---
title: "Ch29 two cheapest experiments content"
description: "Chapter 5, The Ladder Without a Ruler, found that the model has no procedure for deciding where a recursion level lies. Chapter 9, Why None of This Ever Cost…"
url: "https://wiki.cybernetic-futures.org/punch-and-judy/part-05-put-model-at-risk/ch29-two-cheapest-experiments/ch29-two-cheapest-experiments-content"
modified: "2026-11-12T15:26:25.960Z"
language: "en"
site: "Cybernetic Futures"
---

# The Two Cheapest Decisive Experiments

## Two experiments the accounting produced

Chapter 5, *The Ladder Without a Ruler*, found that the model has no procedure for deciding where a recursion level lies. Chapter 9, *Why None of This Ever Cost Us Anything*, and Chapter 14, *The Evidence: Ostrom*, found that it has no evidence base. Each finding produced an experiment, and these two come first in the program because they are the cheapest decisive work available to us and the least excusably undone, not because they matter most. They came out of the book's own accounting rather than a wish list. Chapter 28 states the conditions under which a result from either would count against the model, and we do not restate them here. Chapter 30 sets out what the whole program costs.

We describe these two in more detail than the rest, because each has to be legible enough that a reader who thinks the design is wrong can say where. For each there is a question, a design, a decision rule fixed before the data is seen, a price where the work we have done supports one, and the negative result the experiment would establish. Where a design choice is contestable we say so and name the alternative. A proposal that hides its choices cannot be criticized, and so cannot be improved.

## Experiment one: measure the recursion level

*The criterion Chapter 5 showed the model has never had.*

**The claim under test.** Beer's test for whether a candidate is a recursion level is, in our paraphrase, that the contained system must be viable in its own right and one of the units that do what the containing system does.[^ch29-beer-paraphrase] It is a good test, and it measures nothing. Beer claimed more for it: the principle of recursion "is by no means loosely expressed: it is rigorously determined by the Law of Cohesion."[^ch29-beer-rigor] What he supplied to apply it are rules of thumb, such as the warning that more than seven or eight operational units means "you may be missing a whole level of recursion."[^ch29-beer-heur] Two competent analysts who disagree about a boundary have no procedure for settling the disagreement beyond those heuristics. The same hole appears one level down, in variety itself. Variety is a count of distinguishable states, and Beer says plainly who does the distinguishing: the observer, relative to the purpose attributed to the system. "Everything will depend on how the system is defined," and two people "will not necessarily agree" on its boundaries.[^ch29-beer-observer] His only arbiter between observers is predictive usefulness, and the model has no procedure for agreement. The experiment asks whether the levels a trained diagnostician draws correspond to any structure recoverable from behavior, or whether the diagnostician produced them.

The design applies three instruments to the same organization, and each is informative whichever way it comes out.

## The three instruments, and what could go wrong with them

First, the rate-separation study. Simon defined a level as a separation in interaction rates: parts within a block interact much faster than blocks interact with each other.[^ch29-simon] That is computable from data that did not exist when Beer worked and now exists everywhere, in message logs, ticketing systems, transaction graphs, incident propagation and deployment dependencies. Take a set of organizations that have been diagnosed with the VSM, build the interaction matrix, look for block structure, and ask whether the blocks coincide with the recursion levels the diagnostician drew. Simon's blocks and Beer's recursions are related but not identical objects. Beer's criterion, set out in Chapter 10, is viability, not interaction density, and his System Two exists because the operational units inside one recursion affect each other enough to oscillate without it.[^ch29-beer-pathologies] Coincidence would therefore be a convergence of two criteria, not a confirmation of one.

The contestable choice is which trace stands as the proxy. Message logs record who talks to whom and not what the talk decides. Ticketing systems record work formalized enough to be ticketed and miss the rest. Deployment dependencies describe a technical architecture that may or may not be the organization's. Each can be argued to measure the wrong thing. The alternative design builds the matrix from two or three independent traces in the same organization and treats agreement between them as a precondition for treating any of them as evidence about the diagnosis. It costs more, and it removes the easiest objection: that a negative result merely reflects a badly chosen data source.

