Holarchy Without Hierarchy

Volume VI — Governance and the Commons

THE PLATE

A small group walking and talking together through a warm office, lamps lit overhead. — Plate VI.03The Room Where It Was Already Decided.A nested structure is not a room without authority. It is a room that holds the authority for the thing it is about, and hands nothing upward that it can finish itself.

THE LETTER

You have probably been in both kinds of organisation, and you have probably noticed that the second kind is not simply the first kind with the managers removed.

You have sat in the one where a decision that four people could have taken in ten minutes took eleven weeks, because it had to travel to the level that held the budget and then travel back with somebody's edit on it. And you have sat in the one where the decision was taken in ten minutes and then quietly unravelled over the following year, because nobody could say afterwards who had taken it, and the person who disliked it most simply kept raising it until it went away.

Both of those are failures of the same variable, and this chapter is about that variable. It is not hierarchy. It is not flatness. It is where a decision is allowed to stop.

The vocabulary for this is sixty years old and unusually precise. Arthur Koestler gave us the holon — a thing that is simultaneously a whole and a part, with a face turned inward toward what it contains and a face turned outward toward what contains it. Elinor Ostrom gave us the eighth design principle — that a commons of any size holds only if its appropriation, monitoring, enforcement and conflict-resolution activities are organised in multiple layers of nested enterprises. Herbert Simon gave us the reason both of them work: complex systems that survive are nearly decomposable, which is to say almost all of their interactions happen inside their parts and only a few cross between them.

What none of the three gave us is a number. So that is what this chapter is for. We are going to compute the coordination cost of a nested structure against a hierarchy at a stated span and a stated depth, and we are going to end with a single measurable quantity — the containment rate — that tells you which one your organisation should be running, at your size, this year.

The answer is not the one you expect, and it is better than the one you expect.

— The Editors

DISCOVERY

What is already working

Buurtzorg, in the Netherlands. Jos de Blok founded it in 2006 with four nurses, on a simple structural bet: a district nursing team of no more than twelve people, holding its own intake, its own rostering, its own hiring and its own clinical judgement, with no team leader. By 2017 it was carrying on the order of 10,000 nurses. The coordinating cadre that sits above those teams is about 50 coaches, and the shared back office is about 50 people — a total non-nursing establishment of 1.00% of the front line.

That is the headline, and the KPMG evaluation of 2015 supplies the outcome beside it: Buurtzorg delivered a given client's care in roughly 40% fewer hours than comparable providers, with overhead running at about 8% of revenue against a sector figure near 25% — a gap of 17 points. KPMG's national estimate, if the model were generalised across Dutch home care, was on the order of EUR 2.0 bn a year. Buurtzorg has also been named the Netherlands' best employer repeatedly, which matters here because it is the variable most flat structures lose.

Morning Star, in California. Chris Rufer's tomato processor runs roughly 400 full-time colleagues on revenue of about $700m — $1.75m per colleague — and handles a quarter to a third of all processing tomatoes in the United States. It has no managers at all. What it has instead is the Colleague Letter of Understanding: an annual negotiated agreement in which each person writes down their personal commercial mission, the activities they will perform, and the metrics by which they will be judged. Gary Hamel's 2011 account gives the shape of an average CLOU precisely — around 30 activity areas and some 50 metrics, negotiated directly with about 10 colleagues, the people whose work touches yours.

The CLOU is the invention. It is not a job description, because nobody issues it; it is a contract between peers, and it converts the question "who is in charge of this" into "who did I agree this with."

Haier, in Qingdao. Zhang Ruimin began the RenDanHeYi transition in 2005 and carried it through a restructuring that removed on the order of 10,000 middle-management roles — at a European loaded cost that would be around £700.0m a year, though Haier's own figure is not published. What replaced them is roughly 4,000 microenterprises, of which about 200 face the customer directly and the rest supply them, each contracting with the others on terms they negotiate, each able to raise outside venture capital, each able to hire and to dismiss. Haier bought GE Appliances in 2016 for about $5.6bn and ran the same model through it.

