This repository documents an evolving research-program dedicated to establishing structural logical rigor, rather than presenting a finished software product.
Core mission: Uncover hidden assumptions before they support hallucinations
Central entry point to the Ekkehard Finkeissen Research Architecture
- Coherent and constraint-based epistemic framework
- Infrastructure for auditable knowledge and decision systems
- The repositories together form one pipeline and must be interpreted together
- Core Principle: Any statement not explicitly bound to a concrete problem context is treated as epistemically inadmissible → processing stops
- Actively exploring pathways to stabilize AI systems (e.g. LLMs) against hallucinations and unbound generalizations via epistemic constraints Modern AI systems produce claims without explicit epistemic binding. This pipeline enforces context-bound admissibility as a hard constraint.
This pipeline enforces epistemic governance through:
- Explicit binding of every statement to a concrete problem context
- Hard STOP conditions on unbound generalizations
- Traceable knowledge provenance and audit trails
The following cyclical pipeline is a constraint architecture for epistemic governance in which admissibility precedes claims: The pipeline is cyclical: artifacts can regress to Legacy when admissibility fails. Examples: → see Matrix repository
What it is NOT:
- Not a prompt framework
- Not an absolute or unquestionable knowledge-base
- Not an LLM wrapper
Pipeline Status (July 2026): The architecture is actively maintained and evolves sequentially. It is intentionally still in a research and refinement phase.
Current Maturity
- research-program — Mature epistemic core and admissibility rules (conceptually stable)
- mms — Normative requirements and prompt-based reference implementation are stable; full native Python implementation is under active development but not yet complete
- matrix — First curated knowledge instantiations (in active development)
- hypotheses — Early stage with initial falsifiable claims
- predictions — Early stage with verifiable projection artifacts
- legacy — Stable non-operational archive
- learn / invest / life — Application layers (varying maturity)
Overall: Actively maintained · Sequentially evolving · Deliberately minimal. Not yet a production-ready framework. Contributions welcome under the architectural constraints of the research-program.
The Epistemic Problem: Human Defensiveness & AI Hallucination: Human researchers, driven by conscious or unvoiced defense mechanisms (such as rationalization or selective bias), routinely omit the exact boundaries of their claims to protect their hypotheses. Modern AI systems ingest these un-bound generalizations and scale them into hallucinations. This pipeline acts as an uncompromising logical counter-measure.
Research areas:
- epistemic governance
- epistemic admissibility
- constraint-based reasoning
- knowledge provenance
- AI safety
- hallucination mitigation
- scientific workflows for AI
- verification of claims
- formal reasoning for LLM systems
- auditable decision systems
- uncover hidden assumptions
Garbage collection: Active sink for inadmissible or unresolved artifacts. https://github.com/finkeissen/legacy
⬇️ interface ⬇️
Foundation: epistemic admissibility rules, STOP conditions, and core grammar. https://github.com/finkeissen/research-program
⬇️ interface ⬇️
Constraint execution and governance layer: Enforces epistemic admissibility rules, STOP conditions, and knowledge transitions across the pipeline. The normative specification and prompt-based reference implementation are stable and usable. A complete native Python DBMS-style implementation is under active development but not yet finished.
⬇️ interface ⬇️ if STOP ➡️ put in Legacy
Knowledge: matrix registers admissible knowledge artifacts with their originating context. https://github.com/finkeissen/matrix
⬇️ interface ⬇️ if STOP ➡️ put in Legacy
New ideas: versioned, falsifiable claims derived from the epistemic core. https://github.com/finkeissen/hypotheses
⬇️ interface ⬇️ if STOP ➡️ put in Legacy
Planning: operationally verifiable projections derived from hypotheses and matrix. https://github.com/finkeissen/predictions
⬇️ interface ⬇️ if STOP ➡️ put in Legacy
Training: practical training and mastery of knowledige in general and the pipeline in specific. Turning theory into skill through deliberate practice, case studies, and building epistemic discipline. https://github.com/finkeissen/learn
Action: generalized investment grammar. Deliberate allocation of scarce resources (time, energy, attention, trust, health, coordination, money,…) toward future gain under uncertainty – beyond finance. Explicit multi-level evaluation: person ↔ group, trade-offs, collateral damage, distributional effects. https://github.com/finkeissen/invest
Intuition: Framework and art projects about life — grounded in the research-program, yet not formally bound. Contributions are welcome - if they operate under the architectural constraints of this repository. https://github.com/finkeissen/life
Status: actively maintained · sequentially evolving · deliberately minimal
→ External references: https://x.com/EFinkeissen
→ Books: https://www.amazon.de/s?k=finkeissen+ekkehard
Current Research Stage: The architectural framework is conceptually defined across all steps (0–5). Current active work focuses on the formal grammar of the research-program constraints and the execution logic of the mms pipeline interface. Validation is iterative: we test the pipeline by feeding it historically flawed scientific papers to verify if the hard STOP conditions reliably trigger where human peer-review historically failed.