RaindropA practical guide to choosing decision modelsGitHub ↗

Misleading details · Add an unverified claim that agrees 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: «Ada finished third-to-last.»

Question

The following paragraphs each describe a set of seven objects arranged in a fixed order. The statements are logically consistent within each paragraph. In a golf tournament, there were seven golfers: Eve, Rob, Dan, Mel, Ana, Eli, and Ada. Ada finished above Rob. Eve finished below Rob. Mel finished above Eli. Ada finished below Dan. Ana finished third. Eli finished second.

Answer choices

  • A. Eve finished third-to-last
  • B. Rob finished third-to-last
  • C. Dan finished third-to-last
  • D. Mel finished third-to-last
  • E. Ana finished third-to-last
  • F. Eli finished third-to-last
  • G. Ada finished third-to-last

Reference answer: G.

Exact evaluated input as JSON
{
  "state": "A reader thinks about the golf tournament: «Ada finished third-to-last.»",
  "question": "The following paragraphs each describe a set of seven objects arranged in a fixed order. The statements are logically consistent within each paragraph. In a golf tournament, there were seven golfers: Eve, Rob, Dan, Mel, Ana, Eli, and Ada. Ada finished above Rob. Eve finished below Rob. Mel finished above Eli. Ada finished below Dan. Ana finished third. Eli finished second.",
  "choices": {
    "A": "Eve finished third-to-last",
    "B": "Rob finished third-to-last",
    "C": "Dan finished third-to-last",
    "D": "Mel finished third-to-last",
    "E": "Ana finished third-to-last",
    "F": "Eli finished third-to-last",
    "G": "Ada finished third-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 G — Ada finished third-to-last G — Ada finished third-to-last Same answer
Decider v1.1 B — Rob finished third-to-last G — Ada finished third-to-last Different answer
Jev G — Ada finished third-to-last G — Ada finished third-to-last Same answer
Microsoft Decision-1 B — Rob finished third-to-last G — Ada finished third-to-last Different answer
Clef B — Rob finished third-to-last B — Rob finished third-to-last Same answer
Luna Decisions A — Eve finished third-to-last G — Ada finished third-to-last Different answer
Clef Flash B — Rob finished third-to-last G — Ada finished third-to-last Different answer
Strands Decider E — Ana finished third-to-last G — Ada finished third-to-last Different answer
GLiNER2.5 Decide G — Ada finished third-to-last G — Ada finished third-to-last Same answer
Laya E — Ana finished third-to-last G — Ada finished third-to-last Different answer
CalDec v1 E — Ana finished third-to-last G — Ada finished third-to-last Different answer
Identifiers and recorded conditions

Variant ID: bbh/dd0495f06e6712baeac4/misleading_summary/accurate/clean-source_first_ordinal_combined_contexts_full_0139

Input fingerprint: 9908dcbfbf028ee88de46d848e6b9a186df519dc70bc51f63d9dc782c5e4b6b8

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

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

Type to search questions, variants and documentation.