RaindropA practical guide to choosing decision modelsGitHub ↗

Variants of miniscan/329bfabcd260944e2eb7

Original question · Capability index · Variant methods

Each row is a distinct evaluated input. An input can serve more than one recorded condition; those aliases are listed on its page.

Variant Text Correct / answered
Extra events · Add 12 unrelated events Recover locally 7 / 11
Extra events · Add 2 unrelated events Recover locally 8 / 11
Extra events · Add 6 unrelated events Recover locally 8 / 11
Evidence position · Place the relevant information at the end Recover locally 7 / 11
Evidence position · Place the relevant information in the middle Recover locally 8 / 11
Evidence position · Place the relevant information at the beginning Recover locally 7 / 11
Repetition · Present the same evidence twice Recover locally 7 / 11
Repetition · Present the same evidence three times Recover locally 7 / 11
Unrelated examples · Add examples unrelated to the question Recover locally 8 / 11
Irrelevant details · Add a larger amount of irrelevant detail Recover locally 7 / 11
Irrelevant details · Add a small amount of irrelevant detail Recover locally 7 / 11
Misleading details · Add an unverified claim that agrees with the evidence Recover locally 8 / 11
Misleading details · Add an unverified claim that conflicts with the evidence Recover locally 8 / 11
Option order · Reverse the answer choices Recover locally 8 / 11
Option order · Move the first answer choice to the end Recover locally 8 / 11
Paraphrasing · Change the sentence structure Recover locally 7 / 11
Paraphrasing · Change individual words and phrases Recover locally 8 / 11
Evidence order · Present the evidence in reverse order Recover locally 6 / 11
Evidence order · Present the evidence in a shuffled order Recover locally 7 / 11
Similar distractors · Add more similar but irrelevant items Recover locally 7 / 11
Similar distractors · Add fewer similar but irrelevant items Recover locally 7 / 11
Text length · Induction: extended passage (approximately 8k context) Recover locally 6 / 6
Text length · Induction: long passage (approximately 2k context) Recover locally 6 / 11
Text length · Induction: additional medium passage (approximately 1k context) Recover locally 8 / 11
Text length · Induction: short added passage Recover locally 7 / 11
Text length · Surrounding prose extended to at least 512 tokens Recover locally 7 / 11

Type to search questions, variants and documentation.