And the design principle underneath all three. Ostrom's Governing the Commons set out eight design principles common to long-enduring common-pool resource institutions. The eighth is the nesting one. Cox, Arnold and Villamayor-Tomás reviewed 91 subsequent case studies in 2010 and found the principles well supported — restating them as eleven, which is to say they subdivided three rather than discarding any.

Four cases, four decades, one pattern: in each, the unit that holds the work also holds the decision rights over the work, and the layer above holds only what genuinely spans units. That is nesting. It is not the absence of structure, and it is emphatically not the absence of authority.

THE ARITHMETIC

What works, what does not, and where the line sits

Now the cut, and it is the whole chapter.

A nested structure and a hierarchy are the same tree.

Take 10,000 front-line people in teams of 12. That is 834 teams. Group the teams at a span of 5 — the span that FEMA's incident-command doctrine gives as optimal, with a workable range of three to seven — and the tree is 6 levels deep. That is true of Buurtzorg and it is true of a conventional provider of the same size. Same span. Same depth. Same number of levels. Nesting does not remove a single layer from the diagram.

What it changes is one rule. In a hierarchy, a decision rises until it meets authority. In a holarchy, a decision rises only when it cannot be contained. Everything measurable below follows from that single substitution, and nothing follows from flatness — which is why the organisations that got the benefit kept their trees, and the ones that tried to delete the tree mostly lost it.

The escalation ledger. Let c be the containment rate: the share of decisions a unit can close without sending anything upward. A decision escapes one level with probability 1 − c, so the expected number of levels it traverses is a truncated geometric sum:

        L(c, d)  =  ( 1 − (1 − c)^d ) / c

  c  containment rate      d  depth of the tree in levels

At d = 6, with an authority-threshold rule holding c = 0.55 and a containment rule holding c = 0.85:

  hierarchy    L = 1.8031 levels      0.8031 escalations per decision
  nested       L = 1.1765 levels      0.1765 escalations per decision
  difference        0.6266 levels

At 24 coordination-requiring decisions per person per year that is 240,000 decisions across the organisation, and at four people for ninety minutes each — £270 a level — the escalation line alone is £52.04m against £11.43m.

The headcount. A hierarchy at this shape needs a leader per team and a node per group of five: 1,043 coordination posts, 10.43% of the front line. Run the span all the way down to individuals rather than to teams and it is 2,500 posts, 25.0% — which is, to within a point, the overhead the Dutch home-care sector actually carried. Buurtzorg's equivalent establishment is 100 people. The difference is 943 posts, about £66.01m a year at a loaded £70,000.

And the cost of nesting, which is real and quadratic. A team coordinated through a hub has 11 ties. A team coordinated as a mesh has 12 × 11 / 2 = 66 — 6.0× as many. Put the whole ledger together:

                              HIERARCHY        NESTED
  coordination posts              1,043            50
  posts, cost                   £73.01m        £3.50m
  inside-team coordination       £4.95m       £29.72m
  escalation                    £52.04m       £11.43m
  ------------------------------------------------------
  total                        £130.00m       £44.66m

£85.35m a year, or £8,535 per front-line person. That is the prize, and it is large enough that the interesting question is not whether to pursue it but what would have to be true for it to be false.

So here is the instrument. Hold everything except the nested structure's containment rate, and ask how low it can fall before nesting stops paying. The answer at this span and this depth is:

  break-even containment   c*  =  37.8 %

37.8%. A nested structure at this size has to close only three decisions in eight inside the unit to beat the hierarchy, because it is carrying 993 fewer coordination posts to start with. Against a modelled 85% that is a margin of 47.2 points. This is an unusually forgiving threshold, and it is the reason Buurtzorg works in a sector nobody thought it would.

Now the honest negatives, and there are three.