Second, the time-span study, on the same organizations and preferably in different hands. Jaques's time-span of discretion attaches a number to a role, and a stratum follows from it.[^ch29-jaques] Measure it at the boundaries the diagnostician drew. If Beer and Jaques are describing the same structure, the boundaries should coincide. Beer warned against exactly that reading: "If the term *recursion* is allowed to become a meaningless appendage to the term *organizational level*, then we do not have a valid model of a viable system."[^ch29-beer-level] A time-span stratum is a managerial stratum inside one hierarchy; a recursion is a viable system with its own metasystem. On Beer's terms the two need not coincide, and where they do, the diagnostician may have drawn levels of management and called them recursions. The registration has to say in advance how coincidence and non-coincidence will each be read, and we register the rule here. Where time-span strata coincide with diagnosed recursions and the units fail Beer’s viability test, the reading is management levels drawn as recursions. Where diagnosed recursions cross strata and the units pass the test, the reading is support. Partial overlap is reported, not scored. The limit is availability. Time-span can be established only where role records or interviews support it, which is why the receipt in Chapter 5 makes this instrument conditional on what the records permit.

Third, the variety instrument. Galbraith's descendants spent four decades building construct-validated measures of information-processing requirements, of information-processing capacity and of the fit between them, and testing them on large samples.[^ch29-oipt] Nothing prevents the same being done for the variety claims, except that nobody has done it. A fit measure of that kind suits Beer's own use of variety better than any absolute count. He ruled the count out himself ("Of course we could not") and said the measure's use lay in "matching state generators," not in "the naive use of counting states."[^ch29-beer-matching] This instrument is the least developed of the three and we do not claim it is ready. It would have to be adapted rather than applied.

**The decision rule, registered in advance.** The block-detection rule is fixed and written down before the interaction data is inspected: what counts as a block, at what threshold, and how a block is matched to a diagnosed level. The rule we propose, as in Chapter 5, is a nested degree-corrected stochastic block model with the number of levels chosen by minimum description length, which leaves no free threshold for the blocks themselves.[^ch29-sbm] Two analysts work blind, one holding the diagnosis and the other the matrix, and neither of them makes the comparison afterward. Registration matters more here than the threshold chosen. A rule chosen after the matrix has been seen can be tuned until the blocks agree with the levels, and nobody, the analyst included, will be able to tell afterward whether it was.

**Receipt — Experiment one, first stage.** One organization, one interaction matrix, one existing diagnosis, two analysts each blind to the other's material, a block-detection rule registered before the data is seen, comparing detected blocks, diagnosed recursion levels and, where records permit, Jaques's time-spans of discretion. Priced in Chapter 5 at two analyst-weeks plus data preparation. Negative result: the diagnosed levels have no proxy in observed interaction structure. That counts against the equivalence of Beer's criterion with Simon's or Jaques's, and against any claim that recursion levels can be recovered from interaction data; it does not by itself count against viability as Beer's criterion. The multi-organization extension — several diagnosed organizations, several traces each, the variety instrument adapted and applied — is the natural continuation and is not yet priced; it should not be commissioned until the single-organization stage has produced a result either way.

## What experiment one would settle, and whom it would cost

The design bears on more than the headline question. Where diagnosis and proxy diverge, which one is wrong? The blind design keeps that open rather than settling it, since neither instrument has prior standing over the other. Is Beer's criterion equivalent to Simon's, to Jaques's, to neither, or to both at different scales? And underneath all of them: is there any quantity in this model that two observers can measure independently and agree on?

The worthless version of this study is a paper that proposes a metric and applies it to one sympathetic case. Two things separate the useful version from it. The first is inter-observer agreement: the study is only worth doing if two analysts work blind and their results are compared. The second is the publication of a negative result, on the terms set out under the conditions shared by both experiments. A finding that VSM recursion levels do not track rate separation would be the most informative single result this community could produce.

