GEODESIC

Is this just another AI tool?

No. AI gathers the evidence. Galileo does the math.

Here is where that math came from, why you have not heard of it, and what changed.

Every AI vendor looks and sounds the same.

Most people cannot tell them apart. Trust one blindly and you can pay for it in public.

Google · 2023

$100 billion

Google’s chatbot got one fact wrong in its own launch ad. That is what Alphabet lost in market value that day.

Reuters, February 2023

Deloitte · 2025

A$440,000

The price of a government report that carried a fabricated quote from a court judgment and cited academic papers that do not exist. Deloitte refunded part of it.

Associated Press, October 2025

A New York law firm · 2023

6 fake cases

Lawyers let ChatGPT do their research. None of the six court decisions it cited existed. A federal judge fined them $5,000.

Mata v. Avianca, S.D.N.Y., June 2023

The difference matters most on your biggest decisions.

AI is good at plenty. Strategy is not on the list.

Good at

Writing a first draft.
Summing up a long report.
Answering routine emails.
Turning notes into slides.

Not good at

Who to target.
Who to stop chasing.
How to stand apart.
Which message to fund.

AI vendors promise strategy. They do not deliver it.

What you are promised, and what you get.

What AI vendors promise

Fast answers.

What you actually get
  • Ask twice, get two different answers.
  • Always sounds sure.
  • Goes along with whatever you push.
  • You cannot see how it got there.
How Geodesic differs

Measured answers.

What you get
  • The same answer every time.
  • It says when the evidence is not enough.
  • Built on real customer evidence.
  • Every source shown.

An answer you can defend to your CFO.

What it measures

  • Which audience segments to target
  • How customers see you and your rivals
  • How far you are from what they want
  • How hard each group is to move
  • Which ad concept lands hardest

When AI gets those wrong, you pay for it. You fund the wrong ad. You chase people who will never buy. You find out after the money is spent. And you cannot explain any of it to your CFO.

It is not an LLM that was trained on one of your studies.

There are four parts to this. Three of them are machinery. The fourth is a person, and the person checks every step.

See the four parts

This is the whole engine.

Nine shelves hold the evidence. The sweep fills them under gates. The Atlas reads them as one body. The mathematics and the operating system sit over all of it and never change between clients.

The mathematics

Heavily based on tensor analysis and Riemannian geometry.

Metric multidimensional scaling on ratio-scaled data. Every measurement sits in one space, where the distances between them mean something.

The operating system

What counts as a finding, settled in advance.

The same for every client, every time. Gates that can stop a run. Built here, and the one part of this that is not for sale.

The shelves one stack, nine shelves
  • 0 Master Corpus
  • 1 Foundational Data
  • 2 Proprietary Data
  • 3 Syndicated Data
  • 4 Corporate Goals
  • 5 Project Briefing
  • 6 Internal Comms
  • 7 Geometry
  • 8 Results
The Atlas one concept space learned from your whole library

Lineage stripped before anything counts twice, so nothing is agreed with by the same document wearing two covers.

How to read it
  • A clusterSources that say the same thing. Useful, or the same document wearing four covers.
  • A thick linkTwo pieces of evidence that sit close together. Check whether one is quoting the other.
  • A big dotBacked up by more separate places. The only kind of agreement worth leaning on.
  • A lone pointEvidence connected to nothing else you own. Either a unique asset or an orphan.
  • Empty spaceA question nothing in your library answers. That is next year’s research plan.
The sweep what fills the shelves
PopulationRegisterDate

Three gates, then lineage stripped.

What it sweeps
  • Regulators and federal statistics
  • Filings and registrations
  • The company’s own pages
  • Peer-reviewed and preprint
  • Sector bodies and foundations
  • Consultancies
  • Syndicated vendors
  • Trade press
  • Customer voice
  • Search and media observatories
  • Competitors’ advertising

Refused or unavailable, and recorded as such: Reddit, Google Play, Glassdoor, anything licensed, and anything whose terms bar commercial or AI use.

Nothing enters the Atlas without a date, a source, and who paid for it.

There is no third input. That is the whole of it.

The engine does not change between clients. What changes is what you put on the shelves.

But what is Galileo?

My father built Galileo. I helped.

Dr. Joseph Woelfel created it. He built it across Wisconsin–Madison, Chicago, Illinois at Urbana–Champaign, Michigan State and Michigan. It was programmed and first commercialised at Albany. It has been in the peer-reviewed literature under its own name ever since, which means you can go and check it.

In 2009 the U.S. Army funded a RAND review of ways to measure and change attitudes at scale. Of the social-science approaches RAND looked at, this was the one it recommended be tested for operational use. The report calls it the closest any approach came to a usable framework for planning and assessing attitude change.

Two of the other approaches were judged already mature. This one was not. That difference matters and we will not blur it.

RAND, Foundations of Effective Influence Operations, MG-654-A, 2009.

One peer-reviewed paper tested its precision and stability head on, and found it agreed with traditional measures. The authors called that evidence preliminary. We use their word.

Gillham and Woelfel, Human Communication Research 3, no. 3 (1977): 222–234.

The published record, and what is still open

“If it is so great, why have I not heard of it?”

Fair question. Here is the honest answer.

  • Too complex
  • Too expensive
  • Too long to complete
  • Too hard to understand
  • Too hard to run

It was the perfect academic tool. And impossible to sell.

What made it impossible to sell is gone.

Too complex→Simple
Too expensive→Affordable
Too long to complete→Days or hours
Too hard to understand→Plain English
Too hard to run→I run it for you

It was never a mathematics problem. It needed more computing than anyone would spend on a marketing question. That changed. Nothing else about it did.

AI is not the answer. It is the engine that makes the answer possible.

What we have not established is on the evidence page.

No published hit rate against real commercial outcomes. Two-analyst reliability has not been run. We put that on the site ourselves rather than leave you to find it.