Why does the number of questions change when I change shape?

Because two of them ask about independent review, and an organization where nobody is independent of the people building and using AI has no honest answer to give. Those controls carry a weight of zero for that shape and drop out entirely, taking their weight out of the divisor as well, so a dropped control is not a silent failure. Every other control is asked at every shape.

Why are the basics weighted heavier for a small team?

Because there is nothing else there to carry them. In a larger organization an incomplete inventory is partly compensated for by procurement, a security function and a joiners process that all happen to know what is in use. In a company of six the list is the only thing that knows, so the same control is doing more of the work and the weight says so.

Where do the bands come from?

From a product RedHub already ships in this lane. The AI Risk Register & Treatment-Tracking System puts its top band at 75 and its middle band at 45, so this assessment uses the same two numbers. Choosing different thresholds by feel would have made a free assessment disagree with the paid tool it recommends. The names are not identical: that product says GOVERNED flat, because it grades the entries in your register. This one reads what you told it, so its top verdict is GOVERNED AS DESCRIBED.

Why can a gate only ever lower a verdict?

Because a weighted average is the wrong instrument for a structural failure. Nine strong controls can carry a score while the one thing everything else depends on is missing, and averaging is exactly what hides that. A gate is the place where one specific fact overrides the summary, and letting it work in the other direction would make it a bonus instead of a limit.

Why does one missing precondition fail the whole thing, when the security assessment needs a whole chain to be weak?

Because the two describe different structures. A chain of defenses only fails when the last one does, so three strong links and one weak one is a system with three things still standing. Governance is a sequence of preconditions, and a precondition is not compensated for by what got built on top of it. You cannot write a policy for AI you have not inventoried. One absence is enough, where in a chain it is not.

Can I check the score by hand?

Yes, and that is the point of publishing every weight on this page. Multiply each mark by its weight, divide by three, sum them, divide by the weight in scope and multiply by a hundred. The result is deterministic: the same answers always produce the same score, the same band and the same first thing to fix.

How this assessment works

The assessment marks up to 22 controls from 0 to 3 on what you could show today, weights them for the organization shape you picked, and renormalizes over what is in scope. The result is a score out of 100 and one of three verdicts. Two gates can hold a verdict down regardless of the score, and neither can lift one.

It is deterministic. The same answers always produce the same score, the same band and the same first thing to fix. There is no model in the scoring path, which is why the whole method fits on one page.

The arithmetic

Each control carries a weight for each organization shape, and the weights for a shape sum to exactly 100. You mark every in-scope control from 0 to 3 for what you could show someone today. The score is the weighted marks divided by the weight in scope, as a percentage, rounded to a whole number.

earned  = sum(mark / 3 * weight)   over in-scope controls
inScope = sum(weight)              over in-scope controls
score   = round(earned / inScope * 100)

Marking a control not applicable removes its weight from the divisor instead of scoring it zero, so declaring something out of scope never counts against you. Controls your shape has no surface for were never in the divisor at all: an organization with nobody independent of the people building and using AI is not graded on independent review, because grading it there would invent a finding.

The three preconditions are the exception. They cannot be marked not applicable, because nobody running an AI governance assessment can truthfully say that a list of the AI they use, a rule for using it, or a person accountable for it is out of scope. An assessment you can opt out of the gate on is a score with a back door rather than a verdict.

Bands are 75 for GOVERNED AS DESCRIBED and 45 for GAP, anything below that is UNGOVERNED. Those two numbers are not ours: they are the thresholds RedHub already ships in this lane, in the AI Risk Register & Treatment-Tracking System. Picking different ones by feel would have made this assessment disagree with the products it recommends.

Why two gates, and why they only ever hurt

A gate can lower a verdict and can never raise one. That is the whole mechanic, and it is the reason a verdict here means something a weighted average does not.

The foundation. Three of the 22 controls are preconditions, not contributions: a list of the AI in use, a written rule, and a named person accountable. If any one of them reads zero, the verdict is UNGOVERNED at any score. Not because those three matter most, but because everything else is scored on top of them. A vendor register attached to no inventory is a register of the vendors somebody remembered. A policy with nobody accountable for it is a document.

The threshold is zero and not weak, deliberately. A mark of one is a real if thin answer, and a thing that exists badly can be improved. A zero is the absence, and nothing downstream of an absence was earned.

The dark surface. A top verdict cannot sit on an in-scope control that reads zero. It is held at GAP instead. Without this a weighted average quietly fills the hole in: nine strong controls carry the number while the tenth, which you have not started, disappears into the mean.

Every control and its weight

22 controls across ten domains. A weight of zero means the shape has no surface for that control and it is not scored. The three marked as preconditions are the ones the foundation gate reads.

