Research

Preprint · 2026

The Average-Fit Loss

What does it cost to serve a whole population with one design tuned to its centre?

A single population-level number for the welfare given up when one design serves everyone. It brings four results from separate literatures under one metric, and measures all four on the ANSUR II anthropometric survey: 6,068 people, 93 body measurements, the modern descendant of the 1952 cockpit study. One self-contained script reproduces every figure in the paper.

What it found

  1. 01

    Under a smooth fit model the loss equals the population's dispersion, which is the mismatch cost that address models of product differentiation already price. In the quadratic case it is simply the total variance.

  2. 02

    A system measuring itself understates its own loss, because the people it excluded are absent from its data, and the understatement grows as it excludes more. At 30 percent excluded a system sees 63 percent of the misfit it imposes.

  3. 03

    In more than a few dimensions almost nobody is near-average on all of them. Of 6,068 people, 16 are average on ten dimensions and none on all 93. The cost is borne broadly, not by a nameable minority.

The other half

Single Human-AI Fit takes this metric to a system where fit is conversational rather than physical, and where the deviation is a single fact a person states about themselves.

Single Human-AI Fit

Read it

Cite as
Podgortsev, M. (2026). The Average-Fit Loss. Zenodo. 10.5281/zenodo.22364499
Licence
CC BY 4.0
ORCID
0009-0007-5561-1916

Get in touch

If you want to write something together, have a speaking invitation, or just want to talk about the ideas, leave a note.