Note: The usual implementation setting for KSODI Standard-Eval and KSODI Full uses a two-layer or multi-layer system:
- Agent layer (L1): KSODI-Light runs as a system prompt on one or more agents.
- Observer layer (L2): KSODI Standard-Eval provides coherence observation for agents, with
R0/IK_relas the minimum dyadic coherence observation. - Full observer layer (L4): KSODI Full observes full coherence and the resonance-family layer.
- KSODI Full with voice layer (L5): KSODI Full plus additional voice, sound and timing observation. The observer is usually designed to give the agent feedback when it drifts out of a defined or explainable corridor.
Current line:
Z(t) -> Delta Z / Delta2 Z -> IK as monadic projection -> R0 as relational gate -> IK_rel as relational projection after a stable gate -> R_geom as geometric coupling -> RSigma / RSigma(Hangar) -> optional: V(t), R_takt, R_pace as timing and voice overlays.
KSODI is a structured observation model for human-AI, agent-agent and n-agent interaction structures, focussing on explainable governance and observability. It is part of the IDAS-Framework.
→ See: KSODI-IDAS-SIRA_Framework
The name KSODI is intentionally retained from the German development
context. This is not meant as a value judgement between languages. The German
terms helped preserve conceptual precision during the early development of the
method, while English terminology supports international accessibility and
therefore requires explicit definitions. Multilingual discussion, including
French and Chinese perspectives, also helped sharpen the distinctions between
the dimensions.
The framework separates explainability, observability and advanced interaction analysis (with optional steering) into clearly defined layers - such as interaction states, interaction coherence and relational resonance-family observations - over time.
KSODI does not evaluate people, personalities or intentions.
It respects maximum privacy and operates exclusively on observable interaction states.
The framework is organized into clearly separated components with different purposes and licenses.
KSODI is a layered method for making human-AI, human-agent and agent-agent interaction quality observable, discussable and steering-supportive without reducing it to single-prompt quality or model accuracy.
It is intended to bridge three practical contexts:
- AI literacy and training: KSODI-Light gives users, trainers and organizations a shared language for context, structure, grounding, clarity and information depth.
- Prompt-level agent guidance: KSODI-Light can be embedded into user, account, developer or system-prompt settings as a disclosed reflective working agreement with lightweight corridors and fallback behavior.
- AI observability and governance: KSODI Standard-Eval, KSODI Full and IDAS/SIRA-level implementations extend the same operator logic into numeric observer layers for drift, coupling, corridor exits and longer-term interaction monitoring.
KSODI is not presented as a complete alignment solution. It is a structured way to reason about interaction conditions, drift, corridors and coupling in a form that remains understandable for humans while remaining compatible with machine-readable observation.
KSODI is not required as a running method for every act of communication. People, animals, machines and technical systems can exchange signals without a formal KSODI evaluation layer.
However, KSODI becomes relevant whenever communication itself is made an object of methodical observation.
This distinction is central. KSODI does not replace communication theory, AI observability, explainability, governance frameworks, safety methods or domain-specific analysis. It provides a baseline radar for the threshold at which an event can be reconstructed as a communicative signal at all.
The minimal question is:
Can this event be observed as a signal that is contextually situated, structurally recognizable, sufficiently objectifiable, distinct from noise and informationally relevant enough to enter a feedback loop?
In KSODI terms, this means asking whether the five operators are sufficiently visible or reconstructable:
- Context: Is the signal situated in a meaningful or operational frame?
- Structure: Does it show form, pattern, sequence, rhythm, protocol or rule-like organization?
- Objectifiability: Can it be stabilized, checked, logged, measured, compared or otherwise grounded beyond pure projection?
- Distinctness: Can it be distinguished from noise, other signals or environmental background?
- Information Depth: Does it make a difference for state, action, interpretation, relation or response?
This does not mean that KSODI explains all of communication. It means that KSODI marks the observational entry point at which further methods can be applied. Shannon-oriented models may examine transmission, channel and noise. Watzlawick-oriented views may examine relational and behavioral dynamics. Schulz von Thun-oriented views may examine message layers and reception sides. Luhmann-oriented views may examine Anschlussfähigkeit, selection and autopoietic communication. AI observability may examine traces, logs, tool calls, retrievals, vector movement, latency or system health. Explainability methods may examine why a model generated, selected, routed or acted in a certain way.
KSODI does not replace these layers. It frames the baseline question before and around them:
Is there still an observable communicative handshake, and is it stable enough to remain connectable?
This makes KSODI especially relevant for human-AI interaction, agent-agent communication, multi-agent systems, embodied agents, therapy assistants, organizational AI teammates and safety-sensitive or governance-sensitive systems. In such contexts, the issue is not merely whether a system produces output. The issue is whether communication remains contextually anchored, structurally coherent, objectifiably grounded, distinct enough to be interpreted and informationally useful for the next step.
