A walled, self-perpetuating advantage can assemble itself out of individually rational decisions made on true information, with no agreement, no policy and no conspiracy. Which means there is no defendant to sue, no statute to repeal, and no one who did it.
Links: Positional Goods (the payoff structure that decides valence), Market Efficiency and Human Limits, Aggregation vs. Sorting (this is sorting that turns into aggregation), The Slack in the Law (permission converting to requirement with no text change), Inequality Optimum, Constructed, Not Arbitrary, The Gödel Governance Problem, Word War final — genetic engineering
The page was first written as “The Citadel Effect” and renamed the same day. Chris flagged the mash himself: “I found the term ‘citadel’ from Antoine de Saint-Exupéry, but there is also a more modern version called ‘cathedral’.” He is right that these are different, and one of them is not this concept at all:
| Term | Source | What it actually names |
|---|---|---|
| Citadelle | Saint-Exupéry, posthumous, 1948 (The Wisdom of the Sands) | ❌ A different idea entirely — that structure, constraint and ritual are what make life meaningful; the citadel is built in men’s hearts. Nothing to do with emergent coordination. Related instead to constraint-gives-meaning themes |
| “The Cathedral” ① | Curtis Yarvin / neoreaction | ✅ This mechanism, in the ideological domain — universities, press and civil service converging on one line without conspiracy, because incentives and information are shared |
| ⚠ “The Cathedral” ② | Eric S. Raymond, The Cathedral and the Bazaar (1997) | ❌ The opposite — a small core team, controlled releases, explicit central coordination (his examples: GNU Emacs, GCC) |
| ⭐ “The Bazaar” | Raymond, same essay | ✅ This mechanism, in the production domain — thousands of uncoordinated contributors, release early and often, and a coherent Linux kernel comes out |
| Tacit collusion | economics / antitrust | ✅ This mechanism, in the pricing domain |
⚠⚠ Note what rows ① and ② do to each other — Chris found this: “I also came across the term bazaar… the cathedral was looked at as top-down, where bazaar was more bottom up.” There are two Cathedrals in circulation and they sit at opposite ends of the very axis this page is about. Raymond’s cathedral has a coordinator; Yarvin’s Cathedral has none — that is its whole claim. ⭐ So “cathedral” cannot be used unglossed: the word does not tell you whether central control is present or absent, which is the only thing anyone needs to know. A second reason the descriptive name is the right one.
⚠ One correction to Chris’s gloss, since it matters for the axis: Raymond’s cathedral is not “top-down without explicit control mechanisms” — it is top-down with them, a restricted core team and gated releases. The bazaar is the one with no explicit control, which is why it belongs on this page and the cathedral does not.
⭐⭐ And the most interesting thing here: Yarvin’s Cathedral, Raymond’s bazaar and antitrust’s tacit collusion are the same formal object, discovered independently in three literatures that never cite each other. One was built to explain oligopoly pricing, one ideological convergence, one open-source software — and all three describe correlated incentives plus shared information producing a coordinated outcome with no coordinator.
⚠ Using the Cathedral’s mechanism does not mean adopting its politics, and the two are separable. The descriptive claim — that institutions with aligned incentives converge without agreeing — is analytically ordinary and applies to any correlated-incentive bloc regardless of its direction: industry-funded research, professional-association orthodoxy, trade-press consensus, a donor class, a regulatory bar. The normative and historical claims bolted onto the term are a separate argument and are not imported here. ⭐ And the defector test below applies to the Cathedral exactly as it applies to a cartel, which is what converts it from an unfalsifiable vibe into a checkable claim — in either direction.
So the recommended usage: “coordination without a coordinator” for the mechanism, “tacit collusion” when the domain is prices, “the Cathedral” when the domain is ideology and you are willing to carry the term’s politics. Drop citadel — it is Saint-Exupéry’s and it means something else.
Chris referred back to “the citadel effect (tacit collusion) you describe” on 2026-09-11 — and a search of the whole vault found no page, note or memory containing either term. The concept had been discussed and never written down, which is the rot pattern the OS manual records for positional goods. It surfaced only because a citation failed — and the failed citation turned out to be a terminology failure as well as a storage one, which is its own small lesson: a concept with no page also has no settled name, and the two decay together.
Tacit collusion is the narrow, load-bearing case. Firms in a concentrated market can reach a supra-competitive equilibrium with no agreement at all — each observes the others’ incentives, anticipates the response to undercutting, and independently declines to cut. No meeting, no signal, no contract. The outcome looks like a cartel and is produced entirely by unilateral rational choice.
⭐ Antitrust is built around this gap and has never closed it. Liability generally requires an agreement; parallel conduct on its own is not enough, however suggestive the pattern. The law needs a defendant, and this has none.
