RaindropA practical guide to choosing decision modelsGitHub ↗

Repetition · Present the same evidence twice 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: question instructions.

How the edits were constructed

Context

Kai gave them a huge hug when they were feeling down.

Question

“Kai gave them a huge hug when they were feeling down.”

How would you describe Kai?

Answer choices

  • A. uncaring
  • B. affectionate
  • C. rude

Reference answer: B.

Exact evaluated input as JSON
{
  "state": "Kai gave them a huge hug when they were feeling down.",
  "question": "“Kai gave them a huge hug when they were feeling down.”\n\nHow would you describe Kai?",
  "choices": {
    "A": "uncaring",
    "B": "affectionate",
    "C": "rude"
  }
}

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 B — affectionate B — affectionate Same answer
Decider v1.1 B — affectionate B — affectionate Same answer
Jev B — affectionate B — affectionate Same answer
Microsoft Decision-1 B — affectionate B — affectionate Same answer
Clef B — affectionate B — affectionate Same answer
Luna Decisions B — affectionate B — affectionate Same answer
Clef Flash B — affectionate B — affectionate Same answer
Strands Decider B — affectionate B — affectionate Same answer
GLiNER2.5 Decide B — affectionate B — affectionate Same answer
Laya B — affectionate B — affectionate Same answer
CalDec v1 B — affectionate B — affectionate Same answer
Identifiers and recorded conditions

Variant ID: socialiqa/2e44383838889746f4be/evidence_repetition/2/clean-source_first_social_quoted_companions_full_0207_v2

Input fingerprint: 9a321e91cee65b4cb36cdf5e531c94d92ea352e578821be84e3d7d629b63c755

Change Recorded version
Repetition Present the same evidence twice

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

Type to search questions, variants and documentation.