invariantfinder.com
The 2026 problem
AI systems will pursue whatever you ask them to pursue. They're exceptionally good at it.
Which means the bottleneck is no longer goal-achievement. It's whether the way you operate is principled enough to be worth amplifying.
Prescribe a goal: AI optimizes toward it, including in ways you didn't anticipate and wouldn't endorse.
Prescribe a process — a set of principles that are genuinely invariant, that hold under pressure and in novel conditions — and your AI systems reason from the same structure that makes your organization coherent.
The difference is the constitutional layer. Most organizations don't have one. They have a mission statement, a set of values, a strategic plan. These describe what the organization aspires to. They are not a model of how it actually operates.
The constitutional profile finds the actual model.
What we do
We build constitutional profiles of organizations: the minimum description of how you actually operate, consistent with all your observed decisions, predictive of novel situations.
Revealed decision logic
Not your stated mission, but the operating principle visible in what you've actually done — including decisions made under pressure when the stated principles were costly to hold.
Invisible assumptions
What your organization treats as given without examining. The things you'd never think to question because questioning them feels like questioning the ground.
Untested constraints
Limits you believe exist but haven't tested. The things you don't do not because they've failed, but because you've never tried.
Actual versus formal authority
Who actually makes decisions, versus the org chart. Where the information bottlenecks are. What decisions require whose approval that the structure doesn't reflect.
Update history
What has and hasn't changed when evidence challenged it. This reveals whether your principles are genuinely invariant or merely stable because they haven't been tested.
The result is three sentences that are the true causal model of your organization. The gap between those sentences and what you currently say about yourself is where you'll find the decision inefficiencies, the AI deployment mismatches, and the scale risks you haven't named yet.
Before we begin
One question precedes everything else.
Can the people who will maintain your constitutional document update their model — not just their conclusions — when evidence challenges the framework they've been using to reason?
Not: can they accept being wrong about a fact? Most people can.
But: can they accept being wrong about the way they were reasoning about the problem?
A constitutional profile maintained by people who can't do this will produce a document that justifies the existing model. More formal documentation of the same entropy.
We run a probe before beginning. It takes two hours. The result determines what kind of engagement is possible.
The diagnostic
Five questions. Take ten minutes.
01.
When your AI system encounters a situation your operating instructions don't cover, what does it do?
02.
When was your AI system's operating document last updated to reflect what actually changed in your organization?
03.
After each significant AI-assisted session, what's the compressed decision record — what was decided, why, and what evidence would change it?
04.
What's the gap between how your organization describes itself and what it actually does under pressure?
05.
Can your leadership team produce genuine challenges to their own strategic positions — arguments that, if correct, would require changing direction, not just updating a belief within the current framework?
If question five produced discomfort: that's the signal.
The engagement has already started.
Entry
We take a small number of engagements at any time. Each produces a corpus entry that compounds the instrument's accuracy. If the diagnostic resonated:
g@computerfuture.meFind what actually holds.