Structure Liberates: How Constrained Sensemaking Produces More Novel Research Output

2026-05-01

AI Summary

Scientific creativity generation is often oversimplified, lacking effective modeling of systematic thinking. This work proposes SCISENSE, a novel framework that introduces constrained sensemaking into the scientific ideation process, formalizing it into eight cognitive stages. The authors construct SCISENSE-Traj, a large-scale dataset, and train SCISENSE-LM—a suite of large language models spanning 3B to 70B parameters—using citation-conditioned trajectory reconstruction, reasoning, and model distillation. Remarkably, Target-mode training enhances both novelty and diversity of generated content (+2.0% trajectory quality) while maintaining goal-directedness, thereby significantly improving the quality of executable scientific artifacts produced by downstream agents. These findings challenge the conventional assumption that looser supervision better facilitates exploratory creativity.

Abstract

Scientific discovery is an extended process of ideation--surveying prior work, forming hypotheses, and refining reasoning--yet existing approaches treat this phase as a brief preamble despite its central role in research. We introduce SCISENSE, a sensemaking-grounded framework that operationalizes ideation as a structured sequence of eight cognitive stages (Pirolli \& Card, 2005). We construct SCISENSE-Traj, a 100K-scale dataset of citation-conditioned research trajectories in two modes: Target, where an LLM reconstructs the ideation path leading to a known paper from its cited works, and Infer, where the LLM proposes novel directions from the same citations. We distill these into SCISENSE-LM, a family of sensemaking LLMs spanning 3B to 70B parameters. Contrary to the assumption that looser supervision promotes greater exploration, Target-trained models achieve a 2.0\% improvement in trajectory quality over Infer-trained models while also producing more novel and diverse outputs. This advantage propagates downstream: coding agents conditioned on Target trajectories produce research artifacts with higher executability and quality than those conditioned on Infer trajectories. This suggests that targeted ideation reduces cognitive burden on downstream agents, freeing them to explore more creatively. SCISENSE offers both a practical tool for augmenting LLM-driven research workflows and a principled testbed for studying how planning shapes scientific discovery.

Problem

• scientific discovery
• ideation
• sensemaking
• research workflow
• cognitive burden

Innovation

• structured sensemaking
• scientific discovery
• research ideation
• trajectory learning
• LLM for science