Representation-Boundary-Guided Scientific Hypothesis Generation: An Exploratory Framework and 200-Case Study

Abstract

Representation-Boundary-Guided Scientific Hypothesis Generation (RBG-HG) is an exploratory framework for using explicit representation-boundary detection as a practical mechanism for scientific hypothesis generation. The proposed mechanism identifies assumptions or commitments in a problem representation that may constrain the reachable hypothesis space, deliberately explores alternative representations, generates hypotheses made salient or reachable by those transformations, formulates distinguishing tests, and evaluates the resulting incremental hypothesis yield as representation gain. Positive, neutral, and negative representation probes can subsequently inform future representation search while preserving exploration of previously unknown boundary types. The framework was developed through iterative human-AI investigation during the development of the Solverscue project and explored across approximately 200 heterogeneous scientific frontier cases. A structured second set comprises 100 new cases and 300 retained raw representation-boundary-guided hypotheses with associated representation shifts, distinguishing tests, and exploratory ratings. This deposit establishes a timestamped, falsifiable description of the framework, its proposed hypothesis-generation mechanism, exploratory evidence, protocol, and supporting corpus before confirmatory experimentation. The reported ratings are exploratory AI-assisted assessments and are not probabilities of scientific truth or independent scientific validation. The principal next test is a controlled comparison between ordinary hypothesis generation, matched creativity-enhanced hypothesis generation, and representation-boundary-guided hypothesis generation under comparable information and inference conditions.