DecisionOS — Build · module

Architecture

What has already been decided about how AI runs here, which of those decisions nobody actually made, and which will be expensive to reverse.

When you reach for it

Before a platform commitment. When a second vendor is about to be introduced. When an enterprise architect has been asked for a target state and knows that half of it has already been decided by whoever moved first.

What it asks you

13 blocks, in this order. Each one exists for a stated reason, and the run tells you that reason as you answer it.

01
Three facts the coherence checks read
Three of the six coherence rules compare an architecture decision against something established elsewhere. Where an Operating Model or Governance run exists, these answers belong to it and should be taken from there rather than re-formed here.
02
1. Model sourcing
The current de facto answer — what is actually true today, not the target state and not what was agreed in principle — and then three questions that are the same for all ten: was it decided or did it happen, who could change it, and what would changing it cost.
03
2. Deployment boundary
The current de facto answer — what is actually true today, not the target state and not what was agreed in principle — and then three questions that are the same for all ten: was it decided or did it happen, who could change it, and what would changing it cost.
04
3. Knowledge and retrieval
The current de facto answer — what is actually true today, not the target state and not what was agreed in principle — and then three questions that are the same for all ten: was it decided or did it happen, who could change it, and what would changing it cost.
05
4. Agent orchestration, and where state lives
The current de facto answer — what is actually true today, not the target state and not what was agreed in principle — and then three questions that are the same for all ten: was it decided or did it happen, who could change it, and what would changing it cost.
06
5. Identity for non-human actors
The current de facto answer — what is actually true today, not the target state and not what was agreed in principle — and then three questions that are the same for all ten: was it decided or did it happen, who could change it, and what would changing it cost.
07
6. Integration pattern
The current de facto answer — what is actually true today, not the target state and not what was agreed in principle — and then three questions that are the same for all ten: was it decided or did it happen, who could change it, and what would changing it cost.
08
7. Logging and evidence
The current de facto answer — what is actually true today, not the target state and not what was agreed in principle — and then three questions that are the same for all ten: was it decided or did it happen, who could change it, and what would changing it cost.
09
8. Evaluation placement
The current de facto answer — what is actually true today, not the target state and not what was agreed in principle — and then three questions that are the same for all ten: was it decided or did it happen, who could change it, and what would changing it cost.
10
9. Data egress and secrets control
The current de facto answer — what is actually true today, not the target state and not what was agreed in principle — and then three questions that are the same for all ten: was it decided or did it happen, who could change it, and what would changing it cost.
11
10. The escape hatch
The current de facto answer — what is actually true today, not the target state and not what was agreed in principle — and then three questions that are the same for all ten: was it decided or did it happen, who could change it, and what would changing it cost.
12
The constraints
Absolute requirements only — facts rather than preferences, the same distinction the Vendors module draws between a knockout and a weight. A constraint stated as a category cannot be checked against an architecture, so each one declared is then stated as it would appear in a contract.
13
Who answered
A decision register from one seat is one seat's view of a landscape that may vary by team. Recorded so the report can name the three decisions that usually sit somewhere else.

The method

How the result is produced

Ten decisions, the same ten for every organisation, each recorded with its current de facto answer and three fixed follow-ups: whether it was decided or simply happened, who could change it, and what changing it would cost in a year. The deliberateness audit is the cross of the first against the third — decisions nobody made that would take a re-platform or more to unmake. Six published pairwise rules then check the decisions against each other and against three facts that belong to Operating Model and Governance; each rule names the two things it reads. Finally the instrumentation check derives three architectural capabilities from decisions 3, 5, 7 and 10 — a record of what a system did, a route from an output back to its sources, and a mechanism that stops a running system — and where one is absent the corresponding Cognitive Coverage dimension is reported as unachievable rather than low. Nothing here is scored, averaged or levelled.

What this is not

This is a stated architectural position, not a neutral measurement. The ten decisions and the six coherence rules are ours, they are opinionated, and they are printed in full so that a reader who disagrees knows exactly which rule to disagree with.

Nothing was scanned, inspected, or connected to. No repository was read, no cloud account enumerated, no traffic observed. Every answer is what somebody told us.

This is not a maturity assessment and produces no level. Architecture is a set of decisions, and a level would obscure the only thing worth knowing — which of them nobody took.

A decision register built from one architect's answers reflects one architect's view of a landscape that may genuinely vary by team, and 'varies by team' is a legitimate answer rather than a failure to know.

Where it stops

Everything here is what you told us. Nothing was read, inspected, profiled or connected to.

There is no peer comparison behind any statement in this report, and none is implied. We will not show one until there is enough real data to state an n.

This run reflects the answers of whoever completed it. A second respondent answering the same questions frequently disagrees, and that disagreement is usually the finding.

The reversal costs are estimates by whoever answered, and the deliberateness audit turns on them entirely. They are the numbers in this report most worth a second opinion.

The six coherence rules are not exhaustive. An incompatibility that is not on the list is not reported, and its absence from this report is not evidence of its absence from the architecture.

Three of the checks read facts that belong to Operating Model and Governance. Where those runs exist the answers should come from them; asked here, they are one person's recollection of another module's finding.

The register records the current position at one moment. Its value comes from being compared against — a decision that was deliberate last year and has quietly become expensive is invisible without a previous run to compare with.

What it connects to

Answers established elsewhere are carried in with their age shown, so a stale answer reads as stale rather than being silently reused.

Reads saved results from

The rest of the Build line

Readiness

Given what we have decided to do, the one thing that stops us — and what becomes the next constraint after we fix it.

Open

Operating Model

Who does what in this workflow, who approves what, what it costs to supervise, and who is accountable when the agent is wrong.

Open

Data Foundation

Whether the data this initiative needs can be reached, is right enough, is permitted, and is fresh enough for the decision it feeds.

Open

Talent

Whether we can staff and supervise what we have decided to do — and whether we still will be able to when the people who can catch the machine's mistakes have moved on.

Open

See what it costs All twelve modules