The general form builds a wall. Take that mechanism and drop it into a positional good, and it builds a wall:
| Requirement | Why it matters |
|---|---|
| Individually rational decisions | every actor’s behaviour is defensible in isolation |
| Shared, accurate information | nobody has to lie; the signal is true |
| Correlated payoffs | everyone’s best move points the same way |
| A positional good | the advantage is relative, so gains for insiders are losses for outsiders by construction |
The result is an advantaged enclave that nobody built, nobody agreed to, and nobody can be held responsible for — which is also why it is very hard to dismantle.
A framework that defines coercion as state action has no vocabulary for this. Every actor in a citadel is exercising liberty: choosing customers, pricing risk, hiring, moving house. Nobody’s rights were violated at any step, and the aggregate is still a wall.
You cannot shrink your way out of an actuary.
⚠ This is not an argument that markets are bad — it is an argument that the state/non-state boundary is the wrong place to look for coercion. The mirror failure exists on the other side: a framework that sees only private power misses that most such structures are held open by licensure, zoning and credential requirements that are state action. Both errors are the same error — attributing a distributed outcome to whichever agent the framework already dislikes.
The page as first written had a bug: every specimen was malign. Raymond’s bazaar fixes it, because it is the same mechanism producing something excellent. Nobody directs Linux; thousands of contributors act on their own incentives with shared information, and a coherent kernel comes out. That is coordination without a coordinator, and it is the best-documented success case there is — alongside the one Hayek spent a career on, the price system, which is this mechanism’s founding example.
⭐⭐ So the mechanism explains nothing about the outcome. What determines valence is the payoff structure of the thing being coordinated on.
| Emergent order | Good it coordinates on | Sum | Result |
|---|---|---|---|
| Price system | allocation of scarce resources | positive | ✅ a market |
| The bazaar | code — non-rival, copyable at zero cost | positive | ✅ a kernel |
| Tacit collusion | price, against buyers | transfer | ❌ a cartel |
| Genetic / credential stratification | a positional good | ⭐ zero by construction | ❌ a wall |
Read down the third column and the page’s thesis falls out: distributed choice sorts when the good is positive-sum and walls when the good is positional. Which is aggregation vs. sorting again — and it says the sort/aggregate distinction is a property of the good and the dispersion of preferences over it, not of the institution doing the choosing. ⚠ That is a correction owed to that page, and it is now load-bearing here.
“We often see companies start ‘borrowing’ policies when it turns out another company gets away with it.”
This is a sharper specification of the signal than the pricing literature usually gives, and it deserves its own line. In textbook tacit collusion the observed variable is a price. Here it is an enforcement outcome: firm A adopts a legally or reputationally risky practice, nothing happens to them, and B, C and D copy it. No communication, no agreement, not even a shared price — just a public experiment whose result everyone can read.
⭐⭐⭐ Which wires this page directly into The Slack in the Law, and it is not a decorative link. That page distinguishes designed slack (the jury), accidental slack (enforcement cost) and undesignated slack (nobody was appointed to hold it). This is what undesignated slack does when left alone: it gets discovered, published by example, and copied.
An unenforced rule does not stay quietly unenforced. It becomes an industry norm — because the first firm to test it runs the experiment for everyone else for free.
Specimens to check, all of which spread by imitation rather than agreement: non-compete and forced-arbitration clauses, gig-economy worker classification, dark patterns, data-retention practice, unpaid internships, and algorithmic personalised pricing.
⚠ And it strengthens the who-holds-the-slack criterion beyond what that page claimed. The argument there was that undesignated slack is illegitimate because nobody was appointed to hold it. The stronger argument is that it does not stay put: discretionary non-enforcement is read by everyone with an incentive to read it, and converts into collective practice — so the choice is never between enforce and quietly not enforce. It is between enforce and publish a loophole.
This pattern is dangerously easy to invoke, because it explains any durable inequality without requiring evidence of anything. Three honest objections:
⭐ The discriminating test: would a defector profit, and does anyone defect? A genuine citadel is one where defection is available and nobody takes it — which implies the equilibrium is being held by anticipated retaliation or by a shared signal rather than by underlying costs. If defection is unprofitable, there is no collusion to find; there is just a cost structure. Anything asserted as coordination-without-a-coordinator without this test is a vibe.
The Word War final is the reason this page got written, and it supplies the sharpest version of the mechanism. Correlated parental preference is an input to exactly this: distributed choice preserves diversity only when preferences differ, and for traits with a perceived universal ranking — height, cognition, disease resistance — they do not. Sex-selective abortion already ran that experiment and skewed national sex ratios from purely private choices, with no mandate anywhere.
⭐ Which is why “the parents choose, so nobody decides” is only half an answer. Nobody deciding is compatible with everybody choosing the same thing — and that is exactly the configuration that builds walls.