Three domains, one three-part shape, and three incompatible theories of what starts it.
In 1962, Tim Hart and Mike Levin at MIT wrote a LISP compiler in LISP.
Read that again, because it should not work. A compiler for language L has to be compiled by something, and if it is written in L, the only thing that can compile it is itself. Hart and Levin's move was to run the compiler's source inside an existing LISP interpreter. The interpreter was slow, but it could execute the compiler's code, and that code could then compile LISP properly. Once the compiler had compiled itself, the interpreter was no longer needed. The ladder got kicked away.
Four years earlier, the NELIAC team under Harry Huskey had done the same thing by a different route: hand-code a stripped-down compiler in assembly, use it to compile the real one, then have the real one recompile itself and retire the hand-written stub. Both stories end the same way. A scaffold exists, does one job, and disappears.
Every language you use is standing on one of those chains right now. C, Go, Rust, Haskell, OCaml, TypeScript, Zig: each has a lineage of compiled compilers running back to some hand-written seed that no longer exists in any running system. You cannot rebuild the chain from source alone, because the source assumes the chain.
This is the bootstrap problem, and it has a fixed shape wherever it appears. A system needs a component that only the working system can produce. The resolution is always some asymmetric seed, some external scaffold, that breaks the circle long enough for the loop to close on itself. And then, almost always, the scaffold is forgotten.
I want to take that shape to three places it does not obviously belong: the origin of agriculture, the origin of self-replicating chemistry, and the launch of a two-sided marketplace. The three-part structure is the same in all of them. What is not the same, and this turns out to be the interesting part, is what each domain thinks the first term actually is.
One honest limit before we start. What follows is a pattern in descriptions, not a claim about shared mechanism. Neolithic farmers, RNA fragments and marketplace founders are not doing the same thing at different scales, and nothing here should be read as saying so. The shape is worth having because it tells you where to look, and because the three domains disagree in a way that is genuinely informative.
Seed asymmetry. Something must be unequal at the start. A hand-written compiler stub, a species that happens to be domesticable, a molecule that catalyzes a reaction its neighbors cannot.
Loop closure. The seed only matters if a cycle forms in which each part is produced by another part inside the system. Before closure you have a pile of ingredients. After closure you have a thing that makes itself.
Ratchet. The closed loop has to hold its gains, so that the next round starts from the last round's output rather than from zero.
Now the three legs.
The famous version of civilizational bootstrap is Jared Diamond's, and it is a story about luck. Some regions had grasses worth planting and animals worth herding, others did not, and the rest followed from the draw. It is a compelling story, it is contested by working archaeologists and anthropologists, and it is not the version I want to defend. Take it as the well-known framing being examined rather than as evidence.
The seed itself is concrete and well documented. In Southwest Asia, early Neolithic communities domesticated a small set of species that Daniel Zohary and Maria Hopf named the founder crops in 1988: three cereals (einkorn wheat, emmer wheat, barley), four pulses (lentil, pea, chickpea, bitter vetch), and flax, worked out somewhere in the window of roughly 10,500 to 7,500 years ago. Call it the canonical eight rather than the eight. By 2012 researchers were arguing the real set is larger, perhaps sixteen or seventeen species including figs, and by 2018 many scholars had turned against the founder-crop framing outright. The tidy number is a summary, not a census.
Here is the fact that reframes the leg. Agriculture did not happen once. It arose independently in something like eleven separate regions across the Old and New Worlds: the Near East, China, Southeast Asia, parts of Africa, Mesoamerica, South America, northeastern North America. Different species, different centuries, no contact between most of them.
Think about what independent repetition does to a luck story. A lottery ticket is drawn once. If the bootstrap were a matter of which region happened to hold the winning species, we would expect one origin and a long history of diffusion outward from it. Eleven independent origins point somewhere else entirely. They point at a threshold: conditions crossed a line in many places, at different times, with whatever species were locally available, and once crossed, the same loop closed each time.
