Skip to content

What we believe,
and how to prove it wrong.

Versionv1.0

Written

Next review

One place to read what this house believes about building businesses whose operations are agent-run and whose ownership is on-chain. It is assembled from positions already stated in public, on this site and in the articles. Nothing here is new doctrine.

Read the ten ↓

What this is

A document that can lose

A thesis that cannot be scored is a brochure.

It is a numbered, versioned document on purpose. Each hypothesis below is written so that it could turn out to be wrong, and each carries the finding that would show it. Later versions mark every one right, wrong or still open, with the evidence that decided it. The review date above is when that happens next.

Two constraints on how to read it. First, none of these are results. They are the bets this practice is currently making, stated precisely enough to lose. Second, several are untestable with public data today, and that is noted rather than hidden.

The hypotheses

Ten, each written to lose

Every hypothesis carries one of three states. In v1.0 all ten are open: nothing here has been scored yet, and the header names the date it will be.

01
Still open

The intersection is one category, not two markets.

Entity design for a business whose operations are agent-run and whose ownership is on-chain is a distinct discipline, because the two changes interact: an agent acting without a signature and a membership deciding without a manager break accountability at different joints, and the repair for each depends on the other.

Wrong if

Buyers consistently and successfully purchase the two pieces of work from separate providers with no coordination cost, and their outcomes are no worse for it.

02
Still open

Reversibility, not model capability, is the operative constraint on delegation.

What a business can safely hand to a system is bounded by which of its actions can be undone, and that boundary moves far more slowly than model capability does.

Wrong if

Operations delegated without any reversibility classification show no higher incidence of unrecoverable loss than operations that drew the line first.

03
Still open

The reversibility line can be drawn in advance, by people, on real operations.

The rule is only useful if an operator can classify actions before the first agent runs.

Wrong if

Two independent, competent reviewers looking at the same operation cannot agree, better than chance, on which of its actions are reversible. If the classification is not reproducible, the rule is a slogan.

04
Still open

The order of operations is causal, not stylistic.

Prove the business, then decentralize ownership, then decentralize decisions one mandate at a time, with any token last. The claim is that ventures taking these steps out of order fail more often, and specifically that governance built before there is something to govern exhausts the people who were supposed to use it.

Wrong if

Ventures that decentralize governance before they have revenue survive and ship at comparable rates.

05
Still open

Ownership is a retention instrument, not a compensation instrument.

Real ownership, registered and readable, changes how long people stay in a way that pay and perks do not, because it changes what leaving costs them rather than what staying pays them.

Wrong if

Comparable people in comparable roles show no tenure difference between owning and not owning once pay is controlled for.

06
Still open

Autonomy is a property of a workflow, so a business sits at many levels at once.

The L0 to L4 ladder applies to operations, not to companies, and any single number attached to a whole business is an average that hides the thing worth knowing.

Wrong if

Operating businesses in practice cluster at one level across all their workflows, which would make the company-level number the more informative one.

07
Still open

A level can be assigned from an operating log rather than from a claim.

Each rung carries an observable test, and the point of the tests is that two independent readers of the same log should reach the same number.

Wrong if

Inter-reader agreement on real logs turns out to be poor, in which case the ladder is a vocabulary and not a measurement.

08
Still open

The L2 to L3 step is where operating models break, and where programmes stall.

Every other step on the ladder adds capability. This one moves a job, because approval stops happening action by action and becomes a written policy. The prediction is that stalls concentrate at this step rather than distributing evenly across the ladder.

Wrong if

Failures are spread evenly across the ladder, or concentrate elsewhere.

09
Still open

Participation is a relevance problem before it is a mechanism problem.

Members vote on decisions that touch them, so narrowing mandates and keeping votes close to the work should raise meaningful participation more than any voting mechanism does.

Wrong if

Matched organizations with narrow, work-adjacent mandates show no better participation than comparable organizations with broad ones.

This house has no participation data of its own and does not claim otherwise until it does.

10
Still open

The co-operative record is the best available prior for on-chain organizations.

Member qualification, one member one vote against capital weighting, a named dispute forum, and written exit and redemption terms are solved problems in co-operative practice, and on-chain organizations that adopt them should outlast those that rediscover them.

Wrong if

Organizations that ignore the co-operative playbook survive at the same rate, which would mean the technology changes the governance problem rather than the plumbing under it.

What would change our mind

Three findings

Three findings would force a rewrite, not an amendment.

Each is stated with the hypotheses it takes with it, so the consequence can be checked rather than taken on trust.

A long, independently verified record of businesses run at the top of the ladder without a reversibility discipline.

If durable, audited operations at L3 and L4 turn out to be common among firms that never classified their actions, hypotheses 2 and 3 fall together and the rule this practice is built around is revealed as an expensive habit rather than a constraint.

Would overturn

Evidence that governance built before revenue works.

If ventures that decentralize decisions first prove more durable, or merely equally durable, then hypothesis 4 is inverted and the order of operations was a preference dressed as a finding. The sequencing judgement is the thing this house sells, so this is the one that matters most.

Would overturn

Confirmation that on-chain ownership and agent-run operations do not reinforce each other.

If the two turn out to be independent, or worse, to add two governance overheads to the same firm without any interaction, hypothesis 1 goes and with it the case for treating this as one discipline. That would not make the individual pieces of work wrong. It would make the intersection a marketing position rather than a thesis, and the honest response would be to say so in v2.

Would overturn

Tell us where
this is wrong.

We use cookies to understand how the site is used and to improve it. You choose what we keep. See our Cookie policy.