First: the benefit saturates, early. As depth grows, both L values converge on 1/c, so the gap can never exceed 1/0.55 − 1/0.85 = 0.6417 levels per decision, ever. At about 1,500 people the structure has already captured 88.5% of that ceiling; at 37,500 it has 97.6%. Most of what nesting can do for a decision, it does by fifteen hundred people. Going to ten thousand adds the rest of the headcount saving but almost nothing per decision. Nesting is not a large-organisation technology. It is a technology that finishes early — which is good news for a firm of two hundred and a caution to anyone selling it as a scale play.

Second: the mesh is quadratic, and it has a cap. Set the mesh's maintenance cost equal to a hub's plus a manager's integration time and solve for team size: h² − 3h − 2(m − 1) = 0. At sixty person-hours a month of integration the answer is 12.47 people, with a sensitivity band of 10.46 to 14.16 across plausible integration loads. Buurtzorg caps its teams at twelve. A Morning Star colleague negotiates a CLOU with about ten counterparties. The field found the number before the model did, which is the right order for these things. Across the whole organisation, at equal containment, the managed hub overtakes the mesh at a team size of 18.82 — so a self-managed team that drifts past eighteen is not a flatter organisation, it is a more expensive one.

Third, and this is the one that costs lives rather than money: an emergency sets containment to zero. An emergency is, definitionally, an event that spans units — that is what makes it an emergency rather than a problem. At c = 0, L(0, 6) = 6: the decision climbs the entire tree by consent. At five working days a level that is 30 working days against 5 for a structure with a pre-named commander and unity of command. Twenty-five working days lost, and on a same-day consent cadence it is still six against one.

There is a fourth thing that is not a negative about nesting but about flatness mistaken for nesting, and it is the most important paragraph in this chapter.

A structure that declares no hierarchy still has one. Jo Freeman named the mechanism in 1972 and nothing since has improved on it: there is no such thing as a structureless group, and where structure is undeclared it forms along pre-existing friendship networks, is unaccountable because unacknowledged, and advantages exactly the people who arrived with standing. Give it a number. A team of twelve has 66 possible ties. Nobody sustains 66 relationships; the sustained close-tie count is about five per person, which realises 30 ties — 45.5% of the mesh. The other 54.5% still has to carry information, and it carries it through whoever already had the ties.

The measured record bears this out. Zappos adopted Holacracy from 2013; in March 2015 it offered severance to anyone unwilling to commit, and 210 people — of roughly 1,500, so 14.0% — took it. Total turnover that year was about 30% against a historical 20%: an excess of about 150 people. Zappos retired the Holacracy label in 2020. Medium abandoned it in 2016 at around fifty people, and Andy Doyle's account names the mechanism precisely — codifying responsibilities in explicit detail hindered action, and coordinating across circles cost more than the circles saved. Valve's celebrated flat structure was described by Puranam and Håkonsson as carrying exactly the informal hierarchy Freeman predicted.

On whether exit falls hardest on people without prior standing: that is the mechanism, and it is not a published figure. Zappos has never broken its attrition out by tenure, level or demographic, and no honest number can be offered here. What can be offered is the adjacent evidence, which is strong: Castilla and Benard found that organisations which describe themselves as meritocratic show more bias in reward allocation, not less, because a declared absence of structure removes the thing that makes discretion checkable. An undeclared hierarchy is a hierarchy with the audit trail deleted.

So what must a structure add back to keep the benefit without the cost? Three things, all of them cheap, and every one of them already exists in a working system. A declared exception clause — Holacracy's own constitution carries one in Article 4.1, permitting any partner to break the rules in an emergency with a duty to communicate and restore afterwards. A named integrative role without line authority — Buurtzorg's coach, one per fifteen to twenty teams, who may not decide anything. And a written contract between peers — the CLOU, which converts an informal tie into a record that a third party can read. Freeman's own remedy, forty years earlier, was the same: delegation by explicit agreement, rotation, and information diffused by rule rather than by friendship.

