cns-tinker

Deploy Chiral Narrative Synthesis with a Tinker-based training and evaluation pipeline.

Updated Nov 8, 2025

Search Tags:#contradiction #narrative-synthesis #tinker-api #lora-fine-tuning #mult-agent-debate #fisher-information #sci-fact-fever

Ready to run

npx skills add https://github.com/North-Shore-AI/tinkerer --skill cns-tinker

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: cns-tinker
Source: https://github.com/North-Shore-AI/tinkerer/tree/main/.claude/skills/cns-tinker
Command: npx skills add https://github.com/North-Shore-AI/tinkerer --skill cns-tinker

SYSTEM DOCUMENTATION & REQUIREMENTS

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.

Frequently Asked Questions about cns-tinker

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I detect contradictions across multi-source narratives?▼

Contradiction detection across multi-source narratives is handled by fine-tuning models using SciFact and FEVER data within this Tinker-driven pipeline to identify conflicting claims and generate a reconciled narrative.

What is Chiral Narrative Synthesis and how does the Tinker API support it?▼

Chiral Narrative Synthesis generates unified narratives from conflicting sources using the Tinker API to orchestrate LoRA fine-tuning, multi-agent debates, and topology-informed synthesis for coherent reconciled outputs.

How do I use multi-agent debate orchestration for evidence scoring?▼

Multi-agent debate orchestration evaluates evidence through RL-based scoring using Fisher Information, enabling models to weigh conflicting sources and synthesize coherent narratives via the Tinker API.

Can I fine-tune a base model with LoRA for contradiction detection on Tinker?▼

Yes, LoRA fine-tuning requires a compatible base model and access to Tinker resources to train the contradiction detector using SciFact and FEVER datasets with defined evaluation metrics.

Do I need to prepare data with SciFact or FEVER formats before using this workflow?▼

Yes, the end-to-end CNS workflow requires data preparation using SciFact and FEVER datasets to train the contradiction detector and evaluate multi-source narrative synthesis accurately.

What are the limitations of using topology-informed synthesis for narrative reconciliation?▼

Topology-informed synthesis requires a compatible base model, LoRA fine-tuning, and Tinker API access with clearly defined prompts, limiting use without proper environment setup and evaluation metrics.