// Kedion · Sean McClure, PhD · Independent scientist and technical advisor

Scientific diagnosis for complex technical systems.

I investigate AI systems, software architectures, and technical claims to determine what is actually happening, and give leadership an evidence-based account they can act on.

Technical due diligenceSystems diagnosisIndependent AI advisory

I go deep into the system and come back with an explanation everyone can act on.

The questions that lead here.

Each one is a scientific investigation applied to business-critical technology. If one of these is sitting on your desk this quarter, we should talk.

01
What does this technology actually do?

An AI initiative where no two people give the same answer about what it does, and budget is about to be committed on the strength of vendor decks.

02
Why is this system behaving this way?

A production system that passes every test and is still producing something wrong.

03
Are the vendor's claims true?

A major purchase resting on performance claims that nobody internal is positioned to evaluate.

04
Is the architecture based on sound assumptions?

A roadmap or system design that assumes capabilities the models may not reliably have.

05
What class of problem are we actually dealing with?

A system everyone depends on and nobody wants to touch, because the people who built it are gone and nobody can say what kind of thing it is.

06
What are we actually acquiring?

An acquisition, and two stacks that have to be understood before they can be merged.

Down into the system, then back up into something you can act on.

Three moves, in sequence: descent, causal reconstruction, classification. The third is what makes the first two worth paying for: anyone can produce a 60-page findings document nobody reads.

01

Descend.

Into the code, the documents, the decision history, the interviews, as deep as the material goes.

02

Reconstruct the causal chain.

What actually happens, in what order, driven by what. Not the architecture diagram, the real one.

03

Classify.

Identify what kind of system this actually is and place it in a class whose behavior is already understood, so its known properties become available to your team.

The method as a diagram A causal chain runs left to right. The extracted chain is highlighted. A dashed line lifts it up to a box labelled a known class of system. a known class of system properties your team can use input step ◆ the real chain extracted output
The method, drawn once: descend into the system, reconstruct the real causal chain, then classify it, placing it in a class of system whose behavior is already known, so its properties are available to your team.

What the work looks like.

A few representative engagements, each an investigation into a technical reality: a production AI system, a vendor's claims, an architecture's assumptions, a leadership team's picture of the technology. The range runs from a boardroom to a production codebase.

Executive technical briefing · the state of the art

What the technology actually does

A leadership team was set to commit budget to an AI plan assembled from vendor decks. I gave them an independent read of what the technology can and can't actually do, in terms they could act on, before the spend.

Systems diagnosis · an AI tool in production

An accurate model of their own tools

A team had wired an AI tool into a production workflow and couldn't say where to trust it. I mapped where it was reliable and where it wasn't, so they leaned on it where it earned trust and stopped where it didn't.

Independent advisory · a roadmap's assumptions

A roadmap that assumed too much

A proposed roadmap assumed capabilities the models don't reliably have. Tracing the assumptions back to what is actually known reclassified the hard part and reordered the plan around what would work.

Technical due diligence · a vendor's claims

The product, not the pitch

Ahead of a major purchase, a team wanted a vendor's performance claims checked against evidence. I verified and reproduced what could be, so the decision rested on what the product does, not what the deck says.

Sean McClure

I'm an independent scientist and technical advisor. I come in from the outside with no stake in the existing explanation, architecture, or proposed solution. My job is to determine what is actually happening.

I diagnose complex technical systems by reconstructing how they actually work, identifying what kind of system they are, and translating the result into decisions executives and engineers can act on together. The critical step is classification: identifying what class of system or problem you're actually dealing with, which properties matter, and what established body of knowledge applies to it. A problem that is classified correctly becomes much easier to diagnose.

The second advantage is independence. I have nothing to sell downstream and nothing to gain from the answer being reassuring. I'm there to establish the technical reality and give you an account of it you can act on. Finding what's broken along the way is often just a byproduct.

Sean McClure, PhD, computational chemistry.
Independent scientist and technical advisor.
Enterprise delivery for Fortune 500s over many years.
Author of Discovered, Not Designed (2024).
Research published on Zenodo.

Notes2Tree makes structure visible.

Notes2Tree takes your material: notes, meeting transcripts, backlogs, code…and turns it into representations you can reason with. See hierarchy, sequence, process, causality, argument, and relationships that are difficult to see in the raw material.

Different views expose different properties of the same system, helping you understand how it works, diagnose where problems lie, and decide what to investigate next.

Have a system, an architecture, or a claim no one can fully explain?

That's the job. Book a call and describe it: an AI system, a codebase, a data pipeline, a vendor's claims, a proposed architecture, an acquired stack. I'll tell you what a proper investigation would look like.

Work is done under NDA by default, and anything published is de-identified.