The thinking behind the instruments.

Understanding is not recall, and a healthy system is not the sum of its parts.

Two convictions run through everything on this page. The first is that a person understands something when they can explain it in their own terms and derive it again, not when they can repeat it. The second is that the health of a person, a team or an organization depends as much on the connective medium between its parts as on the parts themselves. Knowledge systems that store facts and ignore relationships fail on both counts.

Understanding is not transmitted. It is constructed in conversation and shown by teachback.

Pask's Conversation Theory treats learning as an exchange between two perspectives, human or machine, that agree on what a topic means. Knowledge is held as an entailment mesh: concepts linked by what each requires to be understood, rather than by the order in which they were taught. Because the mesh has no fixed starting point, learners can enter wherever makes sense to them.

A learner has understood a topic when they can teach it back in their own words and rebuild it from the concepts it depends on. That test does two jobs in CoExplorer. It is how the Workbench confirms that an expert's knowing has been captured correctly, and how the Wayfinder confirms that a newcomer has grasped it. The same principle, applied to AI agents, becomes the teachback gate described below.

Shapes: Workbench · Wayfinder · Augur (Conversation lens)

Only variety can absorb variety. A regulator must be able to respond in as many ways as the situation can go wrong.

Ross Ashby's Law of Requisite Variety sets the basic test for any response: does it have enough options to match the disturbances it faces? Mick Ashby extends this into ethics. His Ethical Regulator Theorem sets out what a regulator needs in order to be ethical as well as effective, and his Law of Inevitable Ethical Inadequacy holds that no regulator is ever fully adequate, so governance has to be a standing function rather than a finished document.

James Bryant's behavioral cybernetics asks what must be in place for people to act as an outcome requires, across ten dimensions: person, state, understanding, decision, action, way, time, position, resource and outcome. Stafford Beer's Viable System Model gives the recursive structure that lets the same questions be asked of a team, a department and a whole organization.

Shapes: Augur (Variety lens) · Transducer · the agent architecture

The observer is part of the system being observed. Knowing is something we do together.

Second-order cybernetics puts the observer back into the picture. Von Foerster asked what it means to observe a system that includes oneself. Glanville developed the idea into a practice of design as conversation, in which the designer and the thing designed change each other. For CoExplorer this means every record keeps its author, every interpretation says whose it is, and the people building the system use it on their own work first.

Shapes: Workbench · provenance in the graph

An issue is a sign. The useful question is what it points to.

Peirce's semiotics treats anything that stands for something else as a sign to be interpreted. His abduction is the kind of reasoning that produces new hypotheses: given a surprising fact, what would make it unsurprising? His pragmatism tests a meaning by the difference it makes to conduct. Together they give a group a way to move from "what is happening" to "what might explain it" to "what would we do differently if that were true."

Shapes: Augur (Meaning lens) · Transducer

Living systems act on models of what is coming, not only on what has already happened.

Rosen defined an anticipatory system as one that contains a predictive model of itself and its environment and changes its present behavior according to that model. His relational biology distinguishes a system's metabolism, the work it does, from its repair, the capacity to maintain and renew itself. A threat to repair is often more serious than a threat to output, and easier to miss.

The Transducer is built on this idea. It makes an organization's model of its future explicit, connects decisions to it, and keeps the model with the decision so it can be checked when the future arrives.

Shapes: Transducer · Augur (Anticipation lens)

Reality is process. What looks fixed is usually a slow event.

Whitehead's process philosophy treats the world as made of occasions of experience rather than enduring things. In The Function of Reason he argues that reason serves "the art of life" by joining practical skill with speculative vision, and that a method which excludes vision grows tired and dogmatic. We use this to read what is actually changing in a conversation, and to notice when a group has mistaken something in motion for something settled.

Shapes: Wellspring · Augur (Process lens)

Meaning is found upstream, before it hardens into finished knowledge.

Drawing on Goethe's way of science, Bortoft distinguished upstream attention, where meaning is still coming into being, from downstream thinking, which handles finished products. He warned against the counterfeit whole: an abstraction assembled by adding parts together and mistaken for the living whole. The Wellspring exists to protect upstream attention, and provenance in the graph keeps a trace of how each finished statement emerged.