If the levels do track something, the practical yield is that a diagnostician could show a client the boundaries were not chosen arbitrarily. Disputes about the system in focus would acquire evidence without thereby becoming settleable, since who may draw a boundary is a separate question from what the data supports. And a failed diagnosis could be told apart from a badly bounded one, which at present it cannot.

The audiences already exist. Organizational network analysis has the data and the methods and no theory of levels. The information-processing tradition in information systems and operations management has the instruments. The requisite-organization practitioner network has fifty years of time-span data and a strong interest in independent corroboration. Complex-systems and modularity researchers treat near-decomposability in real organizational data as an open empirical question. And every consultancy selling organizational design would rather have a number than an argument.

As far as our searches found (Crossref and web, September 2026), nobody is working on the joint question.[^ch29-search] Organizational network analysis measures interaction structure and does not ask about viability. The requisite-organization network measures time-span and does not talk to cyberneticians. The information-processing literature measures fit and has rarely engaged System Four. The three instruments live in separate literatures, and our search has not yet found a published study that puts any two of them on the same organization.

The cost of a result falls on us first. A measurable criterion means a diagnosis can be shown to have been drawn wrong, and our diagnostic authority currently rests on the fact that it cannot. The pedagogy is exposed too, since a criterion that disagreed with the model's own would have to be taught alongside it. And so is the model, if the answer is that nothing recoverable from data corresponds to a recursion level at all.

## Experiment two: test whether a VSM diagnosis predicts

*The evidence base Chapters 9 and 14 showed the neighbors have and we do not.*

**The claim under test.** Ostrom's field built a coding instrument of four hundred and fifty-five questions, coded eighty-six cases against it, and read fourteen in depth, five of them failures. The result was then independently reviewed and reanalyzed three times, and the reviews reported which of her principles the evidence did not support.[^ch29-ostrom] The most recent systematic review of the Viable System Model started from seven hundred and sixty-eight Scopus records and forty-five more found by hand, read a hundred and twenty in full, kept twenty-one studies, and concluded that most of what exists is model and simulation rather than implementation.[^ch29-fathi] The gap is not one of counting. It is that one field built an instrument, ran it, and printed the findings that went against it. The experiment asks whether a VSM diagnosis predicts anything a simpler account does not.

## The register, and the test built on it

The instrument is a retrospective register of VSM interventions, designed to test predictions rather than collect success stories. The register itself is unglamorous and overdue. Each entry records the system in focus and who named it, the boundary judgments made and by whom, the diagnosis, the intervention, the success criteria stated in advance, the outcome, and the client's own account of the outcome alongside the consultant's. The two accounts stay separate on the record rather than being reconciled into one, because the divergence between them is itself a finding.

None of this has to be invented. Ostrom's coding instrument is published and could be adapted. So could the sociotechnical tradition's, which reviewed a hundred and thirty-four experiments in 1982 and said candidly, in the same paper, that most of them had not done the thing the theory claimed.[^ch29-pasmore] Adapting a published instrument is cheaper, and easier to criticize: a reader can set our version beside the original and see what we dropped.

**The decision rule, registered in advance.** Diagnoses and predictions are coded from documents created before outcomes were known, by someone who did not conduct the intervention. The VSM claims are then compared with a simple baseline and with at least one neighboring framework, on criteria fixed before the comparison is run. Each case faces the same test, in the same terms for all three: did the diagnosis identify, in advance, a failure mode or outcome that the baseline and the neighboring framework did not? Around that test, the register makes configural analysis, structural disconfirmation and the intervention question possible across cases. Those are the reasons to build a register rather than run a single comparison.

Three choices are contestable. The first is retrospection. Coding from historical engagement documents is fast and cheap, and it depends entirely on what was written down at the time. A prospective design, registering diagnoses and criteria before interventions begin, is stronger and returns its first result years later. We propose the retrospective version because it can start now, and we accept that its findings are weaker in exactly the way that dependence implies. The second is the comparator. A badly chosen baseline makes the model look good, so the baseline has to be named in the registration, not selected once the cases are in hand. The third is breadth against depth. Ostrom's field coded eighty-six cases and read fourteen in depth; a small register can imitate the depth or the breadth but not both, and for the opening stage we have chosen depth.

