Telling apart the sources of variability in a system
An exploratory look at variance itself: separating variation that comes from the world from variation introduced by the instrument and by the model.
Research question
When a measurement scatters, how much comes from the system, how much from the instrument, and how much from the model — and can those be attributed rather than lumped together?
Overview
A noisy result has at least three possible authors: a genuinely variable world, an imperfect instrument, and a model that does not fit. Confusing them is one of the most common ways to be confidently wrong.
The Variance Project asks whether these can be attributed rather than pooled. It is at the earliest stage: a stated question and a small set of test cases where the answer is known by construction.
Approach
- Construct cases where the true breakdown of variance is known by design.
- Test whether standard methods recover that breakdown, and characterise where they fail.
- Report the failure modes as plainly as the successes.
Log
July 8, 2026 · Concept logged
Opened as an exploratory concept. Assembling synthetic cases with known variance structure.
Findings to date
Exploratory. No results — the question is still being framed.