Survival of Fitness

MRI

Magnetic Regression Index

A category, ranking, model or forecast begins as a way of describing the world.

Magnetic Regression is the possibility that, once it becomes a center, it starts helping shape the world it was built to describe.

MRI is where I put that possibility against the evidence: locating a consequential decision, restoring what the abstraction compressed, and watching whether the decision changes.

Start with you — or an idea, a capability, a discovery, something that does not yet have a settled place.

Move the slider.

  1. 1. YOU ARRIVE

  2. 2. A CENTER IS ALREADY THERE

  3. 3. WHAT FITS BECOMES EASIER TO SEE

  4. 4. SELECTION ACCUMULATES

  5. 5. THE RESULT BECOMES THE RECORD

You. An idea. A capability. A discovery. Something that does not yet have a settled place.

FIRST TEST / MRI-001 47 → 1

AI is becoming a power problem before we've fully separated the kinds of work creating the demand. Models become tokens. Tokens become compute. Compute becomes chips, data centers, electricity and capital.

But AI work is not one thing. Frontier reasoning, coding agents, medical systems, private inference and millions of short, repetitive actions may not require the same kind of intelligence. MRI-001 asked what happens if you start there.

A public benchmark ranked Claude 1 and Qwen 47. For one defined, repeatable kind of work, using the same underlying public evidence:

Qwen ranked 1. Claude ranked 28.

This does not mean Qwen is the better model. It means the model you select can depend on the work you are actually asking it to do.

If the work varies, how much of what we call “AI demand” should vary with it before it becomes compute, infrastructure and power?

A stronger follow-up test for magnetic pull failed. That result is included too.