DREAM

What becomes ordinary

In an organisation built this way, the first thing a new person is given is not an org chart. It is a domain — a written statement of what this unit decides without asking, what it must consult on, and what it may not decide at all. It fits on one side of a page and it is the most-read document in the building, because it is the one that answers the question everybody actually has.

The escalation rule is the same for everyone and it is written down: a decision rises when it cannot be contained, never because someone senior wants to see it. When something does rise, the unit says why in one line — this touches three other units and we could not close it — and that line is the whole justification required. Nobody rehearses a case for a decision they were already entitled to take.

The layer above holds two things and nothing else: what genuinely spans units, and the standard. It does not hold approval, because approval is what a containment rule replaces. A person at that layer spends their week on the interfaces — the places where two domains touch and neither owns the seam — and their performance is read off how few seams are in dispute, not how many decisions crossed their desk.

Team size is watched the way a bank watches a covenant. Twelve is the number; fourteen is a conversation; eighteen is a split, and the split is scheduled rather than argued about, because everybody has seen the quadratic and nobody thinks it is a personal judgement about them.

And the emergency clause is rehearsed. Twice a year, somebody declares one that is not real, a pre-named commander takes command inside the hour, the structure does what it is told, and then it reverts automatically at a stated time without anybody having to ask for their authority back. The rehearsal is what makes the reversion credible, and the credible reversion is what makes people willing to hand over command in the first place.

None of this requires anyone to be unusually good. It requires four documents that most organisations do not currently have: a domain statement, an escalation rule, a peer contract and an exception clause.

DESIGN

The structure that gets there

One. Write the domains before you remove anything. A unit's domain names three lists: decide alone, consult first, may not decide. The third list is the one that makes the first two safe, and it is the list that flat structures omit — which is why their members end up asking permission informally, from whoever has standing, which is Freeman's tyranny arriving by the back door.

Two. Set the escalation rule as a reason, not a threshold. A monetary threshold is the wrong instrument, because it asks about size rather than about span. The rule that works is: escalate what you cannot contain. A unit that sends something up states which other domains it touches. This has a useful property — it is auditable. You can count, at the end of a quarter, how many escalations named a real cross-domain interaction and how many named a preference. That count is your containment rate, and it is the number this chapter exists to give you.

Three. Cap the holon at twelve and mean it. The mesh cost is quadratic; the cap is not a cultural preference. Write the split rule in advance — at what headcount, on what axis, by whose decision — so that a growing team splits on schedule instead of negotiating its way to nineteen.

Four. Replace the manager with a coach at a ratio, and remove line authority explicitly. One coach per fifteen to twenty teams, holding no decision rights whatsoever. The removal must be written, because an integrative role that retains even implicit authority collapses back into a manager within two quarters and the containment rate falls with it.

Five. Put a peer contract on every seam. A CLOU, or something with its grammar: what I will do, with what metrics, agreed with the ten people whose work touches mine. Thirty activity areas and fifty metrics is the observed shape; half of that is enough to start. The contract does the work that a reporting line used to do, and it does it in a form a third party can read.

Six. Write the exception clause on day one, not after the first emergency. Name who may declare, who commands, what the time box is, and what reverts automatically. Rehearse it twice a year against a scenario nobody believes.

Seven. Measure containment quarterly and compare it against your own c. Compute c for your span and depth — the arithmetic is in this chapter and in lib/verify.py — then count your escalations. Two numbers, one comparison, on one page.

DESTINY

How it holds when nobody is pushing

Three forces pull a nested structure back into a hierarchy, and each has a known counter.

The coach re-acquires authority. It happens gradually and always for a good reason — a team is struggling, the coach steps in, the intervention works, and now the team consults before deciding. The counter is structural: the coach has no budget, signs nothing, and is evaluated on the containment rate of the teams they serve, which falls every time they decide something.

The teams grow. Splitting is emotionally expensive and the quadratic is invisible, so teams drift to sixteen and eighteen and the mesh cost rises silently past the point where a hub would have been cheaper. The counter is the pre-written split rule and a standing report of team sizes.

