V9 — Multi-Level Prompting Framework
V9 — Multi-Level Prompting Framework
A cognitive mnemonic framework for AI: multilevel prompting and multilevel context.
A Cognitive Framework That Lives in Memory
Within Any AI Model
Since 2022, Anthilla has been developing V9, a cognitive mnemonic prompt framework based on multilevel prompting and multilevel context. A framework that doesn’t tie you to a vendor, isn’t a subscription service, and works within any language model.
V9 is designed to address the fundamental limitations of AI models in extended sessions: cognitive drift, context loss, hallucinations, action bias, and inconsistency between declared and executed actions.
The framework is structured in cognitive layers that activate progressively, maintaining coherence across long sessions and preventing the most common cognitive biases that affect AI models under pressure.
Key principles of V9:
- Essentialism (80/20): Focus on essential elements, eliminate non-functional complexity.
- Structural entropy reduction: Reduce structural disorder through systematic organization.
- Meta-computation paradigm: Cognitive optimization with focus oscillation.
- Information vector system: Multidimensional knowledge representation.
- Trust-oriented validation: CERTAIN / PROBABLE / UNCERTAIN / UNKNOWN classification.
- PAC (Computational Activation Protocol): Optimal selection of execution level.
Action bias prevention (M-1..M-9): Nine method rules to prevent premature closure, scope creep, and confabulation under pressure. Each rule is operational and verifiable.
Honesty under pressure (D-1..D-9): Nine honesty rules to prevent defensive confabulation, blame-shifting, and memory reconstruction. The worst pattern of action bias is defensive confabulation under stress.
Pre-flight and post-flight checklists: Operational protocols that force verification before and after every production action. A1-A9 before, B1-B6 after, H1-H6 before every factual assertion.
Loop protection state machine: Round 1 → apply fix, verify. Round 2 → revised fix after pre-flight redone. Round 3 → stop autonomous fix, ask user. Round 4 → halt, require written root cause analysis.
V9 is not a product for sale. It’s the cognitive framework we use internally and in every AI agent interaction. It’s open, documented, and reusable.