Shapes: Wellspring

Dysfunction often begins in the connective medium, not in the components.

Ground Regulation comes from the biology of the extracellular matrix, the medium through which every cell in the body is supplied and regulated. Pischinger and Heine showed that disease often starts there before it appears in any organ. We apply the idea to people and organizations: we assess nine dimensions of the ground and four stages of degradation, and we look for the error of treating components as primary while ignoring what connects them. The same reading applies at every scale, which is why it can inform a clinical encounter and a board discussion alike.

Shapes: Augur (post-session reading) · Wellspring

People handle complexity at different scales, and that capacity grows over a working life.

Jaques mapped the levels of complexity that work involves and the rate at which people's capacity to handle it develops. Stamp's appreciative approach recognizes capability through conversation rather than scoring, and gives people language for where they are now and where they are heading. We use both to pace learning in the Wayfinder and to match the Workbench conversation to the person taking part.

Shapes: Workbench · Wayfinder

Before a group can think together, someone has to understand who is involved and how they already work.

Three readings help us understand a situation before a Workshop begins: a stakeholder analysis of who is involved, what they bring and what they need; Etienne Wenger's communities of practice, which looks at shared domain, patterns of participation and shared repertoire; and an action framework that records current practice, commitments, habits and gaps.

We then choose a dialogue method to fit the situation:

• Syntegration (Stafford Beer) for a large group that needs to cover a complex agenda without a hierarchy.
• Structured Democratic Dialogue (John Warfield and Aleco Christakis) for building a shared map of how factors influence one another.
• Open Space (Harrison Owen) for self-organizing around the questions people care about most.
• World Café (Juanita Brown and David Isaacs) for cross-pollinating ideas through rounds of small conversations.

Any of these can run in a room in a day, or in Confabula over weeks.

Shapes: Workshop · Confabula

Augur puts five of these traditions to work in parallel on one convergence item, so a group sees five kinds of question side by side.

Confabula holds the conversation: conferences, items and threaded responses, each with its author and history. When a group opens a convergence item, Augur's companions read it. The Conversation lens (Pask) keeps the protocol and tracks what has been taught back. The Meaning (Peirce), Variety (Ashby) and Anticipation (Rosen) lenses pose questions in turn. After the session, the Process lens (Whitehead) and a Ground Regulation reading look at what changed. Stakeholder, community and action readings then propose people, actions and decisions for the graph.

The companions ask; they do not vote. Nothing they propose enters the graph until the group has confirmed it.

Shapes: Workshop · Transducer

Trust in a system that uses AI is not asserted. It is shown, through records of how every conclusion was reached.

CoExplorer records the source conversation, the model that proposed each link and the person who approved it. That record is what lets an organization show conformance with the frameworks that now govern AI:

• International: ISO/IEC 42001:2023 for AI management systems, with ISO/IEC 23894 on AI risk management and ISO/IEC 42006 on certification bodies.
• United States: the NIST AI Risk Management Framework (2023) and its Generative AI Profile (2024).
• United Kingdom: certification through BSI under UKAS accreditation, and the Alan Turing Institute's Trustworthy and Ethical Assurance method.
• European Union: the AI Act (Regulation (EU) 2024/1689), whose obligations have been phasing in since February 2025.

Sovereignty

Where material is sensitive or regulated, the system runs on open-weight models in your own country, on your own air-gapped system, or both. Nothing needs to reach the internet. The capability registry enforces these boundaries: an agent asks for a capability, and the registry sends it only to a provider allowed to handle that material.

The teachback gate

Before an agent acts or writes to the graph, it must state back what it has understood it is being asked to do, and have that confirmed. Pask's test for a learner becomes a safeguard for AI, including against instructions slipped in from outside. This is in development.

The agents are organized like a team with regulating loops, not like a pipeline.

One layer of agents asks what happened: transcription, context, ontology and provenance. A second asks what it means, applying the lenses on this page. A third asks how the whole system is changing: learning, research, reflection and governance. A supervisor watches all three. Each layer observes and improves the one below it, as Ashby and Beer would expect of a viable system.

Every agent proposes, states back its understanding and hands on. A person remains the point at which a proposal becomes a fact. The frameworks are distributed across the agents rather than stacked in a single prompt, so each can be examined, tested and replaced on its own.