Control	Small team	Growing	Established
Is there one list of the AI tools and features in use here, and does somebody keep it current?Inventory and discovery · precondition	10	9	8
Does the list include AI switched on inside tools you already bought, and not only the AI products you chose deliberately?Inventory and discovery	6	4	3
Do you know which AI tools are being used on personal accounts instead of company ones?Inventory and discovery	3	4	3
Is there a written rule for what AI may and may not be used for that somebody joining next week could read?Policy and acceptable use · precondition	9	8	7
Could the people using AI here state the rule without going to look it up?Policy and acceptable use	5	3	3
Does the policy get revisited when a new tool is adopted, and not on a calendar?Policy and acceptable use	3	3	3
Is there a named person accountable for AI use here, meaning a person and not a team name or a job title nobody currently holds?Named accountability · precondition	9	8	7
Does each AI system on the list have a named owner of its own?Named accountability	6	5	5
When AI output drives a decision that affects a person, is it recorded who made the call?Named accountability	5	5	5
Is there a way to get a new AI tool approved that is faster than just using it without asking?Approval and offboarding	4	4	4
When a tool stops being used, is it removed from the list and its access revoked?Approval and offboarding	3	3	3
Is it written down which categories of data may not be put into an AI tool?Data boundaries	6	5	5
Do you know, per tool, what it keeps and whether your inputs are used to train it?Data boundaries	4	3	3
Is there a register of AI vendors and their sub-processors, showing what data reaches each one?Vendors and sub-processors	6	6	6
For each vendor handling personal or regulated data, is there a data-processing agreement on file?Vendors and sub-processors	5	5	5
Are the AI risks you have decided to accept written down, each with an owner and a review date?Risk and impact	4	5	6
Before an AI system affects a person's access to something, such as a job, credit, a service or a price, is its impact assessed?Risk and impact	3	4	5
If asked next quarter, could you show what an AI system produced on a given date and who reviewed it?Records and evidence	4	4	5
How long would assembling that evidence take?Records and evidence	3	3	3
Do the people using AI in their work have anything on file showing they were told how?People and training	2	3	3
Does anyone independent of the people building or using a system review the decision to deploy it?Independent challenge	n/a	4	5
Is there a standing point where AI decisions get looked at together instead of one at a time?Independent challenge	n/a	2	3

The three shapes

Small team. One person is the whole governance function, alongside their real job. There is no separate compliance, risk, legal or security function, and no standing committee that meets about anything. Pick this when the person who would write the policy is also the person who would break it. Usually under about twenty people, though the test is structural and not headcount: is there anybody here who is neither the builder nor the user?

Growing. Managers and functions exist. Somebody owns AI governance as part of a wider job, usually operations, security or legal, and reports to someone who has other priorities. Pick this when there is a person to ask but no function to escalate to, and when a new tool arrives faster than a decision about it can be made.

Established. A compliance, risk, legal or security function exists separately from the people building and using AI, and somebody outside the company looks at what you do: a customer's questionnaire, an auditor, a regulator, an insurer. Pick this when the cost of not being able to answer is somebody else's decision and not your own inconvenience.

What this assessment cannot do

It cannot verify a single answer. Nothing here touches your systems, fetches a URL you supply or reads a document, and that is a deliberate limit, not a missing feature. Every number it returns is arithmetic on what you told it.

It is not an audit. An audit is performed by somebody else, who examines evidence and can disagree with you. It is not a certification and not a conformity assessment, because nothing here is accredited to issue either. And it is not legal advice: what applies to you depends on your role, your jurisdiction and what your systems do, none of which a questionnaire can see.

That is why the top verdict is called GOVERNED AS DESCRIBED and not governed. The numbers above are borrowed from the paid Risk Register, which grades the entries in your register rather than your account of them, and says GOVERNED flat. This one reads what you told it. The longer name is the difference between the two, and it has to survive being read on a share card by somebody who did not answer any of the questions.

It names no framework on purpose. If a customer, an auditor or a regulator has already named one, the answer you need is measured against that instrument, not against this one, and the arithmetic here will not tell you whether you meet it.

Where the assessment names a gap that nothing on the RedHub shelf addresses, it says so and recommends nothing. What we sell against these gaps is listed in full, and the independent-challenge domain has nothing behind it because no download supplies a person with the standing to refuse.

Common questions

Why does the number of questions change when I change shape?: Because two of them ask about independent review, and an organization where nobody is independent of the people building and using AI has no honest answer to give. Those controls carry a weight of zero for that shape and drop out entirely, taking their weight out of the divisor as well, so a dropped control is not a silent failure. Every other control is asked at every shape.
Why are the basics weighted heavier for a small team?: Because there is nothing else there to carry them. In a larger organization an incomplete inventory is partly compensated for by procurement, a security function and a joiners process that all happen to know what is in use. In a company of six the list is the only thing that knows, so the same control is doing more of the work and the weight says so.
Where do the bands come from?: From a product RedHub already ships in this lane. The AI Risk Register & Treatment-Tracking System puts its top band at 75 and its middle band at 45, so this assessment uses the same two numbers. Choosing different thresholds by feel would have made a free assessment disagree with the paid tool it recommends. The names are not identical: that product says GOVERNED flat, because it grades the entries in your register. This one reads what you told it, so its top verdict is GOVERNED AS DESCRIBED.
Why can a gate only ever lower a verdict?: Because a weighted average is the wrong instrument for a structural failure. Nine strong controls can carry a score while the one thing everything else depends on is missing, and averaging is exactly what hides that. A gate is the place where one specific fact overrides the summary, and letting it work in the other direction would make it a bonus instead of a limit.
Why does one missing precondition fail the whole thing, when the security assessment needs a whole chain to be weak?: Because the two describe different structures. A chain of defenses only fails when the last one does, so three strong links and one weak one is a system with three things still standing. Governance is a sequence of preconditions, and a precondition is not compensated for by what got built on top of it. You cannot write a policy for AI you have not inventoried. One absence is enough, where in a chain it is not.
Can I check the score by hand?: Yes, and that is the point of publishing every weight on this page. Multiply each mark by its weight, divide by three, sum them, divide by the weight in scope and multiply by a hundred. The result is deterministic: the same answers always produce the same score, the same band and the same first thing to fix.

Run the assessment What we sell against these gaps