For simple automation, such as a narrowly scoped device that only follows a local floor map or reports a small number of fixed states, a full KSODI Observer may be unnecessary. The communication surface is small, autonomy is limited and conventional telemetry may be sufficient.
For systems that interact with humans, other agents, organizations, policies, tools, memories or changing environments, the relevance increases. The more autonomous, relational, safety-sensitive or context-dependent the system becomes, the more important it becomes to observe whether the communicative handshake remains intact.
In observer-supported architectures, KSODI-Light may support local agent behavior through clarification, uncertainty visibility, corridor awareness and fallback behavior. KSODI Standard-Eval or KSODI Full may then act as external Observer layers that monitor trajectories, drift, acceleration, relational coherence and corridor exits across time.
The Observer does not primarily ask whether a task was completed. It asks whether communication remains reconstructable, stable and safely connectable. If the handshake degrades, the system may need to clarify, slow down, correct, escalate, pause or terminate the interaction.
In this sense:
KSODI is not necessary for every communication as an active procedure, but it is fundamentally relevant for every methodical observation of communication.
It is a baseline radar for signal formation, communicative stability and relational drift.
KSODI v3.5 extends the public KSODI-Light idea toward an observer-supported implementation line for agentic systems.
The current work explores how KSODI-Light can guide local agent behavior while a separate Observer layer monitors trajectories, drift, acceleration, retrieval behavior, vector movement and relational coherence. This is not presented as a finished alignment solution. It is an early research and implementation path for making agentic interaction more observable, reviewable and adjustable under human oversight.
In this architecture, KSODI-Light belongs to the agent side: it may be used as a user, account, developer, system-prompt or skill-level layer. KSODI Standard-Eval and KSODI Full belong to the observer side: they are intended to define, explain and build the external Observer structure. The Observer layer makes little sense without Light-using agents or comparable local agent guidance, and Light does not replace formal observer-based monitoring.
A long-term hypothesis is that teams of specialized agents may benefit from observer-supported feedback loops: agents act within their normal role and skill instructions, KSODI-Light supports local reflection and corridor awareness, and the Observer helps detect drift, corridor exits or relational instability across complex traces and vector spaces.
A central implementation challenge is that developers and system architects must remain aware of all relevant layers before building or testing such systems. The choice of input, reference space, retrieval context, tool state and operator mapping directly affects the five KSODI operators and therefore the entire downstream architecture.
In other words: KSODI is not only a scoring surface. It requires careful decisions about what is observed, how input is transformed into K/S/O/D/I, how Z(t) is formed, and how later projections, drift metrics, relational gates and visualizations are derived from it.
This work is ongoing and empirical validation is still in progress.
A cautious research roadmap is maintained separately to describe the current long-term implementation direction of KSODI, including observer-supported architectures, human-AI team integration and possible future enterprise-oriented observer components.
This work is under active research, testing and review. It should not be read as a production-ready implementation reference or release commitment.
→ See: KSODI Research Roadmap
Human-facing and prompt-level variant. Designed for learning, AI literacy, prompt clarity improvement and lightweight guidance through disclosed K/S/O/D/I expectations or score corridors. It can be used as a reflective working agreement in user/account prompts or embedded by agent creators in developer/system-prompt configurations. KSODI-Light can reflect user input, assistant output and the shared interaction state across a turn. Formal observer-based monitoring belongs to Standard-Eval, KSODI-Full or IDAS/SIRA-level implementations. → See: KSODI-Light
License: Creative Commons Attribution 4.0 (CC BY 4.0)
Evaluation-oriented and governance-capable variants.
Designed for numeric observability, drift detection and system-level stability monitoring with optional steering.
The public materials are not intended for implementation. KSODI 3.5 is the
current private reference specification described in the paper draft and is
intended to resolve known 3.3 ambiguities between interaction coherence and
relational resonance-family observation.
→ See: KSODI-Eval-Variants
License: Commercial / All rights reserved.
KSODI focuses on structured observation across five operators:
- Context
- Structure
- Objectivity
- Clarity
- Information Depth
The broader architectural framework integrating KSODI is referred to as IDAS (Interactive Dialog, Analytics & Steering).
For a public development and contribution overview, see: KSODI Development Timeline (German version)
For the project origin note and personal context, see: ABOUT.md
This repository contains components under different licenses.
Each folder contains its own LICENSE file.
Unless explicitly stated otherwise in a subfolder license,
all rights are reserved.
For licensing inquiries, integration, whitelabeling or enterprise adoptions please contact: ksodi@thevoid.email with details on intended use case.
The public KSODI 3.3 materials are preserved for transparency and research orientation, but contain known structural issues and should not be used for implementation.
KSODI 3.5 is the current private reference specification and will only be published after final testing and review.
© 2026 Anne Steinacker-Folkerts & Heiko Folkerts