cns-tinker

What problem does it solve?

Apply CNS 3.0 within a Tinker-driven training and evaluation pipeline to detect contradictions across multi-source narratives and generate coherent, unified narratives.

Core Features & Use Cases

• End-to-end CNS workflow: contradiction detection using SciFact/FEVER data, multi-agent debate orchestration, and topology-informed synthesis.
• Reusable training and evaluation harness: fine-tuning LoRA-based models and scoring evidence with Fisher Information, enabling robust narrative synthesis.
• Use Case: a research team builds an automated CNS evaluation harness to compare competing articles and produce a reconciled narrative with invariants.

Quick Start

Run the CNS-tinker workflow to train the contradiction detector, execute multi-agent debates, and synthesize the unified narrative.