Research Dossier: An Agent That Shows Its Disagreements Instead of Hiding Them
This is a submission for the Sanity Challenge, Path One: Ship an Agent That Queries Real Content . What I Built Research Dossier is a multi-agent research analyst built with LangGraph . Instead of asking one model to answer a research question from its own knowledge, the system routes the question through four stages: Research → Analysis → Writing → Review The research stage is grounded in a…
Title: An Agent That Displays Disagreements Instead of Concealing Them
This submission addresses the Sanity Challenge, Path One: Ship an Agent That Queries Real Content. The project is a multi-agent research analyst named Research Dossier, constructed with LangGraph. Rather than posing a research question to a single model, the system routes the query through four stages: Research, Analysis, Writing, and Review.
The research stage is fueled by a Sanity Knowledge Base accessed via Sanity Context MCP. The core issue explored is not just locating information, but managing scenarios where sources present conflicting claims. When such disagreements arise, Research Dossier does not blend the information into a single confident answer. Instead, it retains the disagreement, displays the sources of the claims, and can flag the final report as CONTESTED.
The Knowledge Base comprises three primary document types: topic, source, and claim. The crucial relationship is claim └── contradicts → claim, which represents disagreement as structured information rather than noise. This design allows the agent to reason about disagreement rather than merely presenting a list of matching passages.
Sanity is chosen for the project because it enables the storage of claims, sources, and relationships between claims. This structure alters the agent's capabilities, transforming the system from merely retrieving text that matches a query to retrieving structured knowledge that can be compared, traced back to its sources, and integrated into a multi-agent reasoning and review pipeline.
The workflow includes a research agent responsible for retrieval, an analysis agent that compares retrieved claims, a writing agent that converts the analysis into a source-linked report, and a review agent that serves as a hallucination gate. The final report is either APPROVED or REVISED based on the original research findings.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
