ARGUS

ARGUS is an evidence-first investigation interface for making persuasive reasoning inspectable.

It turns supplied text, public web pages, and scholarly identifiers into a source-bound graph of claims, reasons, evidence, omissions, dependencies, incentives, and amplification paths. Users can begin with a concise review, follow the questions that matter to them, move through the graph node by node, and return to earlier steps without losing the evidence trail.

ARGUS separates what a source says from what the available evidence supports, contradicts, leaves unresolved, or cannot yet establish. It checks whether cited sources are identifiable, whether they support the exact claim, whether apparently independent sources share dependencies, whether a result transfers to the new context, and whether another reviewer could reproduce the calculation. Contradictory evidence remains visible instead of being collapsed into a single verdict.

Codex CLI with GPT-5.6 performs the specialized close reading. Deterministic code validates exact quotations, source locations, hashes, graph bounds, and export integrity. The interface explains the analysis in plain English so a student can understand both the result and how it was reached.

The repository is private for judging. The source code is available to authorized OpenAI Build Week reviewers at the linked GitHub repository.

Built With

• codex-cli
• crossref
• cytoscape.js
• datacite
• fastify
• gpt-5.6
• playwright
• react
• ruby-on-rails
• sqlite
• typescript
• vite
• vitest
• zod