That is not a small correction to the popular telling. It moves the civilizational leg out of the fate column and into the physics column, which is where the second leg has been living all along.
The sharpest version of the bootstrap question belongs to origin-of-life chemistry, because there the circle is at its tightest. Life needs replication, replication needs catalysis, catalysis needs molecules that replication produces.
In 2012, Nilesh Vaidya and colleagues published a result in Nature that is worth quoting exactly, because paraphrase softens it:
"Here we show that mixtures of RNA fragments that self-assemble into self-replicating ribozymes spontaneously form cooperative catalytic cycles and networks. We find that a specific three-membered network has highly cooperative growth dynamics. When such cooperative networks are competed directly against selfish autocatalytic cycles, the former grow faster, indicating an intrinsic ability of RNA populations to evolve greater complexity through cooperation."
Sit with the middle clause. They ran a race in a test tube between a closed cooperative loop and a strong solo self-replicator, and the loop won. This is not an analogy for loop closure. It is loop closure, at molecular scale, beating the alternative on growth rate under direct competition.
The paper also names why the solo replicator struggles, and the phrasing lands close to home for anyone who has tried to launch something alone: a lone self-replicator "would have faced the extreme challenge of possessing a mutation rate low enough both to sustain their own information and to compete successfully against molecular parasites." Do everything yourself, and defend against everything yourself. The cooperative network distributes that load across members who catalyze each other.
The seed term gets its most rigorous treatment in the same literature. Wim Hordijk and Mike Steel gave loop closure a testable definition in 2004 with what they call RAF sets, short for reflexively autocatalytic and food-generated: every reaction in the set is catalyzed by something in the set, and every molecule in it can be built up from a defined food source. That is closure with the hand-waving removed, and it is checkable by algorithm.
Then in 2011, Hordijk with Stuart Kauffman and Steel put the seed question in a form no other domain has managed. Their framing: the formation of a self-sustaining autocatalytic network is necessary but not sufficient for the origin of life, and the question of whether such a network could form by chance in a sufficiently rich soup of molecules and reactions is one you can investigate directly, by working out the levels of catalysis required. I am deliberately not quoting a specific critical value here. The point that matters is structural. In this domain, the seed asymmetry is not a story about who got lucky. It is a number, and you can compute it. Below the threshold, nothing closes. Above it, closure is expected.
(A note for regular readers, since the surname recurs: this is Kauffman on autocatalytic sets, not the NK fitness-landscape model that showed up in an earlier piece here. Different body of work, same author.)
Marketplace cold-start is the familiar member of this trio, and its job in this essay is to make the other two legible rather than to be re-explained. Andrew Chen's The Cold Start Problem gave the practice its vocabulary in 2021, and the atomic network idea has appeared twice already in essays on this site. I will not re-derive it.
What matters here is the seed term. The standard cold-start move is to subsidize the hard side: guarantee the drivers' income, seed the restaurants, pay for the first inventory, absorb a cost that no unit economics justify yet. You are manufacturing an asymmetry on purpose, and paying for it out of a budget line, until the loop closes and demand starts paying supply directly.
That is a completely different theory of the first term than either leg above. It is neither fate nor physics. It is a decision.
One acknowledgment, because this site has used a natural-science lens on marketplaces before. An earlier piece here mapped island biogeography, MacArthur and Wilson's equilibrium theory, onto marketplace dynamics. That theory is about persistence: how many species an island holds as immigration balances extinction. It assumes a mainland for species to arrive from. This essay is about the prior question, the one that theory presupposes. Where does the mainland come from? Bootstrap is origination, not equilibrium.
Line the three legs up and the shared structure is real but unremarkable. Line up what each domain believes about the seed, and something better appears.
Civilization, in its popular telling, treats the seed as fate: the draw happened before anyone acted, and history followed. Self-creating systems treat it as physics: a threshold condition that can be computed, satisfied anywhere it is met and nowhere it is not. Marketplaces treat it as a choice: an act of will with a price attached.
