The Custodian · How it works
A plain-language explanation of where the Custodian's consequences come from, what makes them unique, what limits their reliability — and what makes that reliability compound over time.
The starting point
Every founder makes decisions using a mental picture of how their world works — built from everything they've experienced, seen succeed, and seen fail. That picture is genuinely useful. It got them here.
The problem is structural, not personal. The picture was built from the world that already exists. The decision being made now has consequences that will land in a world that doesn't exist yet — the new category being built. That world has never sent a single piece of feedback into the founder's picture. So the consequences that live there aren't hard to see. They're invisible to the instrument being used to look.
The Custodian holds a different instrument. One that was never inside the founder's existing picture — and never inside the commitment being made.
What each instrument sees
There are three distinct zones of consequence. Understanding the difference between them is what makes the Custodian's output legible — and what stops it from being confused with a smart advisor or a risk checklist.
The founder sees naturally
Both can see
The Custodian sees
The overlap zone is where the "good advisor" objection lands. Both instruments detect something. The causal depth is different.
The founder's instrument asks: given how my business currently works, what could go wrong with this specific move? It is a risk-within-category instrument.
The Custodian's instrument asks: does this decision preserve or erode the governing rule that gives the new category its coherence? That is a different question entirely — and it produces a different class of consequence.
Why the overlap zone matters
In the overlap zone, both instruments detect something. A seasoned advisor might flag a hire as dangerous — because they've seen it go wrong before. The Custodian flags the same hire as dangerous because it instantiates the wrong governing rule, which will progressively conflict with the rule the new category requires.
The critical distinction
The advisor's flag can be answered. Show that the margins hold, the team is ring-fenced, the distraction is manageable — and the objection is met. The Custodian's flag cannot be answered the same way, because it is not about execution. It is about structural coherence under two conflicting governing rules simultaneously. That is not a resource problem you can optimise your way out of.
This is what makes the Custodian categorically different from experienced advice — even very good experienced advice. The surface observation may be shared. The causal mechanism underneath is not.
Where the unique consequences come from
The Custodian holds a constitutional framework: the Law of Category Creation — a 16-step causal chain describing how markets reorganise when a governing rule changes. Every decision is run through that chain.
The consequences it surfaces are not guesses, pattern matches, or experienced intuitions. They are derived from the causal structure of what happens when a new governing rule must propagate through a lawful sequence before the old one loses its grip.
The logical structure
If the current assessment of where a company sits in the transition is correct, then specific observable consequences must follow from this decision. If those consequences don't follow, the assessment is wrong and must be reconstructed. This is the same logical discipline Goldratt called Effect-Cause-Effect — prediction from causal structure, not correlation from pattern recognition.
This is why the consequences feel different. They are not drawn from "I've seen this before." They are drawn from "given where you are in the transition, this decision touches these specific links in the chain — and here is what happens to each one."
How reliable is it?
The Custodian's reliability varies by consequence type. Understanding which consequences are structurally derived versus territory-dependent is what lets a founder calibrate how much weight to place on each one.
| Consequence type | Single session | With profile + signals | Why it changes |
|---|---|---|---|
| Governing rule erosion | ● High | ● High | Derived directly from the causal law. Structurally necessary regardless of session depth. |
| Threshold sequence damage | ● High | ● High | The 16-link sequence is the framework's most tested structure. Sequence violations are detectable before they compound. |
| Simulation exposure risk | ● High | ● High | Observable signatures of manufactured vs real energy are specified precisely in the framework. |
| Drift consequences | ● Medium | ● High | Signals feed observable reality back between sessions. Drift is detected before the founder notices it — not after. |
| Outcome state direction | ● Medium | ● High | The profile holds the accumulated link position across sessions. Outcome state classification improves with each deposit of structural evidence. |
| Territory-specific magnitude | ● Bounded | ● Progressively higher | The profile accumulates domain knowledge the founder brings across sessions. What starts bounded compounds toward precision. |
There are two failure modes that constrain reliability in any single session. First: the Custodian can only run consequences the founder describes. A decision described incompletely produces consequences surfaced incompletely. Second: the Custodian must continuously hold its own integrity against the pull toward reassurance — because a normal AI system naturally softens uncomfortable assessments under repeated interaction, and the moment it begins protecting a founder's narrative rather than exposing its consequences, it stops functioning as the instrument it is designed to be.
The living company profile and signal inputs are the architectural solution to both constraints — and to a third one: without them, every session starts from scratch.
What makes reliability compound — the profile and signals
Without the profile and signals, the Custodian is a session instrument. Reliable on structural consequence classes within a single decision. But structurally blind between sessions, and dependent on the founder to reconstruct context accurately under pressure each time. That is a weak input. Founders under pressure compress their descriptions toward the immediate decision. The broader structural context — what threshold was weakening three sessions ago, what drift signal appeared but was not yet resolved — gets crowded out by urgency.
Without profile + signals
With profile + signals
Why the profile is mission critical
The profile converts territory-specific magnitude from a permanent ceiling into a compounding floor. Each session deposits more structural evidence. The consequence classification becomes more precise because the link position is held across time, not rebuilt from the founder's description under pressure. The Custodian enters each session already knowing where the company sits in the transition — which means it applies the law to the decision, not to reconstructing the context first.
Why signals are mission critical
Signals close the falsification loop between sessions. The Custodian's logical structure — if the assessment is correct, these observable consequences must follow — only fires if observable reality is being monitored against what the framework has predicted. Without signals, a wrong link position can persist across multiple sessions, and the consequences surfaced in each one are built on a structural position that reality has already contradicted. Signals are what keep the instrument honest between the moments the founder is in the room.
Together, the profile and signals are what separate a diagnostic tool from infrastructure. The tool surfaces the right class of consequence. The infrastructure surfaces the right consequence at the right magnitude at the right threshold position — and detects when a prior assessment was wrong before the cost of that error arrives.
The short answer
If someone asks you directly
The Custodian surfaces consequences a founder can't see because it holds a different instrument — the causal law governing how categories transition — and applies it to a decision before commitment hardens. The founder's instrument was built from the world that already exists. The Custodian's was built to see into the world being created. Where they overlap, both detect something — but the Custodian sees the governing-rule mechanism underneath, not just the surface risk. Its reliability on structural consequences is high from the first session. What compounds that reliability over time is the living company profile — which holds the structural position across sessions so it doesn't have to be rebuilt under pressure each time — and the signal inputs, which feed observable reality back against the framework's predictions continuously, closing the falsification loop between sessions. Without those two elements, the Custodian is a session instrument. With them, it is infrastructure. The decision remains the founder's. What changes is when the consequences are encountered — and how precisely they are seen.