**Receipt — Experiment two, first stage.** Ten interventions, retrospectively coded by someone who did not conduct them, against criteria reconstructed from the original engagement documents rather than from memory, and compared with a simple baseline and at least one neighboring framework. Publish the ten, including those that failed and those where nobody wrote the criteria down. That establishes both the instrument and the honesty of the exercise, and it can be done in a year. Negative result: VSM diagnoses do not identify, in advance, failure modes or outcomes that the baseline and the neighboring framework also identify. The register's continuation beyond the first ten entries is not yet priced.

## What experiment two would settle, and whom it would cost

The questions in reach are the ones a client asks first. Does a VSM intervention improve anything, against criteria fixed before it starts? Which of the five functions, coded absent or disseminated under the coding in Chapter 28, actually predicts the pathologies Beer says it predicts, such as oscillation where System Two is missing and the collapse of the metasystem into System Three where System Four is empty?[^ch29-beer-pathologies] Do those pathologies occur in organizations where the model was never applied, which is to ask whether they are findings or artifacts of our vocabulary? What is the failure rate, and what distinguishes the failures from the rest? And is our record unblemished because the model is good, or because it has never been tried in a way that could mark it?

The worthless version here is a case-study collection, which we already have, written by people with an interest in the outcome. A register differs from a collection in two respects that have nothing to do with size: the criteria were fixed before the outcome was known, and the coder had nothing to gain from the answer.

If the answer is favorable, the register buys a response to the first question any serious client, funder or regulator asks, which at present we cannot answer. It gives training a basis for separating what is known from what is believed. And it answers the reasonable charge that a framework absorbing every outcome is a language rather than a theory.

Again the interested parties are mostly outside the tradition. Metaphorum and the practitioner networks, whose credibility depends on this more than is generally admitted. Evidence-based management, which would take the data seriously and the vocabulary not at all. The commons and sociotechnical research communities, which have run this exercise on their own traditions and know how it goes. Journals that publish registered reports. And the model's critics, for whom a register would be the first thing to argue with rather than about.

No register exists. Quantitative tests do: two earlier St. Gallen dissertations, which we have not read (Chapter 18), and the two we examine there. One is Schwaninger and Scheef's 2016 survey.[^ch29-schwaninger] The other, and the nearer, is Michael D. Pfiffner's 2017 dissertation,[^ch29-pfiffner] an instrument rather than a survey: an operationalization of the model's own conditions, applied to 135 organizations by third-party raters independent of the organizations’ management (22 respondents, five of them consultants or former consultants of the author’s firm; one rated 23 of the 135 cases).[^ch29-pfiffner-raters] Neither is a register, since neither records interventions, criteria fixed in advance, or outcomes over time. But the second changes what this chapter should recommend first, and the change is set out below. The 2025 systematic review is an audit of the literature. Its findings (limited empirical validation, few real implementations, almost no longitudinal work) make the case for this direction and cannot substitute for it. The neighboring fields have done the work and published their instruments, which removes the last excuse.

The cost falls on everyone in the tradition who bills for a diagnosis, ourselves included. A register with pre-stated criteria and published failures is an invitation to be measured, and for fifty years we have been able to decline it. In the short run it also puts the model's reputation at risk, because the first honest register will look worse than the case literature it replaces. Every honest register looks like that at the beginning. A community that cannot survive its own first negative result has no claim to be a science.

## A cheaper check, if the data can be had

We wrote the pages above before we had read Pfiffner's dissertation. Reading it produced a candidate that costs less than both experiments and answers a question neither of them reaches.

The data already exist. One hundred and thirty-five organizations, seventy-four of which had undergone an existential crisis and sixty-one of which had not; thirty-five independent variables; indicators organized by the model's own functions, with separate blocks for Systems One to Five and for the principles of recursion, requisite variety and communication. The expensive part of an empirical study, finding the cases, reaching the informants and collecting the ratings, was done in Switzerland between December 2014 and April 2015.[^ch29-pfiffner-data]

The study did not test whether any function is necessary, and Chapter 18 explains why it could not. The measurement model is reflective and specified downward, with viability treated as a single latent factor of which the functions are interchangeable expressions. In that specification the functions are expected to move together. They duly do, and no possible pattern of results could have shown one of them to be dispensable.

