RaindropA practical guide to choosing decision modelsGitHub ↗

Invented names · Use familiar names 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, question instructions.

How the edits were constructed

Context

def transform (text ):
    text =text .upper ()
    uppercase_count =0 
    for character in text :
        if character .isupper ():
            uppercase_count +=1 
        else :
            return 'no'
    return uppercase_count //2 

Question

What does transform('ax') return?

Answer choices

  • A. 1
  • B. 0
  • C. -1
  • D. 3

Reference answer: A.

Exact evaluated input as JSON
{
  "state": "def transform (text ):\n    text =text .upper ()\n    uppercase_count =0 \n    for character in text :\n        if character .isupper ():\n            uppercase_count +=1 \n        else :\n            return 'no'\n    return uppercase_count //2 ",
  "question": "What does transform('ax') return?",
  "choices": {
    "A": "1",
    "B": "0",
    "C": "-1",
    "D": "3"
  }
}

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 A — 1 A — 1 Same answer
Decider v1.1 B — 0 A — 1 Different answer
Jev A — 1 A — 1 Same answer
Microsoft Decision-1 B — 0 B — 0 Same answer
Clef B — 0 B — 0 Same answer
Luna Decisions B — 0 B — 0 Same answer
Clef Flash B — 0 B — 0 Same answer
Strands Decider B — 0 B — 0 Same answer
GLiNER2.5 Decide A — 1 A — 1 Same answer
Laya A — 1 C — -1 Different answer
CalDec v1 A — 1 A — 1 Same answer
Identifiers and recorded conditions

Variant ID: cruxeval/dac9a78a4a34555b5ff0/unfamiliar_names/familiar

Input fingerprint: bd6af286ef9e829b331996a774392289a6e863407e72c795cbd58dd23d997f9d

Change Recorded version
Invented names Use familiar names

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

Type to search questions, variants and documentation.