Searching for What Matters: Open-Ended Discovery of Multiple Hypotheses with Agentic Data Exploration
What this work does
Searches a large dataset for multiple testable mechanisms using an agentic residual-driven discovery process that removes explained variation before subsequent searches.
Connection to discovery
Directly studies choosing what to investigate and accumulating multiple evidence-supported hypotheses.
Evidence
The body defines hypothesis identities and interventional support, constructs planted-mechanism tasks with hidden SCMs, and compares precision/recall under proposal-budget controls.
Scope and limitations
A designated outcome, finite tabular schema and engineered hidden SCM constrain openness; model-assisted mechanism matching and observational search must be separated from interventional validation.
THE RESEARCH CONVERSATION
Discussion & reactions
Share a question, an insight, or a replication experience. Sign in with GitHub to join the conversation.