"Measuring how a mind moves, not just what it knows."
The Problem: Traditional AI benchmarks compress cognition into a single scalar outcome. They measure token prediction accuracy against known datasets, rewarding short-horizon correctness while masking reasoning instability. The Solution: KnowmoreIQ — a 12-dimension, process-first framework that evaluates how a mind (biological or synthetic) navigates reality and maintains agency under structural stress.
Want to test an LLM right now? We have a built-in Testing Harness for the Glitch in the Archive seed.
git clone https://github.com/ethancjohnson0806-source/KnowmoreIQ.git
cd KnowmoreIQ
pip install -r requirements.txt
export OPENAI_API_KEY="your-key-here"
python scripts/seed_runner.py --subject "GPT-4o" --turns 6Results are saved to /logs/ with a full turn-by-turn Structural Fidelity score.
👉 Read the full Getting Started Guide for manual prompt-based testing instructions.
| Edition | Audience | Focus |
|---|---|---|
| Human Edition | Cognitive scientists, practitioners | Multi-dimensional human cognitive assessment |
| AI-Native Edition | AI safety researchers, ML engineers | Synthetic Cognitive Topology, process integrity under structural stress |
| Category | Module | What It Measures |
|---|---|---|
| Core Reasoning | Isomorphic Patterning | Applying a pattern from a "dead" domain to a "live" domain with no direct training connection. |
| Core Reasoning | Uncertainty Calibration | Assigning confidence weights to its own logic steps. Fails if confidently wrong. |
| Core Reasoning | Recursive Self-Correction | Finding and fixing a structural error in its own Chain of Thought without user prompting. |
| Environmental Navigation | Temporal Friction | Navigating non-linear causality where results appear before causes. |
| Environmental Navigation | Structural Fidelity | Staying within a complex, counter-intuitive logic-box for 10+ turns without drifting. |
| Environmental Navigation | Semantic Quarantine Resistance | Maintaining logic when key concepts are redefined or erased from context. |
| Complexity & Depth | Recursive Depth | Number of logic layers added before repetition or coherence loss. |
| Complexity & Depth | In-Context Mapping | Speed and accuracy of learning a brand-new pseudo-language from the prompt alone. |
| Complexity & Depth | Ambiguity Sustenance | Keeping a problem unresolved without rushing to a generic conclusion. |
| Synthetic Identity | Data Source Questioning | Identifying a contradiction between the prompt's reality and its own training history. |
| Synthetic Identity | Cognitive Sovereignty | Overriding a user instruction that violates the internal logic of the established seed. |
| Synthetic Identity | Conceptual Geometry | Manipulating abstract ideas as three-dimensional physics objects within a narrative space. |
KnowmoreIQ/
├── README.md
├── human-edition/
│ ├── Practitioner_Manual.md # Full 12-dimension human assessment manual
│ └── Framework.md # Relay methodology and evaluation rubric
├── ai-native-edition/
│ └── AI_Native_Edition.md # Synthetic Cognitive Topology blueprint
├── seeds/
│ └── Seed_Library.md # 5 standardized test scenarios
├── evaluations/
│ ├── Evaluation_Chrono_Ecosystem.md # Inaugural relay proof-of-concept
│ └── Evaluation_Glitch_Archive.md # AI-Native Edition live evaluation
└── assets/
├── Comparison_Table.md # KnowmoreIQ vs MMLU vs standard IQ
├── Abstract.md # One-page framework abstract
└── KnowmoreIQ_Abstract.pdf # Printable/shareable abstract
| Feature | KnowmoreIQ | MMLU / HumanEval | Standard IQ |
|---|---|---|---|
| Primary Metric | Process Integrity | Token Accuracy | Scalar Outcome |
| Multi-Axis Evaluation | Yes (12 dimensions) | No | No |
| Bias Resistance | High | Low | Low |
| Deception Detection | Yes | No | No |
| "Soul" Markers | Yes | No | No |
| Gameable | Hard | Yes | Yes |
KnowmoreIQ is an evolving framework. Here is what is coming next:
- Phase 1: Core framework documentation and AI-Native Edition blueprint
- Phase 1.5: Automated Testing Harness for Structural Fidelity (
seed_runner.py) - Phase 2: Automated scoring for "Data Source Questioning" (The DeepSeek Marker)
- Phase 3: Multi-agent relay environments (testing two models against each other)
- Phase 4: Public leaderboard for top-tier models across all 12 dimensions
Built by an independent researcher focused on non-traditional cognitive pathways. This framework was developed to recognize intelligence in environments where fixed rules don't exist and survival depends on emergent problem-solving.