Downloads · 30 days
0
eem-expert
ftl-ai/eem-expert
eem-expert is a machine learning model from ftl-ai. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as mit.
Expert knowledge base for explaining External Epistemic Memory (EEM) to humans and LLM-based agents. Contains 90 justified beliefs covering what EEM is, how it works, why it matters, and empirical evidence for its eff…
Access
Public
Updated May 31, 2026
Repo size
—
Likes
0
Public
Repo makeup
Click a slice to open those files.
.json65.1 KB · 89%
At a glance
- License
- mit
- Access
- Public
- Created
- May 31, 2026
- Updated
- May 31, 2026
- SHA
- bd0662a0
Model card
From the Hugging Face model README
EEM Expert
Expert knowledge base for explaining External Epistemic Memory (EEM) to humans and LLM-based agents. Contains 90 justified beliefs covering what EEM is, how it works, why it matters, and empirical evidence for its effectiveness.
What is this?
This is an External Epistemic Memory (EEM) — a model-agnostic knowledge base that any LLM can use via the reasons CLI or tool calling. Unlike a LoRA or fine-tune, this knowledge is not baked into model weights. It is external, inspectable, correctable, and works with any model.
Stats
| Metric | Value |
|---|---|
| Total beliefs | 90 |
| Status | 90 IN / 0 OUT |
| Premises (observations) | 49 |
| Derived (justified conclusions) | 41 |
| Nogoods (contradictions) | 0 |
| Retraction rate | 0% |
| Max derivation depth | 5 |
Domain Coverage
- What EEM is: three load-bearing properties (external, epistemic, memory), formal definition
- How EEM differs from alternatives: vs RAG, vs context/conversation history, vs parametric knowledge, vs knowledge graphs
- TMS architecture: Doyle 1979 foundations, SL justifications, retraction cascades, nogoods, backtracking
- Empirical evidence: ablation studies, dual-path validation, confidence unreliability, model compensation, six-domain validation
- Design principles: derive-then-review, cognitive budget, wide-not-deep, generate-and-critique
- Practical workflows: expert pipeline, how agents use EEM, how humans use EEM, multi-agent belief tracking
- Construction & cost: amortization argument, automated overnight construction, construction cost measurements
- Staleness & maintenance: staleness detection, source change tracking, stale belief workflows
- Getting started: installation, CLI interface, quick start, HTTP endpoint access
How to Use
Import into a reasons database
reasons init
reasons import-json network.json
Query beliefs
reasons search "what is EEM"
reasons explain eem-definition
reasons show eem-three-properties
Use as an MCP tool or CLI
Any LLM agent that can call reasons search, reasons show, and reasons explain can use this knowledge base. The agent does not need to be told it is an expert — the knowledge base speaks for itself (see belief expert-prompt-paradox).
Key Beliefs
| Node | Summary |
|---|---|
eem-definition | EEM is knowledge that lives outside the model, carries its justifications, and lets you understand how the system knows what it knows |
eem-three-properties | External, epistemic, memory — three load-bearing properties |
eem-works | EEM measurably and dramatically improves LLM performance on domain tasks |
evidence-dual-path | Opus + dual-path achieves 98.5% A/B across 3,853 questions |
evidence-retraction-rate | 13-37% of derived beliefs retracted per review round — self-correction works |
confidence-unreliable | LLM self-assessed confidence does not track accuracy (r=-0.182 to r=0.219) |
ftl-reasons-is-tms | ftl-reasons implements Doyle-style TMS with LLMs as problem solvers |
Sources
Built from exploration of benthomasson/ftl-reasons and empirical studies across 40+ expert knowledge bases ranging from 237 to 13,511 beliefs.
Files
| File | Description |
|---|---|
network.json | Full belief network (machine-readable, portable) |
reasons.db | SQLite database (gitignored, regenerate with reasons import-json network.json) |
CLAUDE.md | Agent instructions for using this knowledge base |
Quality
- All 90 beliefs are IN (none retracted)
- 49 premises grounded in direct observations and published research
- 41 derived beliefs justified from premises via SL justifications
- 0 nogoods — no contradictions detected
- Built and reviewed using ftl-reasons derive and review-beliefs pipeline
Limitations
- Focused on EEM concepts and ftl-reasons implementation — does not cover alternative TMS implementations in detail
- Empirical evidence drawn primarily from code-expert use cases
- No ATMS or assumption-based beliefs (single-context TMS only)
- PostgreSQL multi-tenant patterns not covered
Authors
- Ben Thomasson (@benthomasson)
License
mit
- License
- mit
- Created
- May 31, 2026
- Updated
- May 31, 2026