Most of what we build is meant to be dismantled

Every venture resolves to one of a small set of outcomes: build, scale, spin out, joint venture, license, sell, assimilate, archive. Most of them are not scale, and any system treating everything other than scale as failure will keep dead ventures alive to protect the narrative.

The cost of that is not the capital already spent. It is the capital spent on the next venture, which now starts from a portfolio that has learned nothing and is carrying a corpse. Sunk cost is the obvious trap and the smaller one; the expensive trap is that a portfolio unwilling to record a stop has also lost the ability to record a reason.

Stopping on evidence, not on boredom

The objection to a factory that shuts things down is that it produces serial quitters — teams that abandon hard problems at the first sign of difficulty and call it discipline. That failure is real and the distinction that prevents it is specific.

A stop is legitimate when a gate that was written down in advance was not met. It is not legitimate when the work got boring, when a more fashionable idea appeared, or when the thesis was quietly rewritten to make the current result look like a failure. The protection is that the bar existed before the evidence did. A venture that fails a gate nobody set is not being evaluated; it is being abandoned with paperwork.

The same rule cuts the other way and this is the half that gets forgotten. A venture that clears its gates keeps its capital even when the room has lost interest in it, because the alternative is a portfolio steered by enthusiasm.

Residue is the return

Archival counts as a result only if the work leaves something behind: a reusable technical primitive, a validated market signal, a sharper operating doctrine, a better execution workflow, a stronger evidence pattern, a distribution advantage, a trust asset, or a data asset.

That list is the actual return on a shut-down venture, and it is not small. A validated negative — this market does not have the urgency we assumed, and here is the evidence — is worth real money to the next three ventures that would otherwise have considered it.

So shutting something down is itself a process with deliverables rather than an absence of activity. What is generic gets extracted and hardened. What was learned about the market gets written where the next thesis will find it. What the build revealed about the factory itself gets fed back into the standard. The extraction happens before the lights go off, because nobody ever comes back for it afterwards — the moment a venture is archived, the people who understood it are already on something else.

The second-year effect

In year one this looks slower than the alternative, and it is. Generalising costs more than solving the immediate case, and the cost falls on the venture doing the generalising rather than the one that will benefit. A team measured on its own velocity will never choose it, which is why the choice cannot sit with the team.

The curve crosses somewhere around the third or fourth venture. Before the crossing, a factory looks strictly worse than a studio that ships fast and shares nothing. After it, each build starts further along than the last, and the gap widens continuously because the shelf keeps growing while the alternative keeps paying full price.

That crossing point is the investment case, and it is why the factory gets built before the products. It is also why the honest question to ask a venture factory is not how its current ventures are performing. It is whether the marginal venture is getting cheaper, and whether anyone is measuring it.

What this implies about portfolio size

A crossing point at the third or fourth venture sets a floor. A factory that will only ever run two ventures is paying the generalisation cost and will never collect on it — for that shape, a studio model is genuinely better, and saying so is more useful than defending the architecture.

It also sets a shape for how the shots are chosen. Enough of them have to be in domains where the primitives transfer, or the shelf fills with things only one venture can use. Diversifying across unrelated markets looks like risk management and is closer to running several small studios that share an office.

The compounding asset is the machine, not any product it emits. Every decision about which venture to start, hold, or archive is ultimately a decision about what the machine will be able to do next year.