Essay · The architecture
The fourth system of record
The enterprise learned long ago that an asset cannot be left to the memory of whoever happens to hold it. It moved its capital, its customers and its code into systems of record. Its judgment is still in the room.
The modern enterprise scales by institutionalizing its memory. Three systems of record carry most of the weight. The ERP secures capital: every dollar has a ledger entry, and the entry outlives the bookkeeper. The CRM secures customer relationships: every account has a history, and the history outlives the salesperson. The repository secures code: every line has an author, a date and a reason, and all three outlive the engineer.
None of these existed at the start. Each was built after a company noticed the same thing. When the ledger lived in the bookkeeper’s head, the company could not grow past the bookkeeper. When the client relationship lived in the salesperson’s notebook, it left with the salesperson. The asset had to be taken out of the person and written into the infrastructure, where the next person could find it. That is the whole idea of a system of record, and it is why the enterprise has three of them.
The one that never made the journey
The most consequential thing an enterprise does is decide. How its leaders actually make strategic choices—what each person meant, what the leader decided, who accepted the resulting work and on what conditions—is the function everything else serves. And it has no system of record.
It has a system of ephemera instead: chat threads, call transcripts, meeting notes nobody reopens, and the memory of the people who were there. When a vice president leaves, the context behind three years of allocation leaves with them. Decision debt is the name for what that costs. This essay is about the architecture that would stop issuing it.
What makes a system of record
A system of record is not a place where things are written down. Email is a place where things are written down. Three properties set a system of record apart, and the three existing ones all have them.
It is authoritative. When two accounts of a customer disagree, the CRM wins; nobody argues from a notebook. It is versioned. A ledger entry is corrected by a new entry, never by erasing the old one, so the history of what was believed stays readable. And it is inherited. The systems downstream draw from it rather than reconstruct it: the invoice reads the ledger, the renewal reads the account, the build reads the commit. Nobody retypes the asset at each handoff.
Hold the enterprise’s decisions against those three tests. Authoritative: when two people remember a decision differently, there is nothing to consult. Versioned: when the decision changes, the old reasoning is overwritten, if it was ever written. Inherited: the tracker, the channel and the agent each carry their own reconstruction of what the meeting produced, and the reconstructions drift.
Why a decision log is not one
The obvious repair is a decision log, and many organizations keep something like it: a list of what was decided, by whom, on what date. It fails the test quietly, because it records the outcome and not the judgment. It is a corporate receipt. It tells the successor that the purchase was made and nothing about why, and so it cannot tell them the one thing they need to know: whether the precedent they are about to follow was structurally sound or merely lucky.
A system of record for decisions stores the receipt. A system of record for judgment stores what the receipt was for.
Judgment is more than the choice. It is the evidence that was uncovered and the evidence that was looked for and not found. It is the reservations people stated, in their own words, before the room formed a view. It is each person’s private deliberation, recorded before anyone else could shape it. It is the boundary conditions—what has to stay true for this to remain the right call—and the exact scope each owner accepted. Those are what turn a debate into a firm commitment, and they are what a successor needs in order to tell a sound precedent from a fortunate one.
The fourth system
Sikizana is the system of record for judgment: the fourth system, beside the three the enterprise already keeps. What it holds for each decision is its Record of Judgment, versioned. The record carries six things: the meaning people shared and the question they understood; the alternatives considered and why the others were not chosen; the choice, by whom, on which evidence and assumptions; the acceptance, in the exact scope each owner took on; the conditions that must hold; and the learning, when the outcome is read back against the original reasoning.
As the CRM took the customer relationship out of a salesperson’s private memory and made it enterprise equity, Sikizana takes the architecture of a choice out of the boardroom and makes the company’s judgment a kept, versioned and reusable asset. It records decisions, never people: what a person contributes is held under their own control and shared on their terms, and the record is of the judgment, not of the judged.
The three tests then hold for judgment as they hold for capital. The record is authoritative, because it is where the leader decided and where each owner accepted. It is versioned, because a change of conditions produces a new entry against the old basis rather than a rewrite. And it is inherited: by locking judgment at its source, the execution trackers, the communication channels and the AI agents the enterprise already runs can finally inherit a decision instead of inventing one.
What it is not
It is not a note-taker. A transcript is ephemera at higher resolution. It is not a replacement for the three systems it joins; it sits upstream of them, where the choices that fill them are made. And it is not the AI deciding. Sikizana’s AI does clerical work in the record and asks questions of a person’s own draft; it holds no position and casts no vote. The judgment that the fourth system keeps is human, which is the only kind worth keeping.
Before the next meeting
If the reasoning behind last quarter’s largest decision turned out to be wrong, which of your systems of record would be able to show it?