The emergency clause is never rescinded. This is the serious one. A structure that takes command in a crisis and does not hand it back has become a hierarchy that once had a good reason. The counter is automatic reversion by date, written into the clause, requiring a positive act to extend rather than a positive act to end.

And here is where the whole thing genuinely fails, stated without softening. It fails where the work is not nearly decomposable. Simon's condition is not a preference; it is the load-bearing assumption. Where almost every decision genuinely touches almost every unit — a single continuous process, a tightly coupled engineering programme, a live trading book — containment cannot be raised by structure because the interdependence is in the work itself. Then c sits below c*, nesting costs more than it saves, and the honest answer is a hierarchy with a wide span. Notice what the comparison at other spans says: widening the span from five to seven takes the hierarchy's cost from £130.00m to £123.91m. Widening the span is the hierarchy's own best move — and a wide span is only safe when units are self-contained, which is nesting under another name. The two structures are converging on the same insight from opposite directions.

It also fails where the work is contained but the reward is not. If people are paid on an aggregate they cannot influence, no domain statement will make them behave as though they hold their own. That is Volume V, and it is a real dependency, not a caveat.

DELIGHT

What it feels like

The pleasure is specific and most people who have felt it describe it the same way: the absence of the rehearsal. There is a particular expenditure of spirit that goes into preparing a case for something you already know is right, for a person who will approve it, and the recovery of that expenditure is not subtle. You get the afternoon back, and you get the part of yourself that was composing the argument in the shower.

Then a second one, quieter. In a nested structure you can see the whole of your own domain. Not the whole company — the whole of yours, its edges included, which is a size a human being can actually hold. Koestler's word for this is the Janus face: you are looking inward at something complete, and outward at something that contains you, and both views are in focus at once. People who work this way report a kind of restfulness that has nothing to do with the workload, and it is this: they are not carrying any part of the organisation they cannot see the edge of.

And the best hour in the year is the emergency rehearsal, which is unexpectedly joyful. Command is handed over in ten minutes, the structure moves like one thing, the clock runs out, and command comes back without anyone having to ask. A group that has practised giving up its autonomy and getting it back trusts its own autonomy in a way that a group that has never tested it simply cannot.

OPERATIONALIZE THIS

At the level of finance

Nesting becomes real at the moment a unit's autonomy is written into an instrument that a finance function recognises. Here is that instrument.

The structure: a Nested Authority Schedule with a microenterprise contract and an internal capital facility.

Three documents and one account. It is the Haier microenterprise agreement and the Morning Star CLOU, assembled in the form a group treasurer will sign.

The mechanics.

The Authority Schedule. For each holon: the delegated authority limit A, the domains it decides alone, the domains requiring consultation, and the exclusions. Set A at the value below which 85% of that unit's decisions historically fall — readable from one quarter of purchase-order and decision-log data, and the single most useful hour of analysis in this chapter.
The microenterprise contract. The unit buys from and sells to other units at a stated transfer-pricing basis, holds a notional P&L, and participates in the gain above an agreed baseline — Haier's value-adjustment mechanism, and structurally the same instrument as the shared-savings facility in I.01.
The internal capital facility. Group treasury lends to the holon against its own notional P&L, at the group's internal rate. The unit's right to draw is its autonomy made financial, and it is the clause that stops the Authority Schedule being ceremonial.
The exception clause. Named declarer, named commander, a time box in working days, automatic reversion on a date, and extension only by positive act. Price it: the difference between 30 working days and 5 is what this clause is buying.
The coach. Established at one per fifteen to twenty holons, with no signing authority and no budget, evaluated on containment rather than on outcome.

