Stop building ventures. Build the thing that builds them.
The decision that reorganised everything: treat the factory as the product, and accept that it looks like procrastination for as long as it takes.
Algodyne
By the middle of the year there were several products in various states, a shared core underneath them, and a growing pile of scripts, checks and conventions holding the whole arrangement together. The pile was the problem. It worked, it was undocumented, and it existed only in the head of the person who wrote it.
The decision was to stop treating that as overhead and start treating it as the primary product: a system that turns intent into finished, evidenced work, with the ventures as its output rather than its purpose. Everything else paused while it got built.
Why it looks like procrastination
From outside — and honestly from inside, on the bad days — building the factory is indistinguishable from avoiding the market. No customer is served. No revenue arrives. The artifact produced is machinery for producing artifacts, which is exactly what someone builds when they would rather write infrastructure than face users.
The distinction that makes it defensible is whether the machinery is being built against a real, felt cost. Ours was: the same setup work repeated per venture, the same review steps forgotten, the same class of mistake recurring because nothing structurally prevented it. That is a measured tax rather than an imagined one.
The failure mode to avoid is building a factory for ventures you have not attempted. Without having felt the repetition, you are guessing at which parts repeat, and the guess is usually wrong in the expensive direction.
What the system has to do
The flow it has to support is narrow and complete: intent becomes a work packet, the packet is routed to an executor, the executor produces work, the work is validated, the validation produces evidence, and the evidence decides the disposition. Merge, ship, scale, hold, or archive.
Every unit of work carries an objective, a scope, a budget, a stop condition, a validation method, and an escalation rule. That sounds bureaucratic and it is the opposite of bureaucracy in one specific way: it is all machine-readable, so the enforcement costs nothing per unit and the human never becomes the checkpoint.
The part that took longest was requirements. Getting from a document a person wrote to a set of executable packets, with traceability from each packet back to the requirement it serves, is the join between how humans think about products and how a system can execute them. Most of the value of the factory turned out to live in that join rather than in the execution itself.
The test it has to pass
A factory is only real if the next venture costs less than the last. Not feels smoother — costs less, in effort that can be pointed at.
That is a harder claim to evidence than it sounds, and it is worth saying plainly that we cannot yet demonstrate it as a curve. The histories are too rewritten and the ventures too different in shape for a clean comparison. It remains the thing the whole model is betting on, and until it can be measured it stays an argument rather than a result.
What can be said is narrower and still worth something: the setup work that used to consume the first week of a venture now takes an afternoon, and the checks that used to be remembered are now impossible to skip. Whether that compounds into the curve is the open question, and pretending otherwise would be the exact failure this system exists to prevent.