The Experts Become Stewards

The Incumbent Wrote the Mandate and Never Encoded It said the Operating Inheritance crosses from the incumbent as rules, history and terms, and that the people cross as Stewards or not at all. The Parent Contributes Knowledge, Not Capital bounded the parent's role in the clone's governance. This memo is about the people who do cross, because they are where two claims the corpus has now made collide. The parent's experts are the best source of the inheritance — they are the ones who made the decisions the inheritance encodes. They are also the carriers of everything the inheritance excludes: the handoffs, the approvals, the status meetings, the instinct to ask before acting. The Coordination Tax does not live in the org chart. It lives in people's habits, and the people are the ones the clone is about to make Stewards.

The corpus has already said that a Steward placed over a workflow full of human dependencies is not a Steward but a manager with a more expensive toolset. This memo says the converse: a manager placed over a stack that has removed those dependencies does not become a Steward by appointment. They become one by a transition the clone has to design, and the design is the fourth thing the clone encodes, after the mandate, the thresholds and the surface.

Three positions, in order

Layered Stewardship prescribes one Steward at handover and domain Stewards added only when sustained intervention load in a specific area justifies them, measured by Escalation Rate against calibrated thresholds, never by convention. The clone applies the prescription with one amendment forced by where its people come from.

The founding Steward is from outside the parent by default. This is the amendment, and it is not a judgment about the parent's people. It is a consequence of the definition in The Incumbent Wrote the Mandate and Never Encoded It, and the test behind it is the one any candidate has to pass — can they govern inside the clone's mandate and thresholds without re-creating the parent's approval chain. An insider takes the role by passing that test on evidence, not by origin; the default is outside because the inheritance was extracted from people who could not see their own assumptions, and the test exists so that the exception is decided by the record rather than by the title. The inheritance is the parent's decisions without the parent's organisation, and a founding Steward who spent fifteen years inside that organisation carries it with them — not as disloyalty but as the thousand defaults about who is consulted, what is escalated, and how long a decision is allowed to take. The corpus's Bridge Operator is the profile: an entrepreneur whose pre-AI business-building experience supplies the integrative judgment to recognise a missing step or an unstated assumption in an AI-assembled business, distinct from deep specialism. The founding Steward's job in the first months is exactly that recognition, applied to an inheritance that was extracted from people who could not see their own assumptions because they had never operated without them.

Domain Stewards are drawn from the parent, when the system's data justifies them. This is where the experts enter, and the entry condition is the one #101 set: Escalation Rate in a domain persistently above target despite correctly calibrated thresholds. The parent's pricing lead becomes the pricing domain's Steward when pricing exceptions outrun the founding Steward's capacity, not when the clone reaches a size at which firms conventionally have a head of pricing. Their obligation is the one every Steward carries — resolve the exception, then update the agentic library so the next similar case is handled autonomously — and the role is designed to shrink its own necessity. A domain Steward from the parent who is still resolving the same class of exception after two quarters has not re-encoded it; they have re-created their old job with a better interface.

The builder's operator leaves on evidence, not on a date. The Parent Contributes Knowledge, Not Capital said the builder supplies the first Steward and takes a declining operating fee for the period it does. The handover from the builder's operator to the clone's own Stewards is the Stewardship Handover #101 describes, and the clone has to gate it on the same metrics that gate everything else: MTTI above threshold, Escalation Rate within band across domains, and a Steward in place whose intervention record shows them resolving by updating the system rather than by deciding each case. A builder that stays past that point is a dependency, and the corpus's whole position on Legacy Liability is that dependencies are what a clean sheet exists not to have.

What the Steward inherits, and on whose authority

The reach chapter and the form chapter were written about Stewards who, in the corpus's earlier memos, were assumed to have been present at design time. The clone's Stewards were not. They inherit three things they did not author, and the memo has to say what that means for each.