**The claim under test.** The model asserts that the five systems are necessary: without each, no viability. A reflective model cannot assess that. A formative one can come closer, because it treats the functions as constituting viability rather than expressing it, and then asks what each block contributes that the others do not. Two limits follow from Beer's own statements. He held that "in a viable system all five subsystems are dependent on each other,"[^ch29-beer-dependent] so a block that adds little once the others are in the model may be moving with them rather than doing nothing. And Pfiffner's items rate how well each function works, not whether anyone performs it. The re-analysis tests whether each function's measured quality adds discrimination of crisis that the others do not supply. It tests necessity in Beer's sense only for cases in which absence is coded, as in Chapter 28, and the crisis groups are not the same outcome as Beer's viability.

**The design.** Re-specify the measurement model as formative and re-estimate on the existing data. For each of the five systems and each of the three principles in turn, remove that block and record the change in the model's ability to separate the crisis group from the non-crisis group. Because every run uses the same data, the comparison is internal and clean.

**The decision rule, registered in advance.** Before re-estimation, and in writing: the drop in discrimination a block must produce for its function to count as carrying weight the others cannot supply, and the threshold below which a function is reported as not distinguishable from the rest of the instrument. Fix which statistic is used and what counts as a difference, and how variance shared between blocks will be read, given Beer's statement that the five are mutually dependent. Publish the rule before the first run.

**The contestable choices** are real. Re-specifying someone else's model is an interpretive act, and the original author should be a party to it, not its object. The instrument's items were written for a reflective design and some may not carry formative weight; that would itself be a finding, about the instrument rather than the model. The outcome variable inherits the problem Chapter 18 describes, since it contains inverted viability items. The re-analysis should therefore be run both on that construct and on the original crisis and non-crisis group membership, which the dissertation reports but did not use in the final model. The ratings are clustered, about six cases per rater, and the analysis has to allow for that. And the sample is Swiss, from the mid-2010s, and selected at the extremes.

**Receipt — the formative re-analysis.** Obtain the dataset, re-specify the measurement model as formative, and run the block-removal comparison under a rule published beforehand, on both the composite outcome and the original group membership. Price: one request to the author, and roughly three to four weeks of a statistician competent in structural equation modelling; no fieldwork, no access negotiation, no new instrument. This is cheaper than either experiment above by a wide margin, and it is the only route we have found to the necessity claim that does not require collecting a new sample. The re-analysis cannot code a function as absent in the sense of Chapter 28, because the instrument has no such answer: raters scored how far a described ideal matched the organization, on an eight-point scale with "don't know" as the only other option.[^ch29-pfiffner-scale] If a subsystem's removal makes no difference under the registered rule, the claim that the subsystem carries weight the others cannot supply is in trouble — and Schwaninger and Scheef's System Three result, reached from an entirely different instrument and sample, says where to look first. If every block carries weight, the model has its first corroboration of necessity from a design that could have refused it.

**What would make this worthless.** Running it without the rule fixed in advance. Running it without the author. Reporting only the blocks that behaved. And treating a Swiss sample of 135 as more than it is: one dataset, at the extremes, from one decade.

## Conditions shared by both experiments

The two designs share a small set of conditions, and neither is worth commissioning without them. Criteria and exclusions are registered before results are inspected. Analysts work blind wherever the material permits. Failures stay in the dataset. Missing criteria are recorded as findings, not repaired from memory. Data and exclusions are open to anyone who wants to check the analysis. Negative results are submitted for publication.

The last condition is the one that costs something. A finding that diagnosed levels do not track independent measures, or that VSM predictions do not beat the alternatives, must be publishable and must be published; the conditions under which such a finding would count against the model are the ones Chapter 28 has already set out. Neither experiment requires a theoretical advance. Both require the willingness to be seen losing.

