Here is the work we would take part in and cannot do alone. Most of it comes out of the first theme we have opened, The Law We Quote Is Not the Law Ashby Wrote, and later themes will add to it. Every item is small enough to start, and every one of them could change what we say about the Viable System Model.
Several of the items are questions about what a result would even mean. A few need data, texts or time we do not have. If you are a researcher in one of these areas, a student looking for a topic, or someone who has tried this at work and kept records, one of them may be yours. Tell us which, and we will say what we can offer: sources, access, a co-author, or nothing more than thanks. The receipts in the working paper, A Stress Test for the Viable System Model: A Cumulative Research Programme, set out the same ground in more detail.
Who counts the states
Variety is a count of distinguishable states, and nothing in Ashby’s formulation fixes who does the distinguishing. The observing distinction is made before the arithmetic starts, and it comes back at two other places in the model: the decision about the system boundary and the specification of the transducers. We treat it as a political choice rather than a technical gap, which is a strong claim and an unflattering one. If you can show that the partition is fixed by something we have missed, that is a better result for us than an agreement.
The good regulator theorem
Conant and Ashby’s 1970 paper is cited in our field as proved mathematics. Its authors withdraw the universal quantifier inside the paper, the proof has no inputs, channels or noise, and the regulator is scored by outcome entropy alone, so a regulator that pins its system to a fatal outcome scores perfectly. Nathaniel Virgo and colleagues published a reconstruction in 2025 that replaces being a model with having one. Whether that reconstruction can carry the System Four claim we have hung on this theorem is, as far as we know, untested. A worked analysis of one diagnostic practice, or of one published case, would be enough to start.
Aulin’s derivation
Arvid Aulin published a law of requisite hierarchy in Kybernetes in 1979 that appears to supply the derivation Beer’s recursion argument never gave. OpenAlex recorded roughly 43 citations for it in August 2026, against about 1,040 for Conant and Ashby, which suggests the tradition did not take up an argument that was available to it. We have not yet read it in full, and “appears to” is doing real work in that sentence. Someone who reads it properly, and tells us whether it establishes recursion or only something adjacent to it, would close a gap in Chapter 1 in a week.
Measuring variety at all
Two quantitative tests exist. Schwaninger and Scheef (2016) declared eight hypotheses before seeing the data, which is the hardest thing anyone in the field has done. Michael D. Pfiffner’s 2017 dissertation operationalized the model into a questionnaire and split 135 organizations in German-speaking Switzerland, in advance, into those that had survived an existential crisis and those that had not. Both are better than the field’s norm, and in our reading both constructions leave the model almost unexposed. A replication with the revision rule fixed before the result is known would be the first of its kind. Two earlier dissertations, Frost in 2005 and Crisan Tran in 2006, we know only through a 2016 review that names them without detail, and anyone who has tried to count variety in a real organization and kept the data has material we do not.
Recursion levels
Two questions, one cheap and one not. Do trained analysts agree on where the levels fall when they work on the same organization, and can the stratification be measured without an analyst in the loop, for instance by timescales in the data. The second is one of the four demonstrations in the working paper, and it is the one where a small team could produce a first result without us.
The model as a network
Formal specification of the VSM as a network, ablation of its components, and measures of emergence. This is the fourth demonstration in the working paper. It needs someone who is at home in network science and willing to write down what the model actually claims as a graph, including the parts where the claim turns out to be vague. We would rather find out that the specification is impossible than keep assuming it is easy.
The neighbouring literatures
Parts Two and Three of Punch and Judy claim that sociotechnical design, organization theory, complexity science and Ostrom’s work on the commons reached several of the model’s components independently, and often measured them. If we have those fields wrong, the correction is worth more to us than agreement. Chapter-length surveys, or a precise objection to ours, are welcome.
Where we are probably naive
The part we cannot delegate to a literature check. Point out where we are ignorant, not where we are wrong on a page number: where we do not know that we do not know. A first public draft is the cheapest place to be embarrassed, and the chapters are published one at a time precisely so this can happen early. If a whole literature exists that we have not mentioned, that is not a small thing to tell us. It is the most useful thing you can send.
How to take part
Pick an item and write to us. The Zotero library is open and holds most of what we cite, and we will send anything you cannot get. Chapters discuss on Slack, linked from each chapter page. The WhatsApp group is public, the Buzz community is for people who want to work on this, and new chapters go out through the newsletter. Or reach us directly: Margeret, Krishan.
Krishan Mathis and Margeret Heath