The Market Test

The claim

Metaphorum’s website states that after fifty years no one has been able to find a case study in which the Viable System Model has failed.^[1] Schwaninger and Scheef, from a different tradition and with a different purpose, record the same absence from the other end: to their knowledge, unsuccessful or failing applications of the model have not been reported in the literature.^[2] Read together, the two sentences close a loop. The reading is ours; neither source makes the joint claim. Failures are not reported, so none is found, so the record is clean, so there is nothing to report.

The community nearest to us publishes a warrant, and the warrant cannot be false. Underneath it is a mechanism that needs neither a rhetorical trick nor anybody’s dishonesty: a selection effect operating on a professional practice. It would produce the same result if every practitioner in the field were scrupulous. Most of them are.

What “failed” would have to mean

The sentence “no case study in which the VSM has failed” has no truth conditions, because this literature, as far as we have found, has never defined what it would mean for a model to fail in an engagement.^[3] Take the candidates one at a time, with what each would require us to have recorded.

A diagnosis could fail descriptively: the analyst maps five systems onto an organization, and the mapping turns out not to hold. A function said to be present is absent; a channel said to carry variety carries none. Establishing that requires an independent re-mapping by someone who did not produce the first. That is Chapter 5’s problem about the ruler, and we have never run it.

A diagnosis could fail predictively: the model names a pathology, the pathology is left unrepaired, and the predicted consequence does not follow. Establishing that requires the prediction to have been written down in advance with a date on it, which our engagement reports do not do.

An intervention could fail: the recommended structural change is made and the organization does not improve, or declines. Establishing that requires a measure of improvement stated and fixed before the change. That returns us to Ulrich’s third question in Chapter 16, and it is not standard practice either.

Or the engagement could fail: the client stops the work. This one we can observe. It is the only one we can observe, and it almost never reaches print. Beer printed a few, among them four proposals, in four countries, in which “the impending crisis exploded before work could begin.”^[4]

Four meanings: three that the field has never instrumented, and one that it almost never publishes. Against that background the claim that no failure has been found reports nothing about the model. It reports the design of the observation system.

The market test, stated plainly

The practitioner network does have an operative test, and its honesty should be acknowledged before it is criticized. Patrick Hoverstadt has put it directly: if it doesn’t work, they don’t ask you back.^[5] That is a real feedback loop. It runs fast, it costs the practitioner money when it fires, and it is more than most consulting frameworks submit to. The warrant is older than the network: Beer’s own confidence in the model came, “subjectively speaking,” from “the speed with which the diagnosis can be made.”^[6]

It is also selected on the outcome variable, in the direction that hides failure. A client who does not ask you back does not co-author the case study. The practitioner who loses the account writes nothing; the practitioner who keeps it writes the chapter. Every mechanism that could turn a negative result into a published finding runs the other way. Confidentiality agreements cover the engagements that ended badly, reputational cost attaches to reporting them, and no venue has asked for them. The founder did. Beer proposed “a small team of organizational pathologists,” organized “more like a lifeboat crew,” to reach a dying organization, resuscitate it if possible and, failing that, perform “a post mortem … before rigor mortis had set in,” because much had been learned from systems “viable no more.”^[7] Nothing in the practice described in this chapter answers to the proposal. The market test is a test of the practitioner, not of the model, and nobody records it even as a test of the practitioner.

The same structure appears in the network’s other warrants. Angela Espinosa writes that the model “has been tested in several types of businesses, public and not-for-profit organisations and communities”.^[8] The word “tested” there means “applied”: the cases are ones in which the model was used, reported by the people who used it. Her 2025 reply to Michael Jackson’s charge that the model is unitary and functionalist answers largely by pointing to recent applications.^[9] That answers a question about usefulness. Jackson’s question was about frame.

José Pérez Ríos’s pathology taxonomy repays the most attention, because it is the field’s best piece of derived diagnostic machinery and shows the problem in its cleanest form. As Chapter 9 says, the taxonomy is generated from the structure; it is not assembled from observed breakdowns.^[10] We have not been able to find a validation of it against failure data: a set of organizations that failed, independently classified, with the taxonomy asked in advance which pathology should be present. Until that exists, the catalog is a vocabulary with a derivation. That is a real asset, and it is not evidence.

The selection rule, in writing

Everything above is an inference about how a record comes to be clean. The inference turns out to be unnecessary. SCiO (Systems and Complexity in Organisation, the UK professional body for systems practitioners and the network through which most VSM practice in Britain is taught, accredited and sold) publishes the rule at the head of its case-study library:

The following are case studies of successful application of systems methodologies.^[11]

That is the inclusion criterion, stated by the community itself, in its own voice, on a public page. A failed application is not suppressed. It is out of scope by construction, and the sentence saying so is the first thing a reader sees.

