-4.6v":
+  lab: Z.ai
+  open_weight: true
+  release_date: 2025-12
+  provider_pinned: z-ai
+  quant: fp8                   # official Z.AI endpoint is fp8
+  contamination: clean
+"minimax/minimax-m3":
+  lab: MiniMax
+  open_weight: true
+"google/gemma-4-31b-it":
+  lab: Google
+  open_weight: true
+"xiaomi/mimo-v2.5":
+  lab: Xiaomi
+  open_weight: true

tests/test_leaderboard_html.py

@@ -0,0 +1,21 @@
+import json
+import os
+import sys
+
+import pytest
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..", "scripts"))
+import generate_leaderboard as gl  # noqa: E402
+
+
+def test_merge_metadata_attaches_and_reports_missing():
+    entries = [{"model": "a/b"}, {"model": "c/d"}]
+    metadata = {"a/b": {"lab": "X", "open_weight": True}}
+    missing = gl.merge_metadata(entries, metadata)
+    assert entries[0]["meta"] == {"lab": "X", "open_weight": True}
+    assert entries[1]["meta"] == {}
+    assert missing == ["c/d"]
+
+
+def test_load_model_metadata_absent_returns_empty():
+    assert gl.load_model_metadata("does/not/exist.yaml") == {}