The balance-sheet treatment. While holons are management-accounting constructs, this is segment reporting and internal transfer pricing — no consolidation question arises, and the transfer-pricing basis is the only item your auditors will want to see documented. The moment a holon becomes a separate legal entity and takes outside capital, as Haier's microenterprises do, the control tests under IFRS 10 become live and the answer is no longer obvious. That is a genuine fork and it should be walked into deliberately, with advice, rather than discovered.

The counterparty. Internal first: group treasury to a single business unit, one schedule, one quarter. The external version — a lender underwriting a microenterprise on its own notional accounts — is available only once you have two completed internal cycles and a measured containment rate to show.

The number that decides it. One comparison, on the front page:

      measured containment rate  c        >        break-even  c*

  c   escalations that named a real cross-domain interaction,
      as a share of all decisions logged in the quarter

  c*  solve  L(c*, d) = ( hierarchy cost − nested fixed cost )
                        / ( decisions x cost per escalation )  + 1

For a 10,000-person organisation at span 5 and depth 6, c is 37.8%. Your own depth is ceil(log_span(teams)) + 1 and your own c follows from it in one line of arithmetic. If c clears c*, nesting is not a cultural preference — it is the cheaper structure, and it should be presented in those words.

The first ninety days on a page.

Day	Action	Artifact
1–15	Count the levels; compute depth and your own `c*`	One page of arithmetic
16–30	Log every escalation in one unit; classify by reason	The escalation log
31–45	Measure `c`; set `A` at the 85th percentile of decisions	The Authority Schedule
46–60	Write the domains: decide · consult · excluded	Three lists per holon
61–75	Write the peer contracts on every seam; write the exception clause	CLOUs and the clause
76–90	Rehearse the exception once; publish `c` against `c*`	The one comparison

APPRECIATIVE QUESTIONS

Twelve, for a room

Discovery — what is already working

Think of a decision this year that was taken entirely inside one team and turned out well. What made it possible for that team to close it without asking, and who else here has that same freedom without knowing it?
Which unit of ours already behaves like a whole business — knows its own numbers, holds its own edges? What did it have to be given to get there?
When has someone here broken the rules in a genuine emergency and been right? What did the organisation do afterwards, and what did that teach everyone watching?

Dream — what becomes possible

If every team here knew exactly what it decided alone, what would stop happening on a Thursday afternoon that happens now?
Imagine our escalation log a year from now and it is half the length. What is in the half that remains, and why does it belong there?
If our most junior person could read one page and know precisely what they are entitled to decide, what would they do first?

Design — what we build

What are the three things this unit should decide alone, and what is the one thing it should never decide? Write all four before anyone comments.
Where do two of our domains touch with nobody owning the seam — and who are the ten people who would need to be on that contract?
What is our split rule going to be, and who gets to invoke it without asking?

Destiny — how it holds

What would have to be true for a coach here to still hold no decision rights in three years' time?
What is the first sign we would see that an emergency clause had quietly become permanent, and who in this room would notice it first?
If somebody arrived here tomorrow with no prior standing and no friendships, what would tell them how a decision actually gets made — and is that thing written down?

WORKS CITED

Bernstein, E., Bunch, J., Canner, N. and Lee, M. (2016). "Beyond the Holacracy Hype." Harvard Business Review, July–August, 38–49.

Castilla, E. J. (2008). "Gender, Race, and Meritocracy in Organizational Careers." American Journal of Sociology, 113(6), 1479–1526.

Castilla, E. J. and Benard, S. (2010). "The Paradox of Meritocracy in Organizations." Administrative Science Quarterly, 55(4), 543–576.

Cox, M., Arnold, G. and Villamayor-Tomás, S. (2010). "A Review of Design Principles for Community-based Natural Resource Management." Ecology and Society, 15(4), 38.

Doyle, A. (2016). "Management and Organization at Medium." Medium, 4 March.

Dunbar, R. I. M. (1993). "Coevolution of Neocortical Size, Group Size and Language in Humans." Behavioral and Brain Sciences, 16(4), 681–694.

Federal Emergency Management Agency (2017). National Incident Management System, 3rd edn. U.S. Department of Homeland Security.