[^ch29-beer-paraphrase]: The wording is our paraphrase of Beer and not a quotation. His nearest statements: the operational elements of a viable system exhaust its "basic activities, namely those which exist to do what the system does" (Stafford Beer, *The Heart of Enterprise* (John Wiley & Sons, 1979), 116); a division that "could in principle be hived-off as a separate business" is a viable system, and the accounting system that prepares the group accounts is not (ibid., 118); and "the viable system is necessarily a producer of the organization, and not just an adjunct to it, however important" (Stafford Beer, *Diagnosing the System for Organizations* (John Wiley & Sons, 1985), 8).

[^ch29-beer-rigor]: Stafford Beer, *Beyond Dispute: The Invention of Team Syntegrity* (John Wiley & Sons, 1994), 229; see also Stafford Beer, *The Heart of Enterprise* (John Wiley & Sons, 1979), 307–8, and Stafford Beer, *Diagnosing the System for Organizations* (John Wiley & Sons, 1985), 2, on "an absolutely precise definition of viability."

[^ch29-beer-heur]: Beer, *Diagnosing the System*, 19; see also 4, on considering "a trio of viable systems at any one time."

[^ch29-beer-level]: Beer, *Heart of Enterprise*, 308, where the passage continues: "The levels of recursion in any such organizational model answer to mathematical criteria." Chapter 10 states the criterion in full.

[^ch29-beer-matching]: Beer, *Heart of Enterprise*, 86–87: "the potential use of the variety measure is not the naive use of counting states, but of matching state generators."

[^ch29-beer-pathologies]: Beer, *Heart of Enterprise*, 177 ("without it System One would be unstable — it would go into an uncontrollable oscillation") and 265 ("the whole metasystem collapses into System Three").

[^ch29-beer-dependent]: Beer, *Diagnosing the System*, 128.

[^ch29-beer-observer]: Stafford Beer, *Decision and Control* (John Wiley & Sons, 1966), 247 and 243; Beer, *Diagnosing the System*, 98. On predictive usefulness, *Decision and Control*, 113 and 243.

[^ch29-simon]: Herbert A. Simon, "The Architecture of Complexity," *Proceedings of the American Philosophical Society* 106, no. 6 (1962): 467–82. Chapter 11 quotes the near-decomposability argument with pages.

[^ch29-jaques]: Elliott Jaques, *Measurement of Responsibility: A Study of Work, Payment, and Individual Capacity* (Tavistock Publications, 1956), for the measure, and *Requisite Organization: The CEO's Guide to Creative Structure and Leadership* (Cason Hall, 1989), for the strata; Chapter 11 sets out the empirical record.

[^ch29-oipt]: For example Michael L. Tushman and David A. Nadler, "Information Processing as an Integrating Concept in Organizational Design," *Academy of Management Review* 3, no. 3 (1978): 613–24, https://doi.org/10.5465/amr.1978.4305791, and G. Premkumar, K. Ramamurthy, and Carol Stoak Saunders, "Information Processing View of Organizations: An Exploratory Examination of Fit in the Context of Interorganizational Relationships," *Journal of Management Information Systems* 22, no. 1 (2005): 257–94, https://doi.org/10.1080/07421222.2003.11045841, as cited in Chapter 12, which calls them points on a line, not a survey of it.

[^ch29-sbm]: Brian Karrer and M. E. J. Newman, "Stochastic Blockmodels and Community Structure in Networks," *Physical Review E* 83, no. 1 (2011): 016107; Tiago P. Peixoto, "Hierarchical Block Structures and High-Resolution Model Selection in Large Networks," *Physical Review X* 4, no. 1 (2014): 011047. We have not read either paper in full; the method is to be confirmed with a network methodologist before registration.

