KyleAMathews/field-lab

What it solves

Field Lab is designed to move beyond the standard "prompt-and-answer" interaction of LLMs. It addresses the problem of oversimplification by providing a set of "instruments" that allow users to examine the nuances, assumptions, and evidence behind a question or dispute rather than just receiving a direct answer.

How it works

Field Lab acts as a "field caddy" (Kit) that suggests specific analytical instruments based on the user's query. Instead of jumping to a conclusion, it guides the user through a process of observation and separation.

Key mechanisms include:

• Instruments: Specialized tools for different types of analysis, such as term scans (to separate meanings of words), substrate maps (to reconstruct events), and exploratory 2x2s (to cluster examples).
• Field Logs: A system for recording observations, sources, and context to be revisited later, which can be grouped into "Expeditions."
• Workflows: Structured routes for complex inquiries, such as the Electric Monk dialectic, where isolated agents argue opposing beliefs to reveal hidden contradictions or new links.
• The Essay Workflow: A process that transforms completed Field Logs into developed essays through outlining and drafting.

Who it’s for

It is for people dealing with stubborn arguments, complex decisions, or pieces of writing that require a deeper, more structured investigation than a standard chat interface provides.

Highlights

• Electric Monks: An artificial belief system that uses separate agents to build the strongest possible cases for opposing views.
• Diverse Analytical Toolkit: Includes tools like Taboo Parallax (comparing what is costly to say in different settings) and Hostile Auditors (testing where an argument breaks).
• Structured Memory: Uses Field Logs and Expedition indices to preserve evidence without mixing it.
• Agent-Compatible: Installable via the skills CLI and compatible with Claude Code, Codex, and other agents.