swe datasets: carry version/patch/hints/created_at in metadata

Official make_test_spec indexes MAP_REPO_VERSION_TO_SPECS[repo][version]
-- a None version KeyError'd the official scorer (first agentic run
scored 0 with 'KeyError: None' hidden in details). The base converter
never copied these fields; both variants now carry them, verified with
a live make_test_spec call (spec + 2KB eval script + correct image key).

Co-Authored-By: Claude <noreply@anthropic.com>

evalharness/data/datasets/swe_bench_verified.py

@@ -37,6 +37,10 @@ def _record_to_sample(record: dict) -> Sample:
             'PASS_TO_PASS': record['PASS_TO_PASS'],
             'environment_setup_commit': record.get('environment_setup_commit'),
             'difficulty': record.get('difficulty'),
+            'version': record.get('version'),   # official MAP_REPO_VERSION_TO_SPECS key
+            'patch': record.get('patch'),
+            'hints_text': record.get('hints_text'),
+            'created_at': record.get('created_at'),
         },
     )
 

evalharness/data/datasets/swe_bench_verified_agentic.py

@@ -21,6 +21,18 @@ from ..spec import DatasetSpec
     )
 )
 def swe_bench_verified_agentic():
+    def to_sample(record: dict):
         from .swe_bench_verified import _record_to_sample
 
-    return _record_to_sample
+        s = _record_to_sample(record)
+        # official make_test_spec requires these raw fields
+        md = s.metadata or {}
+        md.setdefault('version', record.get('version') or '')
+        md.setdefault('patch', record.get('patch') or '')
+        md.setdefault('hints_text', record.get('hints_text') or '')
+        md.setdefault('created_at', record.get('created_at') or '')
+        md['problem_statement'] = record.get('problem_statement') or s.input_text
+        s.metadata = md
+        return s
+
+    return to_sample