They inherit a mandate. The Encoded Mandate states what an agent may act on and on whose authority, and the clone's mandate is the parent's inheritance, encoded by the builder, on the authority of a licence the Steward may never have read. The Mandate Was a Sentence said a boundary held as instruction is one the agent can exceed or misread. A boundary held as architecture but authored by a third party is one the Steward can misunderstand, and a Steward who does not know why a class of action is refused will, under pressure, ask for it to be allowed. So the first thing the clone re-encodes is the mandate's provenance: every rule in it carries the exception history it was derived from, in the Operational Ledger, so that a Steward who asks why can be shown the ten cases in the parent's history that produced the rule.

They inherit a surface. The form chapter said the Steward's surface is a Fixed-Function Surface, specified at design time and changed only by committed redesign. A Steward who did not design it has to be trained to it rather than consulted on it, and the temptation to let the incoming expert reshape the surface to match the parent's dashboards is Layout Drift by another route. The surface is the clone's, not the parent's. The parent's expert learns where the state lives; the state does not move to where the expert is used to finding it.

They inherit a channel and a threshold. The Transcoding Layer routes command and verification; the Intervention Threshold says when the system asks. A Steward from the parent is used to being asked constantly, because the parent's workflow was built around asking. The threshold is calibrated against the inheritance's exception history, not against the Steward's comfort, and the Authorization Trap — the operator who cannot remove a gate the system has been verified to handle — is the specific failure a parent-trained Steward is most exposed to. The clone's defence is the one the Steward Experience entry prescribes: symmetric effort between escalation and de-escalation, so that loosening a threshold is as easy as tightening one.

Finding out who holds the rules

The licence the parent granted makes one thing possible that The Incumbent Wrote the Mandate and Never Encoded It could not describe: the builder can now sit with the people who hold the rules and replay the record they would be tested against. The diagnostic is simple in form, and one form of it shows the shape. Sit with a handful of the firm's longest-serving operators — five is an illustration, not a protocol; the number is whatever the firm's size and exception volume make meaningful — and ask each for the recent exceptions they resolved and the rule they applied. Where the rules agree across them, there is a candidate mandate to replay against the transaction record. Where each gives a different rule, the record shows which one the firm was in fact applying and with what outcomes, and the inheritance is the rule that survives plus the list of exception classes the firm never settled, which escalate from day one. The test is the agreement and the replay, not the count.

The same exercise is the first step of the formation path, because it sorts the people as well as the rules. The operator whose stated rule survives replay has been governing by a mandate without the architecture to hold it; that is a domain Steward in waiting. The operator whose rule does not survive, or who cannot state one, has been exercising judgment the clone will not encode, and they enter the clone, if they enter it, as a candidate for formation rather than as a holder of knowledge the clone needs. The rules the memo took from the people who could not see their own assumptions are the rules the replay has to test hardest; the people whose assumptions the replay exposes are the ones the formation path has most to do for.

The transition is a product

What this memo has described is not an HR process. It is a product the builder supplies alongside the architecture: a formation path that takes an expert from an incumbent and produces a Steward, with a curriculum drawn from the corpus — the Stewardship Model's labour split, the threshold discipline, the surface, the channel, the exception-resolution obligation — and an exit criterion drawn from the system's own metrics. The Case for the Studio said the Stewardship Model is a role a studio trains for and deploys consistently, and that a solo founder must discover it in isolation. The clone extends that to the parent's experts. They do not discover stewardship by being promoted into it. They are formed into it, and the formation is measured.

The Operator's Verdict

An incumbent's expert offered a Steward's role in a clone built from their own knowledge has one question to ask before accepting, and it is not about the title. It is whether the mandate they will govern carries its provenance — whether, when the system refuses something they would have approved, they can be shown why. A mandate without provenance is one the Steward will fight, and a Steward who fights the mandate reintroduces the judgment the architecture was built to encode.

An operator building the clone has the mirror obligation. The founding Steward is from outside by default, and an insider takes the role only by passing the same test on evidence; the domain Stewards are from the parent and are added on evidence; the builder leaves on evidence; and every rule in the mandate can answer the question on whose authority. None of this exists in an Arco venture. No expert has been formed into a Steward from an inheritance, and the memo describes the transition as the corpus's principles require it to be designed.

Technology changes what is possible. Who authored the mandate determines what the Steward has to re-encode first.