Fischer, B., Lago, U. and Liu, F. (2013). Reinventing Giants: How Chinese Global Competitor Haier Has Changed the Way Big Companies Transform. Jossey-Bass.

Foss, N. J. and Klein, P. G. (2022). Why Managers Matter: The Perils of the Bossless Company. PublicAffairs.

Freeman, J. (1972–73). "The Tyranny of Structurelessness." Berkeley Journal of Sociology, 17, 151–164.

Frynas, J. G., Mol, M. J. and Mellahi, K. (2018). "Management Innovation Made in China: Haier's Rendanheyi." California Management Review, 61(1), 71–93.

Gray, B. H., Sarnak, D. O. and Burgers, J. S. (2015). Home Care by Self-Governing Nursing Teams: The Netherlands' Buurtzorg Model. The Commonwealth Fund.

Hamel, G. (2011). "First, Let's Fire All the Managers." Harvard Business Review, December.

Hamel, G. and Zanini, M. (2020). Humanocracy: Creating Organizations as Amazing as the People Inside Them. Harvard Business Review Press.

HolacracyOne (2015). Holacracy Constitution, v4.1; (2019) v5.0.

KPMG (2015). The Added Value of Buurtzorg Relative to Other Providers of Home Care. KPMG Plexus, Netherlands.

Koestler, A. (1967). The Ghost in the Machine. Hutchinson.

Koestler, A. and Smythies, J. R. (eds) (1969). Beyond Reductionism: New Perspectives in the Life Sciences. Hutchinson.

Laloux, F. (2014). Reinventing Organizations. Nelson Parker.

Lee, M. Y. and Edmondson, A. C. (2017). "Self-Managing Organizations: Exploring the Limits of Less-Hierarchical Organizing." Research in Organizational Behavior, 37, 35–58.

Michels, R. (1915). Political Parties: A Sociological Study of the Oligarchical Tendencies of Modern Democracy. Hearst's International Library.

Mintzberg, H. (1979). The Structuring of Organizations. Prentice-Hall.

Nandram, S. S. (2015). Organizational Innovation by Integrating Simplification: Learning from Buurtzorg Nederland. Springer.

Ostrom, E. (1990). Governing the Commons: The Evolution of Institutions for Collective Action. Cambridge University Press.

Ostrom, E. (2010). "Beyond Markets and States: Polycentric Governance of Complex Economic Systems." American Economic Review, 100(3), 641–672.

Puranam, P. and Håkonsson, D. D. (2015). "Valve's Way." Journal of Organization Design, 4(2), 2–4.

Robertson, B. J. (2015). Holacracy: The New Management System for a Rapidly Changing World. Henry Holt.

Simon, H. A. (1962). "The Architecture of Complexity." Proceedings of the American Philosophical Society, 106(6), 467–482.

Weick, K. E. (1993). "The Collapse of Sensemaking in Organizations: The Mann Gulch Disaster." Administrative Science Quarterly, 38(4), 628–652.

Weick, K. E. and Sutcliffe, K. M. (2015). Managing the Unexpected: Sustained Performance in a Complex World, 3rd edn. Jossey-Bass.

Williamson, O. E. (1967). "Hierarchical Control and Optimum Firm Size." Journal of Political Economy, 75(2), 123–138.

Zhou, W.-X., Sornette, D., Hill, R. A. and Dunbar, R. I. M. (2005). "Discrete Hierarchical Organization of Social Group Sizes." Proceedings of the Royal Society B, 272, 439–444.

Note on figures. Every figure above is computed in lib/verify/VI_03.py and reproducible with python3 lib/verify.py VI.03. Figures marked there as REPORTED are published by the named source and are not derived here; figures marked ASSUMED are this chapter's modelling choices and carry their sensitivity bands. Haier's unit economics are not independently verifiable: no microenterprise-level accounts are published, and the only Haier figure computed here is the arithmetic of a removed headcount.