[^ch29-ostrom]: As in Chapter 14: Elinor Ostrom, Arun Agrawal, William Blomquist, Edella Schlager, and S. Y. Tang, *Common-Pool Resources (CPR) Coding Manual* (Workshop in Political Theory and Policy Analysis, Indiana University, 1989), with the figure of 455 questions from Michael J. Bernstein et al., "Mapping Ostrom's Common-Pool Resource Systems Coding Handbook to the Coupled Infrastructure Systems Framework to Enable Comparative Research," *International Journal of the Commons* 13, no. 1 (2019): 528–52, https://doi.org/10.18352/ijc.904; Elinor Ostrom, *Governing the Commons: The Evolution of Institutions for Collective Action* (Cambridge University Press, 1990); and the three reviews, Cox, Arnold and Villamayor-Tomás (2010), Baggio et al. (2016) and Gari et al. (2017), cited in full in Chapter 14. Chapter 14 also records that the manual itself lists eighty-five coded cases, not eighty-six.

[^ch29-fathi]: Mohammad Reza Fathi, Soraya Birami, Alireza Payvar, and SeyedAli Doorafshan, "A Systematic Review of the Viable System Model: Applications, Insights, and Future Directions," *Journal of Systems Thinking in Practice* 4, no. 3 (2025): 109–45, https://doi.org/10.22067/jstinp.2025.91552.1136. The selection figures are from the PRISMA diagram, 113 (fig. 2); on reliance on simulation rather than implementation, 131 (table) and 139.

[^ch29-pasmore]: William Pasmore, Carole Francis, Jeffrey Haldeman, and Abraham Shani, "Sociotechnical Systems: A North American Reflection on Empirical Studies of the Seventies," *Human Relations* 35, no. 12 (1982): 1179–1204, https://doi.org/10.1177/001872678203501207, as cited in Chapter 13.

[^ch29-schwaninger]: Markus Schwaninger and Christine Scheef, "A Test of the Viable System Model: Theoretical Claim vs. Empirical Evidence," *Cybernetics and Systems* 47, no. 7 (2016): 544–69, https://doi.org/10.1080/01969722.2016.1209375. Its hypothesis on System Three was rejected; Chapter 18 discusses the result and how the paper explains it.

[^ch29-pfiffner]: Michael Dominik Pfiffner, "System Viability of Organizations and the Aetiology of Organizational Crisis: A Quantitative Assessment of Stafford Beer's Viable System Model" (PhD diss., Utrecht University, 2017), https://hdl.handle.net/1874/356772. Pages below follow the Utrecht repository version, printed pagination.

[^ch29-pfiffner-data]: Pfiffner, "System Viability of Organizations": data collection, 179; thirty-five independent variables in the sample of 135, and the reflective measurement model, 180; seventy-four crisis and sixty-one non-crisis cases, 183. The final path model drops five of the twenty-two VSM items, among them P.1, upward recursion (200–201), so the block list in a registered rule has to be fixed against the final model and not the questionnaire.

[^ch29-pfiffner-scale]: Pfiffner, "System Viability of Organizations," 146 and 159.

[^ch29-pfiffner-raters]: Pfiffner, "System Viability of Organizations": "knowledgeable and unbiased third party assessors," unbiased "as compared to managers or owners of the respective organizations," 133 and n. 23; five respondents addressed "among (ex-) consultants of the author's employing firm," 179; twenty-two respondents from sixteen organizations, an average of 6.1 cases each, one of them contributing twenty-three, 181.

[^ch29-search]: Crossref bibliographic search, 29 September 2026, first eight results for each of "organizational network analysis viable system model", "information processing view viable system model", "time span of discretion viable system model", "interaction matrix recursion viable system model" and "near decomposability organization viable system model"; and one web search the same day combining "viable system model" with the three instruments. The nearest hit, judged from its title, is Gary Preece, Duncan Shaw, and Haruo Hayashi, "Using the Viable System Model (VSM) to Structure Information Processing Complexity in Disaster Response," *European Journal of Operational Research* 224, no. 1 (2013): 209–18, https://doi.org/10.1016/j.ejor.2012.06.032, which is not a study of two instruments on one organization. Manuel E. Sosa, Steven D. Eppinger, and Craig M. Rowles, "The Misalignment of Product Architecture and Organizational Structure in Complex Product Development," *Management Science* 50, no. 12 (2004): 1674–89, https://doi.org/10.1287/mnsc.1040.0289, set product architecture beside organizational structure outside the VSM, as the title reports; we have not read it.