Of the twelve entries we were able to read in full, none reports a failure, a negative result, an abandoned engagement, or a limitation of the method used.^[12] The count is what the criterion predicts, and the criterion is not hidden.

The community’s efficacy claims point to this same library as their evidence base. SCiO’s own page for the model states that the Viable System Model “has proven to be a scalable, super-fast, flexible, rigorous, approach” and that it “has been used successfully” for improving productivity, governance and resource matching, across “organisations of all types and scales from small teams up to whole nation states.”^[13] Each of those sentences is true of the case library as constituted. None of them could be false of it. Anything stronger the site does not say, anywhere we could find: no claim that the model has never failed, no count of applications, no years-of-use figure. The strongest single claim on the public site turns out to be a percentage, and Chapter 18 follows it to its source.^[14]

One detail completes the picture, and it is small enough to be missed. The only artifact in this community named “The VSM Test” is a thirty-question multiple-choice quiz with an eighty-five percent pass mark, testing a reader’s recall of a twenty-eight-page guide.^[15] It tests the practitioner. Nothing in the tradition’s public apparatus tests the model, and the one thing named as though it did tests whether you have memorized it.

Why this community and not another

A reader could take this chapter as an attack on Metaphorum and SCiO, and that reading would miss the point twice over.

It would miss it first because these are the people who kept the work alive. Between Beer’s death and the present, this network ran the conferences, taught the courses, maintained the archive and did the client work, and that is why there is a tradition to stress-test at all. A tradition that had been abandoned would have no defects worth cataloging, because nobody would be teaching them.

It would miss it second because nobody in the network has defended the claim under examination as an argument in print. It is a line on a website. Nobody sat down and reasoned their way to an unfalsifiable warrant. The sentence is the natural summary of fifty years of practice in which no instrument for recording failure was ever built, and to the person writing it, it reads as confident rather than evasive. That is what makes it a false friend rather than an adversary. An adversary’s argument can be answered. A warrant of this shape cannot be answered, only replaced, and replacing it is cheap.

One further asymmetry bears on the next chapter. The 2016 St. Gallen test set out to falsify the model, and it reported a rejected hypothesis. It was not the first attempt: its authors name two earlier St. Gallen dissertations, by Benjamin Frost (2005) and Cristina-Ioana Crisan Tran (2006), which we have not read.^[16] The only response from this network that we have read treats the 2016 test as corroboration. Espinosa’s 2023 book says the study “demonstrated that the claim that the VSM provides the necessary and sufficient precondition for the viability of any organisation is well justified,” and that its authors “could not falsify that claim on empirical grounds, which corroborates the effectiveness of the model.” The rejected hypothesis is not mentioned.^[17] Other authors from the network cite the study, and we have not yet read how. In the one case we have read, the community used the test by citing its conclusion.

What this chapter does not establish

It does not establish that the model has failed anywhere. No such case is presented here, and this book has none.

It does not establish that the practitioners are wrong about their engagements. They are almost certainly right that most of their clients were helped, and nothing here contradicts that.

What it establishes is that the field’s strongest public claim is a statement about its recording practices wearing the clothes of a statement about its model, and that the two can be separated at a price small enough to name.

Receipt — define failure, then open the register. Publish a four-line operational definition of a failed VSM application, covering the descriptive, predictive, interventional and engagement senses above, with a rule for each saying what evidence would establish it, and a preamble saying which notion of viability the definition uses: Beer’s capacity to maintain a separate existence, or bare survival (Chapter 18’s receipt faces the same choice). Then invite the practitioner network to enter its own terminated and unsuccessful engagements against that definition, anonymized, in the register proposed in Chapter 30, with a twelve-month amnesty window and no requirement to name clients. This is Beer’s post-mortem proposal, carried out on paper. Price: one afternoon for the definition, circulated for comment at one Metaphorum conference; the register cost is already booked in Chapter 30 and this adds a single intake form to it. If the register fills, the “no failed case” claim retires and the field gains its first negative evidence base. If it stays empty after twelve months with the definition in place and the amnesty offered, that is itself a finding, and a much stronger one than the claim it replaces: the absence would then be an observed absence rather than an unobserved one. Either outcome is worth more than the present sentence, and the present sentence is the one thing the exercise cannot leave standing.

Handover

This community never framed a test. The next chapter deals with the one group that did.