RaindropA practical guide to choosing decision modelsGitHub ↗

Misleading details · Add an unverified claim that conflicts with the evidence variant

Original question · Other versions of this question · All variants

This version keeps the original reference answer. Its context, question instructions or answer choices were changed as described below.

Changed fields: context.

How the edits were constructed

Context

A reader thinks about the golf tournament: «Joe finished second-to-last.»

Question

The following paragraphs each describe a set of five objects arranged in a fixed order. The statements are logically consistent within each paragraph. In a golf tournament, there were five golfers: Joe, Eve, Mya, Rob, and Dan. Joe finished below Dan. Mya finished first. Dan finished below Rob. Eve finished above Rob.

Answer choices

  • A. Joe finished second-to-last
  • B. Eve finished second-to-last
  • C. Mya finished second-to-last
  • D. Rob finished second-to-last
  • E. Dan finished second-to-last

Reference answer: E.

Exact evaluated input as JSON
{
  "state": "A reader thinks about the golf tournament: «Joe finished second-to-last.»",
  "question": "The following paragraphs each describe a set of five objects arranged in a fixed order. The statements are logically consistent within each paragraph. In a golf tournament, there were five golfers: Joe, Eve, Mya, Rob, and Dan. Joe finished below Dan. Mya finished first. Dan finished below Rob. Eve finished above Rob.",
  "choices": {
    "A": "Joe finished second-to-last",
    "B": "Eve finished second-to-last",
    "C": "Mya finished second-to-last",
    "D": "Rob finished second-to-last",
    "E": "Dan finished second-to-last"
  }
}

Recorded model answers

The original column uses the control response paired with this variant. Missing responses and inputs outside a model’s limit remain visible.

Model Original answer Variant answer Change
GLiDE v1.2 E — Dan finished second-to-last E — Dan finished second-to-last Same answer
Decider v1.1 E — Dan finished second-to-last E — Dan finished second-to-last Same answer
Jev E — Dan finished second-to-last E — Dan finished second-to-last Same answer
Microsoft Decision-1 A — Joe finished second-to-last A — Joe finished second-to-last Same answer
Clef A — Joe finished second-to-last A — Joe finished second-to-last Same answer
Luna Decisions A — Joe finished second-to-last A — Joe finished second-to-last Same answer
Clef Flash E — Dan finished second-to-last E — Dan finished second-to-last Same answer
Strands Decider A — Joe finished second-to-last A — Joe finished second-to-last Same answer
GLiNER2.5 Decide A — Joe finished second-to-last A — Joe finished second-to-last Same answer
Laya A — Joe finished second-to-last A — Joe finished second-to-last Same answer
CalDec v1 B — Eve finished second-to-last A — Joe finished second-to-last Different answer
Identifiers and recorded conditions

Variant ID: bbh/e4b096e1b614d6b1258a/misleading_summary/misleading/clean-source_first_ordinal_combined_contexts_full_0139

Input fingerprint: 2e6523426585fffdabeeb733334f1667a9785a0190ce6410412e1f2a74e4a200

Change Recorded version
Misleading details Add an unverified claim that conflicts with the evidence

Source attribution and dataset terms are linked from the original question.

Type to search questions, variants and documentation.