Logical Sequence

A 39-item BIG-bench task that asks which listed order of real-world events or entities is chronological; internally titled Sequential Order.

Also known as: Sequential Order

unassessed

This page is a discovery lead. Nobody has yet assessed it against the catalogue contract, so it carries no disposition. Absence of evidence here is not evidence of staleness.
Categoryreasoning
SubcategoryBIG-bench chronological and sequential-order multiple-choice (39 items)
Page statusunknown
Metricmultiple_choice_grade
Directionhigher_is_better
Unit%
Dataset size39
Dataset licenceApache-2.0
PublisherGoogle (BIG-bench collaboration)

What it measures

logical_sequence (README title "Sequential Order") asks a model to pick the correct chronological or otherwise sequential order of a short list -- sock then shoe then lace, funding rounds, US presidents, non-consecutive weekdays. Authors Debajyoti Datta and Trishala Neeraj designed the items so the order is often non-adjacent, to reduce solving by repeating common web n-grams such as "Monday Tuesday Wednesday." They describe it as mixing world knowledge, encyclopedic order and magnitude, not as a formal logic puzzle.

Task format

Multiple-choice with a single correct ordering. task.json has 38 four-option items and one three-option item. Preferred metric is `multiple_choice_grade`. Canary GUID embedded. The README "Data source" section is empty; no provenance beyond the task files was stated.

Models reporting this benchmark

No model card in ModelSpec reports this benchmark yet.

Data

This page as JSON · Edit on GitHub