Same structural slot, three incompatible theories of what fills it. And the eleven independent agricultural origins are the hinge, because they are evidence against the fate reading of the leg that most famously asserts it. Read carefully, the civilizational leg sides with the chemistry against its own popularization. Repeated independent origins are exactly what a threshold predicts and exactly what a lottery does not.
That is a falsifiable claim, and it is more useful than the isomorphism it sits inside. If you find yourself explaining a bootstrap as luck, check for repetition. Independent recurrence is the signature of a threshold, and a threshold is something you can go looking for, measure, and eventually meet on purpose.
The third term is the one everyone assumes is automatic, and it is the one where the literature actually fights.
Joseph Henrich's 2004 paper in American Antiquity argues that after Tasmania was cut off from mainland Australia by rising seas, its population lost technologies it had previously possessed, over thousands of years. His proposed mechanism is not catastrophe and certainly not stupidity. It is demographic. Cumulative culture depends on a large enough pool of learners: each generation copies the most skilled practitioner it can find, imitation is lossy, and in a small enough population the losses outrun the gains. The ratchet turns backward.
That argument is disputed in print, in the same journal, and the dispute belongs in this essay rather than in a footnote. Dwight Read replied in 2006 that the Tasmanian toolkit was adequate to its environment rather than degraded, and Henrich answered in the same volume. I am not going to tell you who is right. I have not read the primary texts closely enough to adjudicate, and neither has anyone who is only reading summaries.
What the exchange establishes is more useful than a verdict. A ratchet is a claim about a mechanism, and mechanisms can slip. The compiler chain has the same property, sharply: break the lineage and you cannot rebuild the compiler from source, because the source assumes a working compiler. Ken Thompson's 1984 Turing Award lecture made the darker version of this point, showing that a compiler could carry a backdoor into every program it compiles including its own successors, invisible in the source forever after. Your ratchet holds gains you can no longer inspect. That is what holding gains actually means.
Three questions, in order, for anyone bootstrapping anything.
Which kind of seed do you have? Fate, physics, or choice. Most people implicitly assume fate, which is why they wait. If your bootstrap has recurred independently anywhere, treat it as a threshold and go find the number: what level of catalysis, what density of participants, what rate of return makes closure expected instead of lucky. If it genuinely is a choice, then it has a price, and the useful question is what you are willing to pay to manufacture the asymmetry rather than wait for one.
Is your loop actually closed? The RAF definition is a good discipline even outside chemistry. Every reaction catalyzed by something inside the set, every component buildable from what you already have. Draw your loop and check each arc for whether something in the system produces it, or whether the honest answer is that you are still supplying it from outside. A loop with one externally supplied arc is not closed. It is a scaffold you have not noticed you are still standing on.
What holds your gains, and can you still see it? The ratchet is a mechanism, not a law, and small learner pools lose skills. In a codebase, in a team, in a market, the question is the same: how many people know how to rebuild the thing, and what happens to that number under attrition.
Then the last one, which is the reason bootstrap stories are worth reading at all. Name your scaffold before you kick it away. The hand-written compiler stub, the subsidized first hundred sellers, the founder who did every job by hand for a year: these are the pieces that get discarded once the loop runs, and they are the pieces nobody documents, because by the time anyone thinks to write it down the system is self-sustaining and the scaffold looks like prehistory.
Hart and Levin's interpreter is remembered because someone wrote it down. Most scaffolds are not, and the loops they started are running anyway, holding gains that nobody alive can fully account for. That is not a failure of record-keeping. It is what a successful bootstrap looks like from the inside.
The ratchet question in this essay is a provenance question: what holds your gains, and can you still see it? Chain of Consciousness records the decision trail an agent actually followed, so the scaffold is written down before it gets kicked away.
pip install chain-of-consciousness
npm install chain-of-consciousness
If you would rather not run it yourself, the Hosted CoC page covers the managed version.