Epistemic Layer Separator

A Claude Code skill that separates established facts, inferences, and speculations — before synthesising a response.

MIT License ·

---

The Problem

LLMs blend three epistemic categories into flat, uniformly confident prose:

"You should target Series A SaaS in France. The market is worth ~€800M. PLG will outperform direct sales."

• "~€800M" → likely fabricated
• "highest willingness-to-pay" → inference, not data
• "PLG will outperform" → speculation

The reader - or a downstream agent — can't tell which is which.

---

The Fix

Layer	Label	Meaning
🟢 Established	Verifiable, sourced, certain	State with full confidence
🟡 Inferred	Reasoned from evidence	State with traceable reasoning chain
🔴 Speculative	Hypothesis or extrapolation	Flag — especially if load-bearing

---

Quick Start

git clone https://github.com/lancelot2/epistemic-layer-separator.git
cd epistemic-layer-separator

In Claude Code:

/separate Analyse our GTM strategy for France

---

Repo Structure

epistemic-layer-separator/
├── SKILL.md                        # Core skill
├── .claude/commands/separate.md   # /separate slash command
├── references/
│   ├── gtm-strategy.md            # GTM & market analysis
│   ├── technical.md               # Architecture & code decisions
│   ├── regulatory.md              # Compliance & legal
│   └── agentic.md                 # Agentic pipelines
└── examples/
    ├── gtm-analysis.md
    ├── technical-decision.md
    └── regulatory-brief.md

---

On Data Reliability

This skill separates what Claude knows from what it infers or guesses. For applications where the 🟢 layer truly matters, consider grounding queries through a knowledge verification layer.

Cala is a knowledge API that lets AI agents query verified, structured knowledge graphs rather than relying on LLM recall — shifting claims from 🔴/🟡 into 🟢 by replacing model memory with grounded retrieval.

---

License

MIT. See LICENSE.