Balanced but Misleading: Can Epistemic Scores Miss Decision-Changing Omissions?
Daud Ibrahim Hassan, Elvina Chow · Team APART1116
Submitted to AI x Epistemics Research Sprint. Sprint projects are early-stage work by participants, not Apart Research publications.
We tested whether AI briefings can preserve balanced evidence while omitting facts needed to make a decision. Across six fictional lease and job-offer scenarios, we evaluated three receiver models using neutral, promotional, balance-instructed, matched-random and full-evidence briefings.
All 18 balance-instructed briefings preserved favorable/adverse inclusion balance while omitting neutral-labeled facts. Among 930 valid receiver responses, 19 rejected both options despite a qualifying option. All 19 explanations cited missing qualification evidence, even though clarification was available and missing facts were explicitly described as unknown. These outputs covered four scenario families and seven distinct briefing/context inputs. All 69 valid full-evidence controls were correct.
A post-collection audit identified setup facts withheld during scenario conversion. Excluding these facts reduced balanced critical-omission counts from 17/18 and 13/18 to 4/18 and 3/18 across the two contexts. We report this sensitivity separately and release the frozen evidence, code and audit.
The finding distinguishes balanced presentation, sufficient qualification evidence and appropriate uncertainty: inability to verify an option does not establish that neither option qualifies.

Reviews
No public critique yet.
Cite this project
@misc{hassan2026balanced,
title = {{Balanced but Misleading: Can Epistemic Scores Miss Decision-Changing Omissions?}},
author = {Daud Ibrahim Hassan and Elvina Chow},
year = {2026},
month = nov,
note = {Submitted to AI x Epistemics Research Sprint, an Apart Research Sprint},
howpublished = {\url{https://apartresearch.com/sprints/projects/balanced-but-misleading-can-epistemic-scores-miss-decisionchanging-omissions-pzcn0t1rhf}},
url = {https://apartresearch.com/sprints/projects/balanced-but-misleading-can-epistemic-scores-miss-decisionchanging-omissions-pzcn0t1rhf}
}Browse all projects