Dataset Viewer
Auto-converted to Parquet Duplicate
episodes
int64
by_type
dict
review_image_paths_checked
int64
copied_files_hash_checked
int64
snapshots_byte_identical
bool
all_review_statuses
string
official_manifest_unchanged_sha256
string
extra_asset_folders
list
website_not_modified
bool
inverse_compare_episodes_added
int64
merged_hard30_accepted
int64
merged_hard30_inverse_compare
int64
90
{ "beneath": 13, "compare": 32, "count": 17, "find": 28 }
872
10,502
true
accepted
72344affeab5dd48dd3737edacf16042db35f40b37d4a8b6c8d7ff9eefee3c0f
[ "allsides_objects/allsides_stacking_ring_toy", "allsides_objects/allsides_turtle_figurine" ]
true
12
24
4

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

PROBE hard90

Human-reviewed hard development benchmark built from benchmark-hard50 plus accepted candidates from benchmark-hard30-review.

This bank contains 90 questions: beneath 13, compare 32, count 17, find 28. It is a hard development set, not a balanced or independent holdout.

Rejected hard30 candidates were not included. Accepted inverse compare variants were included as separate duplicate-scene questions.

Downloads last month
52